tfx.components.Trainer

A TFX component to train a TensorFlow model.

Inherits From: BaseComponent

View aliases

Main aliases

tfx.components.trainer.component.Trainer

tfx.components.Trainer(
    examples=None, transformed_examples=None, transform_graph=None, schema=None,
    base_model=None, hyperparameters=None, module_file=None, run_fn=None,
    trainer_fn=None, train_args=None, eval_args=None, custom_config=None,
    custom_executor_spec=None, output=None, transform_output=None,
    instance_name=None
)

Used in the notebooks

Used in the tutorials
TFX Component Tutorial TFX Component Tutorial

The Trainer component is used to train and eval a model using given inputs and a user-supplied estimator.

Providing an estimator

The TFX executor will use the estimator provided in the module_file file to train the model. The Trainer executor will look specifically for the trainer_fn() function within that file. Before training, the executor will call that function expecting the following returned as a dictionary:

estimator: The estimator to be used by TensorFlow to train the model.
train_spec: The configuration to be used by the "train" part of the TensorFlow train_and_evaluate() call.
eval_spec: The configuration to be used by the "eval" part of the TensorFlow train_and_evaluate() call.
eval_input_receiver_fn: The configuration to be used by the ModelValidator component when validating the model.

An example of trainer_fn() can be found in the user-supplied code of the TFX Chicago Taxi pipeline example.

Note: The default executor for this component trains locally. This can be overriden to enable the model to be trained on other platforms. The Cloud AI Platform custom executor provides an example how to implement this.

Please see https://www.tensorflow.org/guide/estimators for more details.

Example 1: Training locally

# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
    module_file=module_file,
    transformed_examples=transform.outputs['transformed_examples'],
    schema=infer_schema.outputs['schema'],
    transform_graph=transform.outputs['transform_graph'],
    train_args=trainer_pb2.TrainArgs(num_steps=10000),
    eval_args=trainer_pb2.EvalArgs(num_steps=5000))

Example 2: Training through a cloud provider

# Train using Google Cloud AI Platform.
trainer = Trainer(
    executor_class=ai_platform_trainer_executor.Executor,
    module_file=module_file,
    transformed_examples=transform.outputs['transformed_examples'],
    schema=infer_schema.outputs['schema'],
    transform_graph=transform.outputs['transform_graph'],
    train_args=trainer_pb2.TrainArgs(num_steps=10000),
    eval_args=trainer_pb2.EvalArgs(num_steps=5000))

Args:

examples: A Channel of type standard_artifacts.Examples, serving as the source of examples used in training (required). May be raw or transformed.
transformed_examples: Deprecated field. Please set 'examples' instead.
transform_graph: An optional Channel of type standard_artifacts.TransformGraph, serving as the input transform graph if present.
schema: A Channel of type standard_artifacts.Schema, serving as the schema of training and eval data.
base_model: A Channel of type Model, containing model that will be used for training. This can be used for warmstart, transfer learning or model ensembling.
hyperparameters: A Channel of type standard_artifacts.HyperParameters, serving as the hyperparameters for training module. Tuner's output best hyperparameters can be feed into this.
module_file: A path to python module file containing UDF model definition.

For default executor, The module_file must implement a function named trainer_fn at its top level. The function must have the following signature.

def trainer_fn(trainer.executor.TrainerFnArgs, tensorflow_metadata.proto.v0.schema_pb2) -> Dict: ...

where the returned Dict has the following key-values. 'estimator': an instance of tf.estimator.Estimator 'train_spec': an instance of tf.estimator.TrainSpec 'eval_spec': an instance of tf.estimator.EvalSpec 'eval_input_receiver_fn': an instance of tfma.export.EvalInputReceiver. Exactly one of 'module_file' or 'trainer_fn' must be supplied.

For generic executor, The module_file must implement a function named run_fn at its top level with function signature: def run_fn(trainer.executor.TrainerFnArgs), and the trained model must be saved to TrainerFnArgs.serving_model_dir when execute this function.
run_fn: A python path to UDF model definition function for generic trainer. See 'module_file' for details. Exactly one of 'module_file' or 'run_fn' must be supplied if Trainer uses GenericExecutor.
trainer_fn: A python path to UDF model definition function for estimator based trainer. See 'module_file' for the required signature of the UDF. Exactly one of 'module_file' or 'trainer_fn' must be supplied.
train_args: A trainer_pb2.TrainArgs instance, containing args used for training. Current only num_steps is available.
eval_args: A trainer_pb2.EvalArgs instance, containing args used for eval. Current only num_steps is available.
custom_config: A dict which contains addtional training job parameters that will be passed into user module.
custom_executor_spec: Optional custom executor spec.
output: Optional Model channel for result of exported models.
transform_output: Backwards compatibility alias for the 'transform_graph' argument.
instance_name: Optional unique instance name. Necessary iff multiple Trainer components are declared in the same pipeline.

Attributes:

component_id: DEPRECATED FUNCTION

Warning: THIS FUNCTION IS DEPRECATED. It will be removed in a future version. Instructions for updating: component_id is deprecated, use id instead
component_type: DEPRECATED FUNCTION

Warning: THIS FUNCTION IS DEPRECATED. It will be removed in a future version. Instructions for updating: component_type is deprecated, use type instead
downstream_nodes
exec_properties
id: Node id, unique across all TFX nodes in a pipeline.

If instance name is available, node_id will be: . otherwise, node_id will be:
inputs
outputs
type
upstream_nodes

Raises:

ValueError: - When both or neither of 'module_file' and user function (e.g., trainer_fn and run_fn) is supplied.
- When both or neither of 'examples' and 'transformed_examples' is supplied.
- When 'transformed_examples' is supplied but 'transform_graph' is not supplied.

Child Classes

class DRIVER_CLASS

class SPEC_CLASS

Methods

`add_downstream_node`

View source

add_downstream_node(
    downstream_node
)

`add_upstream_node`

View source

add_upstream_node(
    upstream_node
)

`from_json_dict`

View source

@classmethod
from_json_dict(
    cls, dict_data
)

Convert from dictionary data to an object.

`get_id`

View source

@classmethod
get_id(
    cls, instance_name=None
)

Gets the id of a node.

This can be used during pipeline authoring time. For example: from tfx.components import Trainer

resolver = ResolverNode(..., model=Channel( type=Model, producer_component_id=Trainer.get_id('my_trainer')))

Args:

instance_name: (Optional) instance name of a node. If given, the instance name will be taken into consideration when generating the id.

Returns:

an id for the node.

`to_json_dict`

View source

to_json_dict()

Convert from an object to a JSON serializable dictionary.

Class Variables

EXECUTOR_SPEC