Créez un pipeline TFX pour vos données avec le modèle Penguin

Voir sur TensorFlow.org

Exécuter dans Google Colab

Voir la source sur GitHub

Télécharger le cahier

introduction

Ce document fournit des instructions pour créer un pipeline TensorFlow Extended (TFX) pour votre propre ensemble de données à l'aide du modèle de pingouin fourni avec le package Python TFX. Le pipeline créé utilisera initialement le jeu de données Palmer Penguins , mais nous transformerons le pipeline pour votre jeu de données.

Conditions préalables

• Linux / Mac OS
• Python 3.6-3.8
• Cahier Jupyter

Étape 1. Copiez le modèle prédéfini dans votre répertoire de projet.

Dans cette étape, nous allons créer un répertoire de projet de pipeline de travail et des fichiers en copiant les fichiers du modèle de pingouin dans TFX. Vous pouvez considérer cela comme un échafaudage pour votre projet de pipeline TFX.

Mettre à jour le pip

Si nous utilisons Colab, nous devons nous assurer que nous disposons de la dernière version de Pip. Les systèmes locaux peuvent bien sûr être mis à jour séparément.

import sys
if 'google.colab' in sys.modules:
  !pip install --upgrade pip

Installer le package requis

Tout d'abord, installez TFX et TensorFlow Model Analysis (TFMA).

pip install -U tfx tensorflow-model-analysis

Vérifions les versions de TFX.

import tensorflow as tf
import tensorflow_model_analysis as tfma
import tfx

print('TF version: {}'.format(tf.__version__))
print('TFMA version: {}'.format(tfma.__version__))
print('TFX version: {}'.format(tfx.__version__))

TF version: 2.7.1
TFMA version: 0.37.0
TFX version: 1.6.0

Nous sommes prêts à créer un pipeline.

Définissez PROJECT_DIR sur la destination appropriée pour votre environnement. La valeur par défaut est ~/imported/${PIPELINE_NAME} , ce qui est approprié pour l'environnement Google Cloud AI Platform Notebook .

Vous pouvez donner à votre pipeline un nom différent en modifiant le PIPELINE_NAME ci-dessous. Cela deviendra également le nom du répertoire du projet où vos fichiers seront placés.

PIPELINE_NAME="my_pipeline"
import os
# Set this project directory to your new tfx pipeline project.
PROJECT_DIR=os.path.join(os.path.expanduser("~"), "imported", PIPELINE_NAME)

Copiez les fichiers modèles.

TFX inclut le modèle de penguin avec le package python TFX. Le modèle de penguin contient de nombreuses instructions pour intégrer votre ensemble de données dans le pipeline, ce qui est l'objet de ce didacticiel.

La commande CLI tfx template copy copie les fichiers de modèles prédéfinis dans votre répertoire de projet.

# Set `PATH` to include user python binary directory and a directory containing `skaffold`.
PATH=%env PATH
%env PATH={PATH}:/home/jupyter/.local/bin

!tfx template copy \
  --pipeline-name={PIPELINE_NAME} \
  --destination-path={PROJECT_DIR} \
  --model=penguin

env: PATH=/tmpfs/src/tf_docs_env/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/opt/puppetlabs/bin:/opt/android-studio/current/bin:/usr/local/go/bin:/usr/local/go/packages/bin:/opt/kubernetes/client/bin/:/home/kbuilder/.local/bin:/home/jupyter/.local/bin
CLI
Copying penguin pipeline template
kubeflow_runner.py -> /home/kbuilder/imported/my_pipeline/kubeflow_runner.py
configs.py -> /home/kbuilder/imported/my_pipeline/pipeline/configs.py
pipeline.py -> /home/kbuilder/imported/my_pipeline/pipeline/pipeline.py
__init__.py -> /home/kbuilder/imported/my_pipeline/pipeline/__init__.py
model.py -> /home/kbuilder/imported/my_pipeline/models/model.py
features.py -> /home/kbuilder/imported/my_pipeline/models/features.py
features_test.py -> /home/kbuilder/imported/my_pipeline/models/features_test.py
preprocessing_test.py -> /home/kbuilder/imported/my_pipeline/models/preprocessing_test.py
preprocessing.py -> /home/kbuilder/imported/my_pipeline/models/preprocessing.py
model_test.py -> /home/kbuilder/imported/my_pipeline/models/model_test.py
__init__.py -> /home/kbuilder/imported/my_pipeline/models/__init__.py
constants.py -> /home/kbuilder/imported/my_pipeline/models/constants.py
local_runner.py -> /home/kbuilder/imported/my_pipeline/local_runner.py
__init__.py -> /home/kbuilder/imported/my_pipeline/__init__.py

Remplacez le contexte du répertoire de travail dans ce bloc-notes par le répertoire du projet.

%cd {PROJECT_DIR}

/home/kbuilder/imported/my_pipeline

Parcourez vos fichiers source copiés

Le modèle TFX fournit des fichiers d'échafaudage de base pour créer un pipeline, y compris le code source Python et des exemples de données. Le modèle penguin utilise le même jeu de données Palmer Penguins et le même modèle ML que l' exemple Penguin .

Voici une brève introduction à chacun des fichiers Python.

• pipeline - Ce répertoire contient la définition du pipeline
  • configs.py - définit les constantes communes pour les exécuteurs de pipeline
  • pipeline.py — définit les composants TFX et un pipeline
• models : ce répertoire contient des définitions de modèles de ML
  • features.py , features_test.py — définit les fonctionnalités du modèle
  • preprocessing.py , preprocessing_test.py — définit les routines de prétraitement pour les données
  • constants.py — définit les constantes du modèle
  • model.py , model_test.py - définit le modèle ML à l'aide de frameworks ML tels que TensorFlow
• local_runner.py — définit un runner pour l'environnement local qui utilise le moteur d'orchestration local
• kubeflow_runner.py — définit un runner pour le moteur d'orchestration Kubeflow Pipelines

Par défaut, le modèle ne comprend que des composants TFX standard. Si vous avez besoin d'actions personnalisées, vous pouvez créer des composants personnalisés pour votre pipeline. Veuillez consulter le guide des composants personnalisés TFX pour plus de détails.

Fichiers de test unitaire.

Vous remarquerez peut-être qu'il existe des fichiers avec _test.py dans leur nom. Ce sont des tests unitaires du pipeline et il est recommandé d'ajouter d'autres tests unitaires à mesure que vous implémentez vos propres pipelines. Vous pouvez exécuter des tests unitaires en fournissant le nom de module des fichiers de test avec l'indicateur -m . Vous pouvez généralement obtenir un nom de module en supprimant l'extension .py et en remplaçant / par . . Par example:

import sys
!{sys.executable} -m models.features_test

Running tests under Python 3.7.5: /tmpfs/src/tf_docs_env/bin/python
[ RUN      ] FeaturesTest.testLabelKey
INFO:tensorflow:time(__main__.FeaturesTest.testLabelKey): 0.0s
I0203 11:08:46.306882 140258321348416 test_util.py:2309] time(__main__.FeaturesTest.testLabelKey): 0.0s
[       OK ] FeaturesTest.testLabelKey
[ RUN      ] FeaturesTest.test_session
[  SKIPPED ] FeaturesTest.test_session
----------------------------------------------------------------------
Ran 2 tests in 0.001s

OK (skipped=1)

Créez un pipeline TFX dans un environnement local.

TFX prend en charge plusieurs moteurs d'orchestration pour exécuter des pipelines. Nous utiliserons le moteur d'orchestration local. Le moteur d'orchestration local s'exécute sans aucune autre dépendance et convient au développement et au débogage car il s'exécute sur un environnement local plutôt que de dépendre de clusters informatiques distants.

Nous utiliserons local_runner.py pour exécuter votre pipeline à l'aide de l'orchestrateur local. Vous devez créer un pipeline avant de l'exécuter. Vous pouvez créer un pipeline avec la commande de pipeline create de pipeline.

tfx pipeline create --engine=local --pipeline_path=local_runner.py

CLI
Creating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" created successfully.

La commande pipeline create enregistre votre pipeline défini dans local_runner.py sans l'exécuter réellement.

Vous exécuterez le pipeline créé avec la commande run create dans les étapes suivantes.

Étape 2. Ingérez VOS données dans le pipeline.

Le pipeline initial ingère l'ensemble de données penguin qui est inclus dans le modèle. Vous devez mettre vos données dans le pipeline, et la plupart des pipelines TFX commencent par le composant ExampleGen.

Choisissez un ExampleGen

Vos données peuvent être stockées partout où votre pipeline peut accéder, soit sur un système de fichiers local ou distribué, soit sur un système interrogeable. TFX fournit divers composants ExampleGen pour intégrer vos données dans un pipeline TFX. Vous pouvez en choisir un parmi les exemples de génération de composants suivants.

• CsvExampleGen : Lit les fichiers CSV dans un répertoire. Utilisé dans l'exemple du pingouin et l'exemple du taxi de Chicago .
• ImportExampleGen : Prend les fichiers TFRecord avec le format de données TF Example. Utilisé dans les exemples MNIST .
• FileBasedExampleGen pour le format Avro ou Parquet .
• BigQueryExampleGen : lit directement les données dans Google Cloud BigQuery. Utilisé dans les exemples de taxis de Chicago .

Vous pouvez également créer votre propre ExampleGen, par exemple, tfx inclut un ExecampleGen personnalisé qui utilise Presto comme source de données. Consultez le guide pour plus d'informations sur l'utilisation et le développement d'exécuteurs personnalisés.

Une fois que vous avez décidé quel ExampleGen utiliser, vous devrez modifier la définition du pipeline pour utiliser vos données.

1. Modifiez le DATA_PATH dans local_runner.py et définissez-le sur l'emplacement de vos fichiers.

   • Si vous avez des fichiers dans un environnement local, spécifiez le chemin. C'est la meilleure option pour développer ou déboguer un pipeline.
   • Si les fichiers sont stockés dans GCS, vous pouvez utiliser un chemin commençant par gs://{bucket_name}/... . Veuillez vous assurer que vous pouvez accéder à GCS depuis votre terminal, par exemple, en utilisant gsutil . Veuillez suivre le guide d'autorisation dans Google Cloud si nécessaire.
   • Si vous souhaitez utiliser un ExampleGen basé sur une requête comme BigQueryExampleGen, vous avez besoin d'une instruction Query pour sélectionner des données à partir de la source de données. Vous devez définir quelques éléments supplémentaires pour utiliser Google Cloud BigQuery en tant que source de données.
   • Dans pipeline/configs.py :
     • Remplacez GOOGLE_CLOUD_PROJECT et GCS_BUCKET_NAME par votre projet GCP et le nom de votre compartiment. Le bucket doit exister avant l'exécution du pipeline.
     • Décommentez la variable BIG_QUERY_WITH_DIRECT_RUNNER_BEAM_PIPELINE_ARGS .
     • Décommentez et définissez la variable BIG_QUERY_QUERY sur votre instruction de requête .
   • Dans local_runner.py :
     • Commentez l'argument data_path et décommentez l'argument de query à la place dans pipeline.create_pipeline() .
   • Dans pipeline/pipeline.py :
     • Commentez l'argument data_path et décommentez l'argument de query dans create_pipeline() .
     • Utilisez BigQueryExampleGen au lieu de CsvExampleGen.

2. Remplacez CsvExampleGen existant par votre classe ExampleGen dans pipeline/pipeline.py . Chaque classe ExampleGen a une signature différente. Veuillez consulter le guide des composants ExampleGen pour plus de détails. N'oubliez pas d'importer les modules requis avec les instructions d' import dans pipeline/pipeline.py .

Le pipeline initial est composé de quatre composants, ExampleGen , StatisticsGen , SchemaGen et ExampleValidator . Nous n'avons rien à changer pour StatisticsGen , SchemaGen et ExampleValidator . Exécutons le pipeline pour la première fois.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}

CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 1
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=1, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'output_data_format': 6, 'output_file_format': 5, 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/1/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/1/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:09:12.848598153    5127 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 1 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 1
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 2
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=2, input_dict={'examples': [Artifact(artifact: id: 1
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886553302
last_update_time_since_epoch: 1643886553302
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/2/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/2/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 2 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 2
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 3
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=3, input_dict={'statistics': [Artifact(artifact: id: 2
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886556588
last_update_time_since_epoch: 1643886556588
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/3/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/3/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 3 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 3
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Vous devriez voir "Le composant ExampleValidator est terminé." si le pipeline s'est exécuté avec succès.

Examinez la sortie du pipeline.

Le pipeline TFX produit deux types de sortie, des artefacts et une base de données de métadonnées (MLMD) qui contient des métadonnées d'artefacts et d'exécutions de pipeline. L'emplacement de la sortie est défini dans local_runner.py . Par défaut, les artefacts sont stockés dans le répertoire tfx_pipeline_output et les métadonnées sont stockées dans une base de données sqlite sous le répertoire tfx_metadata .

Vous pouvez utiliser les API MLMD pour examiner ces sorties. Tout d'abord, nous allons définir quelques fonctions utilitaires pour rechercher des artefacts de sortie qui viennent d'être produits.

import tensorflow as tf
import tfx
from ml_metadata import errors
from ml_metadata.proto import metadata_store_pb2
from tfx.types import artifact_utils

# TODO(b/171447278): Move these functions into TFX library.

def get_latest_executions(store, pipeline_name, component_id = None):
  """Fetch all pipeline runs."""
  if component_id is None:  # Find entire pipeline runs.
    run_contexts = [
        c for c in store.get_contexts_by_type('run')
        if c.properties['pipeline_name'].string_value == pipeline_name
    ]
  else:  # Find specific component runs.
    run_contexts = [
        c for c in store.get_contexts_by_type('component_run')
        if c.properties['pipeline_name'].string_value == pipeline_name and
           c.properties['component_id'].string_value == component_id
    ]
  if not run_contexts:
    return []
  # Pick the latest run context.
  latest_context = max(run_contexts,
                       key=lambda c: c.last_update_time_since_epoch)
  return store.get_executions_by_context(latest_context.id)

def get_latest_artifacts(store, pipeline_name, component_id = None):
  """Fetch all artifacts from latest pipeline execution."""
  executions = get_latest_executions(store, pipeline_name, component_id)

  # Fetch all artifacts produced from the given executions.
  execution_ids = [e.id for e in executions]
  events = store.get_events_by_execution_ids(execution_ids)
  artifact_ids = [
      event.artifact_id for event in events
      if event.type == metadata_store_pb2.Event.OUTPUT
  ]
  return store.get_artifacts_by_id(artifact_ids)

def find_latest_artifacts_by_type(store, artifacts, artifact_type):
  """Get the latest artifacts of a specified type."""
  # Get type information from MLMD
  try:
    artifact_type = store.get_artifact_type(artifact_type)
  except errors.NotFoundError:
    return []
  # Filter artifacts with type.
  filtered_artifacts = [aritfact for aritfact in artifacts
                        if aritfact.type_id == artifact_type.id]
  # Convert MLMD artifact data into TFX Artifact instances.
  return [artifact_utils.deserialize_artifact(artifact_type, artifact)
      for artifact in filtered_artifacts]


from tfx.orchestration.experimental.interactive import visualizations

def visualize_artifacts(artifacts):
  """Visualizes artifacts using standard visualization modules."""
  for artifact in artifacts:
    visualization = visualizations.get_registry().get_visualization(
        artifact.type_name)
    if visualization:
      visualization.display(artifact)

from tfx.orchestration.experimental.interactive import standard_visualizations
standard_visualizations.register_standard_visualizations()

import pprint

from tfx.orchestration import metadata
from tfx.types import artifact_utils
from tfx.types import standard_artifacts

def preview_examples(artifacts):
  """Preview a few records from Examples artifacts."""
  pp = pprint.PrettyPrinter()
  for artifact in artifacts:
    print("==== Examples artifact:{}({})".format(artifact.name, artifact.uri))
    for split in artifact_utils.decode_split_names(artifact.split_names):
      print("==== Reading from split:{}".format(split))
      split_uri = artifact_utils.get_split_uri([artifact], split)

      # Get the list of files in this directory (all compressed TFRecord files)
      tfrecord_filenames = [os.path.join(split_uri, name)
                            for name in os.listdir(split_uri)]
      # Create a `TFRecordDataset` to read these files
      dataset = tf.data.TFRecordDataset(tfrecord_filenames,
                                        compression_type="GZIP")
      # Iterate over the first 2 records and decode them.
      for tfrecord in dataset.take(2):
        serialized_example = tfrecord.numpy()
        example = tf.train.Example()
        example.ParseFromString(serialized_example)
        pp.pprint(example)

import local_runner

metadata_connection_config = metadata.sqlite_metadata_connection_config(
              local_runner.METADATA_PATH)

Nous pouvons maintenant lire les métadonnées des artefacts de sortie de MLMD.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
    # Search all aritfacts from the previous pipeline run.
    artifacts = get_latest_artifacts(metadata_handler.store, PIPELINE_NAME)
    # Find artifacts of Examples type.
    examples_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Examples.TYPE_NAME)
    # Find artifacts generated from StatisticsGen.
    stats_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.ExampleStatistics.TYPE_NAME)
    # Find artifacts generated from SchemaGen.
    schema_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Schema.TYPE_NAME)
    # Find artifacts generated from ExampleValidator.
    anomalies_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.ExampleAnomalies.TYPE_NAME)

Nous pouvons maintenant examiner les sorties de chaque composant. Tensorflow Data Validation (TFDV) est utilisé dans StatisticsGen , SchemaGen et ExampleValidator , et TFDV peut être utilisé pour visualiser les sorties de ces composants.

Dans ce didacticiel, nous utiliserons des méthodes d'assistance à la visualisation dans TFX qui utilisent TFDV en interne pour afficher la visualisation. Veuillez consulter le didacticiel sur les composants TFX pour en savoir plus sur chaque composant.

Examiner le formulaire de sortie ExampleGen

Examinons la sortie de ExampleGen. Jetez un œil aux deux premiers exemples pour chaque division :

preview_examples(examples_artifacts)

Par défaut, TFX ExampleGen divise les exemples en deux divisions, train et eval , mais vous pouvez ajuster votre configuration de division .

Examiner la sortie de StatisticsGen

visualize_artifacts(stats_artifacts)

Ces statistiques sont fournies à SchemaGen pour construire automatiquement un schéma de données.

Examiner la sortie de SchemaGen

visualize_artifacts(schema_artifacts)

Ce schéma est automatiquement déduit de la sortie de StatisticsGen. Nous utiliserons ce schéma généré dans ce didacticiel, mais vous pouvez également modifier et personnaliser le schéma .

Examiner la sortie de ExampleValidator

visualize_artifacts(anomalies_artifacts)

Si des anomalies ont été trouvées, vous pouvez revoir vos données pour que tous les exemples suivent vos hypothèses. Les sorties d'autres composants comme StatistcsGen pourraient être utiles. Les anomalies trouvées ne bloquent pas l'exécution du pipeline.

Vous pouvez voir les fonctionnalités disponibles à partir des sorties de SchemaGen . Si vos fonctionnalités peuvent être utilisées pour construire directement le modèle ML dans Trainer , vous pouvez ignorer l'étape suivante et passer à l'étape 4. Sinon, vous pouvez effectuer des travaux d'ingénierie de fonctionnalités à l'étape suivante. Le composant de Transform est nécessaire lorsque des opérations complètes telles que le calcul de moyennes sont requises, en particulier lorsque vous devez effectuer une mise à l'échelle.

Étape 3. (Facultatif) Ingénierie des fonctionnalités avec le composant Transform.

Dans cette étape, vous définirez divers travaux d'ingénierie de fonctionnalités qui seront utilisés par le composant Transform dans le pipeline. Voir le guide des composants de transformation pour plus d'informations.

Ceci n'est nécessaire que si votre code de formation nécessite des fonctionnalités supplémentaires qui ne sont pas disponibles dans la sortie d'ExampleGen. Sinon, n'hésitez pas à passer rapidement à l'étape suivante de l'utilisation de Trainer.

Définir les caractéristiques du modèle

models/features.py contient des constantes pour définir les fonctionnalités du modèle, y compris les noms des fonctionnalités, la taille du vocabulaire, etc. Par défaut, le modèle de penguin a deux constantes, FEATURE_KEYS et LABEL_KEY , car notre modèle de penguin résout un problème de classification en utilisant l'apprentissage supervisé et toutes les fonctionnalités sont des fonctionnalités numériques continues. Voir les définitions de fonctionnalités de l'exemple de taxi de chicago pour un autre exemple.

Implémenter le prétraitement pour la formation/le service dans preprocessing_fn().

L'ingénierie des fonctionnalités réelles se produit dans la fonction preprocessing_fn() dans models/preprocessing.py .

Dans preprocessing_fn , vous pouvez définir une série de fonctions qui manipulent le dict d'entrée des tenseurs pour produire le dict de sortie des tenseurs. Il existe des fonctions d'assistance telles que scale_to_0_1 et compute_and_apply_vocabulary dans l'API de transformation TensorFlow ou vous pouvez simplement utiliser les fonctions TensorFlow habituelles. Par défaut, le modèle de penguin inclut des exemples d'utilisation de la fonction tft.scale_to_z_score pour normaliser les valeurs des caractéristiques.

Voir le guide Tensflow Transform pour plus d'informations sur la création de preprocessing_fn .

Ajoutez le composant Transform au pipeline.

Si votre preprocessing_fn est prêt, ajoutez le composant Transform au pipeline.

1. Dans le fichier pipeline/pipeline.py , décommentez # components.append(transform) pour ajouter le composant au pipeline.

Vous pouvez mettre à jour le pipeline et l'exécuter à nouveau.

!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}

CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 4
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=4, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'output_file_format': 5, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/4/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/4/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:09:37.596944686    5287 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 4 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 4
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 5
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=5, input_dict={'examples': [Artifact(artifact: id: 4
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886578210
last_update_time_since_epoch: 1643886578210
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/5/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/5/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 5 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 5
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 6
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=6, input_dict={'statistics': [Artifact(artifact: id: 5
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886581527
last_update_time_since_epoch: 1643886581527
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'exclude_splits': '[]', 'infer_feature_shape': 1}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/6/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/6/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 6 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 6
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Si le pipeline s'est exécuté avec succès, vous devriez voir "La transformation du composant est terminée". quelque part dans le journal. Étant donné que le composant Transform et le composant ExampleValidator ne dépendent pas l'un de l'autre, l'ordre des exécutions n'est pas fixe. Cela dit, Transform et ExampleValidator peuvent être le dernier composant de l'exécution du pipeline.

Examiner la sortie de Transform

Le composant de transformation crée deux types de sorties, un graphique Tensorflow et des exemples transformés. Les exemples transformés sont des exemples de type d'artefact qui sont également produits par ExampleGen, mais celui-ci contient à la place des valeurs de caractéristiques transformées.

Vous pouvez les examiner comme nous l'avons fait à l'étape précédente.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
    # Search all aritfacts from the previous run of Transform component.
    artifacts = get_latest_artifacts(metadata_handler.store,
                                     PIPELINE_NAME, "Transform")
    # Find artifacts of Examples type.
    transformed_examples_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Examples.TYPE_NAME)

preview_examples(transformed_examples_artifacts)

Étape 4. Entraînez votre modèle avec le composant Trainer.

Nous allons construire un modèle ML à l'aide du composant Trainer . Voir le guide des composants du formateur pour plus d'informations. Vous devez fournir votre code de modèle au composant Trainer.

Définissez votre modèle.

Dans le modèle de pingouin, models.model.run_fn est utilisé comme argument run_fn pour le composant Trainer . Cela signifie que la fonction run_fn() dans models/model.py sera appelée lors de l'exécution du composant Trainer . Vous pouvez voir le code pour construire un modèle DNN simple à l'aide de l'API keras dans le code donné. Voir TensorFlow 2.x dans le guide TFX pour plus d'informations sur l'utilisation de l'API keras dans TFX.

Dans ce run_fn , vous devez créer un modèle et l'enregistrer dans un répertoire pointé par fn_args.serving_model_dir qui est spécifié par le composant. Vous pouvez utiliser d'autres arguments dans fn_args qui est passé dans le run_fn . Voir les codes associés pour la liste complète des arguments dans fn_args .

Définissez vos fonctionnalités dans models/features.py et utilisez-les selon vos besoins. Si vous avez transformé vos entités à l'étape 3, vous devez utiliser les entités transformées comme entrées de votre modèle.

Ajoutez le composant Trainer au pipeline.

Si votre run_fn est prêt, ajoutez le composant Trainer au pipeline.

1. Dans le fichier pipeline/pipeline.py , décommentez # components.append(trainer) pour ajouter le composant au pipeline.

Les arguments pour le composant formateur peuvent dépendre de l'utilisation ou non du composant Transform.

• Si vous n'utilisez PAS le composant Transform , vous n'avez pas besoin de modifier les arguments.

• Si vous utilisez le composant Transform , vous devez modifier les arguments lors de la création d'une instance du composant Trainer .

  • Remplacez l'argument examples par examples=transform.outputs['transformed_examples'], . Nous devons utiliser des exemples transformés pour la formation.
  • Ajoutez l'argument transform_graph comme transform_graph=transform.outputs['transform_graph'], . Ce graphique contient le graphique TensorFlow pour les opérations de transformation.
  • Après les modifications ci-dessus, le code de création du composant Trainer ressemblera à ce qui suit.

  # If you use a Transform component.
  trainer = Trainer(
      run_fn=run_fn,
      examples=transform.outputs['transformed_examples'],
      transform_graph=transform.outputs['transform_graph'],
      schema=schema_gen.outputs['schema'],
      ...
  

Vous pouvez mettre à jour le pipeline et l'exécuter à nouveau.

!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}

CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 7
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=7, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_base': '/home/kbuilder/imported/my_pipeline/data', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'output_file_format': 5, 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/7/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/7/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:01.173700221    5436 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 7 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 7
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 8
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=8, input_dict={'examples': [Artifact(artifact: id: 7
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886601629
last_update_time_since_epoch: 1643886601629
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/8/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/8/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 8 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 8
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 9
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=9, input_dict={'statistics': [Artifact(artifact: id: 8
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886605023
last_update_time_since_epoch: 1643886605023
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'exclude_splits': '[]', 'infer_feature_shape': 1}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/9/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/9/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 9 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 9
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Lorsque cette exécution s'exécute avec succès, vous avez maintenant créé et exécuté votre premier pipeline TFX pour votre modèle. Toutes nos félicitations!

Votre nouveau modèle sera situé à un endroit sous le répertoire de sortie, mais il serait préférable d'avoir un modèle dans un emplacement fixe ou un service en dehors du pipeline TFX qui contient de nombreux résultats intermédiaires. Encore mieux avec une évaluation continue du modèle construit qui est essentiel dans les systèmes de production ML. Nous verrons comment l'évaluation continue et les déploiements fonctionnent dans TFX à l'étape suivante.

Étape 5. (Facultatif) Évaluez le modèle avec Evaluator et publiez avec pusher.

Le composant Evaluator évalue en continu chaque modèle créé à partir de Trainer , et Pusher copie le modèle dans un emplacement prédéfini du système de fichiers ou même dans Google Cloud AI Platform Models .

Ajoute le composant Evaluator au pipeline.

Dans le fichier pipeline/pipeline.py :

1. Décommentez # components.append(model_resolver) pour ajouter le dernier résolveur de modèle au pipeline. L'évaluateur peut être utilisé pour comparer un modèle avec l'ancien modèle de référence qui a réussi l'évaluateur lors de la dernière exécution du pipeline. LatestBlessedModelResolver trouve le dernier modèle qui a passé Evaluator.
2. Définissez le tfma.MetricsSpec approprié pour votre modèle. L'évaluation peut être différente pour chaque modèle de ML. Dans le modèle de pingouin, SparseCategoricalAccuracy a été utilisé car nous résolvons un problème de classification multi-catégories. Vous devez également spécifier tfma.SliceSpec pour analyser votre modèle pour des tranches spécifiques. Pour plus de détails, consultez le guide du composant Evaluator .
3. Décommentez # components.append(evaluator) pour ajouter le composant au pipeline.

Vous pouvez mettre à jour le pipeline et l'exécuter à nouveau.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}

CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 10
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=10, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'output_file_format': 5, 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/10/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/10/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:24.894390124    5584 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 10 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 10
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 11
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=11, input_dict={'examples': [Artifact(artifact: id: 10
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886625515
last_update_time_since_epoch: 1643886625515
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/11/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/11/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 11 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 11
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 12
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=12, input_dict={'statistics': [Artifact(artifact: id: 11
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886628941
last_update_time_since_epoch: 1643886628941
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/12/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/12/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 12 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 12
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Examiner les résultats de l'évaluateur

Cette étape nécessite l'extension de bloc-notes Jupyter TensorFlow Model Analysis (TFMA). Notez que la version de l'extension de bloc-notes TFMA doit être identique à la version du package TFMA python.

La commande suivante installera l'extension de bloc-notes TFMA à partir du registre NPM. Cela peut prendre plusieurs minutes.

# Install TFMA notebook extension.
jupyter labextension install tensorflow_model_analysis@{tfma.__version__}

usage: jupyter [-h] [--version] [--config-dir] [--data-dir] [--runtime-dir]
               [--paths] [--json] [--debug]
               [subcommand]

Jupyter: Interactive Computing

positional arguments:
  subcommand     the subcommand to launch

optional arguments:
  -h, --help     show this help message and exit
  --version      show the versions of core jupyter packages and exit
  --config-dir   show Jupyter config dir
  --data-dir     show Jupyter data dir
  --runtime-dir  show Jupyter runtime dir
  --paths        show all Jupyter paths. Add --json for machine-readable
                 format.
  --json         output paths as machine-readable json
  --debug        output debug information about paths

Available subcommands: bundlerextension console dejavu execute kernel
kernelspec migrate nbconvert nbextension notebook qtconsole run
serverextension troubleshoot trust

Jupyter command `jupyter-labextension` not found.

Si l'installation est terminée, veuillez recharger votre navigateur pour que l'extension prenne effet.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
  # Search all aritfacts from the previous pipeline run.
  artifacts = get_latest_artifacts(metadata_handler.store, PIPELINE_NAME)
  model_evaluation_artifacts = find_latest_artifacts_by_type(
      metadata_handler.store, artifacts,
      standard_artifacts.ModelEvaluation.TYPE_NAME)

if model_evaluation_artifacts:
  tfma_result = tfma.load_eval_result(model_evaluation_artifacts[0].uri)
  tfma.view.render_slicing_metrics(tfma_result)

Ajoute le composant Pusher au pipeline.

Si le modèle semble prometteur, nous devons publier le modèle. Le composant Pusher peut publier le modèle dans un emplacement du système de fichiers ou dans GCP AI Platform Models à l'aide d'un exécuteur personnalisé .

Le composant Evaluator évalue en continu chaque modèle créé à partir de Trainer , et Pusher copie le modèle dans un emplacement prédéfini du système de fichiers ou même dans Google Cloud AI Platform Models .

1. Dans local_runner.py , définissez SERVING_MODEL_DIR sur un répertoire à publier.
2. Dans le fichier pipeline/pipeline.py , décommentez # components.append(pusher) pour ajouter Pusher au pipeline.

Vous pouvez mettre à jour le pipeline et l'exécuter à nouveau.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}

CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 13
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=13, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_file_format': 5, 'output_data_format': 6, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/13/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/13/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:49.163841363    5734 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 13 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 13
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 14
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=14, input_dict={'examples': [Artifact(artifact: id: 13
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886649739
last_update_time_since_epoch: 1643886649739
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/14/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/14/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 14 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 14
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 15
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=15, input_dict={'statistics': [Artifact(artifact: id: 14
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886653128
last_update_time_since_epoch: 1643886653128
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/15/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/15/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 15 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 15
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Vous devriez pouvoir trouver votre nouveau modèle dans SERVING_MODEL_DIR .

Étape 6. (Facultatif) Déployez votre pipeline sur Kubeflow Pipelines sur GCP.

Comme mentionné précédemment, local_runner.py est bon pour le débogage ou le développement, mais pas la meilleure solution pour les charges de travail de production. Au cours de cette étape, nous allons déployer le pipeline sur Kubeflow Pipelines sur Google Cloud.

Préparation

Nous avons besoin du package kfp python et du programme skaffold pour déployer un pipeline sur un cluster Kubeflow Pipelines.

pip install --upgrade -q kfp

# Download skaffold and set it executable.
curl -Lo skaffold https://storage.googleapis.com/skaffold/releases/latest/skaffold-linux-amd64 && chmod +x skaffold

Vous devez déplacer le binaire skaffold à l'endroit où votre shell peut le trouver. Ou vous pouvez spécifier le chemin vers skaffold lorsque vous exécutez le binaire tfx avec --skaffold-cmd .

# Move skaffold binary into your path
mv skaffold /home/jupyter/.local/bin/

mv: cannot move 'skaffold' to '/home/jupyter/.local/bin/': No such file or directory

Vous avez également besoin d'un cluster Kubeflow Pipelines pour exécuter le pipeline. Veuillez suivre les étapes 1 et 2 du didacticiel TFX sur Cloud AI Platform Pipelines .

Lorsque votre cluster est prêt, ouvrez le tableau de bord du pipeline en cliquant sur Ouvrir le tableau de bord des pipelines sur la page Pipelines de la console Google Cloud . L'URL de cette page est ENDPOINT pour demander une exécution de pipeline. La valeur du point de terminaison correspond à tout ce qui se trouve dans l'URL après https://, jusqu'à googleusercontent.com inclus. Placez votre point de terminaison dans le bloc de code suivant.

ENDPOINT='' # Enter your ENDPOINT here.

Pour exécuter notre code dans un cluster Kubeflow Pipelines, nous devons compresser notre code dans une image de conteneur. L'image sera construite automatiquement lors du déploiement de notre pipeline, et il vous suffira de définir un nom et un registre de conteneurs pour votre image. Dans notre exemple, nous allons utiliser le registre Google Container et le nommer tfx-pipeline .

# Read GCP project id from env.
shell_output=!gcloud config list --format 'value(core.project)' 2>/dev/null
GOOGLE_CLOUD_PROJECT=shell_output[0]

# Docker image name for the pipeline image.
CUSTOM_TFX_IMAGE='gcr.io/' + GOOGLE_CLOUD_PROJECT + '/tfx-pipeline'

Définir l'emplacement des données.

Vos données doivent être accessibles depuis le cluster Kubeflow Pipelines. Si vous avez utilisé des données dans votre environnement local, vous devrez peut-être les télécharger sur un stockage distant tel que Google Cloud Storage. Par exemple, nous pouvons télécharger des données de pingouin dans un compartiment par défaut qui est créé automatiquement lorsqu'un cluster Kubeflow Pipelines est déployé comme suit.

gsutil cp data/data.csv gs://{GOOGLE_CLOUD_PROJECT}-kubeflowpipelines-default/tfx-template/data/penguin/

Copying file://data/data.csv [Content-Type=text/csv]...
NotFoundException: 404 The destination bucket gs://tf-benchmark-dashboard-kubeflowpipelines-default does not exist or the write to the destination must be restarted

Mettez à jour l'emplacement des données stockées dans DATA_PATH dans kubeflow_runner.py .

Si vous utilisez BigQueryExampleGen, il n'est pas nécessaire d'importer le fichier de données, mais assurez-vous que kubeflow_runner.py utilise la même query et le même argument beam_pipeline_args pour la fonction pipeline.create_pipeline() .

Déployez le pipeline.

Si tout est prêt, vous pouvez créer un pipeline à l'aide de la commande tfx pipeline create .

!tfx pipeline create  \
--engine=kubeflow \
--pipeline-path=kubeflow_runner.py \
--endpoint={ENDPOINT} \
--build-target-image={CUSTOM_TFX_IMAGE}

CLI
[Error] --build-target-image flag was DELETED. You should specify the build target image at the `KubeflowDagRunnerConfig` class instead, and use --build-image flag without argument to build a container image when creating or updating a pipeline.

Maintenant, démarrez une exécution avec le pipeline nouvellement créé à l'aide de la commande tfx run create .

tfx run create --engine=kubeflow --pipeline-name={PIPELINE_NAME} --endpoint={ENDPOINT}

CLI
Creating a run for pipeline: my_pipeline
Failed to load kube config.
Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 175, in _new_conn
    (self._dns_host, self.port), self.timeout, **extra_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/connection.py", line 95, in create_connection
    raise err
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/connection.py", line 85, in create_connection
    sock.connect(sa)
ConnectionRefusedError: [Errno 111] Connection refused

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 710, in urlopen
    chunked=chunked,
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 398, in _make_request
    conn.request(method, url, **httplib_request_kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 239, in request
    super(HTTPConnection, self).request(method, url, body=body, headers=headers)
  File "/usr/lib/python3.7/http/client.py", line 1256, in request
    self._send_request(method, url, body, headers, encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1302, in _send_request
    self.endheaders(body, encode_chunked=encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1251, in endheaders
    self._send_output(message_body, encode_chunked=encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1030, in _send_output
    self.send(msg)
  File "/usr/lib/python3.7/http/client.py", line 970, in send
    self.connect()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 205, in connect
    conn = self._new_conn()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 187, in _new_conn
    self, "Failed to establish a new connection: %s" % e
urllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPConnection object at 0x7ff729e34190>: Failed to establish a new connection: [Errno 111] Connection refused

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/bin/tfx", line 8, in <module>
    sys.exit(cli_group())
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 829, in __call__
    return self.main(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 782, in main
    rv = self.invoke(ctx)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1259, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1259, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1066, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 610, in invoke
    return callback(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/decorators.py", line 73, in new_func
    return ctx.invoke(f, obj, *args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 610, in invoke
    return callback(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/commands/run.py", line 94, in create_run
    handler = handler_factory.create_handler(ctx.flags_dict)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/handler/handler_factory.py", line 93, in create_handler
    return kubeflow_handler.KubeflowHandler(flags_dict)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/handler/kubeflow_handler.py", line 62, in __init__
    namespace=self.flags_dict[labels.NAMESPACE])
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp/_client.py", line 197, in __init__
    if not self._context_setting['namespace'] and self.get_kfp_healthz(
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp/_client.py", line 411, in get_kfp_healthz
    response = self._healthz_api.get_healthz()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api/healthz_service_api.py", line 63, in get_healthz
    return self.get_healthz_with_http_info(**kwargs)  # noqa: E501
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api/healthz_service_api.py", line 148, in get_healthz_with_http_info
    collection_formats=collection_formats)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 369, in call_api
    _preload_content, _request_timeout, _host)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 185, in __call_api
    _request_timeout=_request_timeout)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 393, in request
    headers=headers)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/rest.py", line 234, in GET
    query_params=query_params)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/rest.py", line 212, in request
    headers=headers)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/request.py", line 75, in request
    method, url, fields=fields, headers=headers, **urlopen_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/request.py", line 96, in request_encode_url
    return self.urlopen(method, url, **extra_kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/poolmanager.py", line 375, in urlopen
    response = conn.urlopen(method, u.request_uri, **kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 786, in urlopen
    method, url, error=e, _pool=self, _stacktrace=sys.exc_info()[2]
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/retry.py", line 592, in increment
    raise MaxRetryError(_pool, url, error or ResponseError(cause))
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=80): Max retries exceeded with url: /apis/v1beta1/healthz (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff729e34190>: Failed to establish a new connection: [Errno 111] Connection refused'))

Vous pouvez également exécuter le pipeline dans le tableau de bord Kubeflow Pipelines. La nouvelle exécution sera répertoriée sous Experiments dans le tableau de bord Kubeflow Pipelines. Cliquer sur l'expérience vous permettra de surveiller la progression et de visualiser les artefacts créés lors de l'exécution.

Si vous souhaitez exécuter votre pipeline sur Kubeflow Pipelines, consultez le didacticiel TFX on Cloud AI Platform Pipelines pour en savoir plus.

Nettoyer

Pour nettoyer toutes les ressources Google Cloud utilisées dans cette étape, vous pouvez supprimer le projet Google Cloud que vous avez utilisé pour le didacticiel.

Vous pouvez également nettoyer des ressources individuelles en visitant chaque console :

• Stockage en nuage Google
• Registre de conteneurs Google
• Moteur Google Kubernetes