Questa pagina è stata tradotta dall'API Cloud Translation.
Switch to English

Addestra e servi un modello TensorFlow con TensorFlow Serving

Questa guida addestra un modello di rete neurale per classificare le immagini di abbigliamento, come scarpe da ginnastica e camicie , salva il modello addestrato e quindi lo serve con TensorFlow Serving . L'attenzione si concentra su TensorFlow Serving, piuttosto che sulla modellazione e l'addestramento in TensorFlow, quindi per un esempio completo incentrato sulla modellazione e l'addestramento, vedere l' esempio di classificazione di base .

Questa guida utilizza tf.keras , un'API di alto livello per creare e addestrare modelli in TensorFlow.

import sys

# Confirm that we're using Python 3
assert sys.version_info.major is 3, 'Oops, not running Python 3. Use Runtime > Change runtime type'
# TensorFlow and tf.keras
print("Installing dependencies for Colab environment")
!pip install -Uq grpcio==1.26.0

import tensorflow as tf
from tensorflow import keras

# Helper libraries
import numpy as np
import matplotlib.pyplot as plt
import os
import subprocess

print('TensorFlow version: {}'.format(tf.__version__))
Installing dependencies for Colab environment
[K     |████████████████████████████████| 2.4MB 4.6MB/s 
[?25hInstalling TensorFlow
TensorFlow 2.x selected.
TensorFlow version: 2.1.0-rc1

Crea il tuo modello

fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()

# scale the values to 0.0 to 1.0
train_images = train_images / 255.0
test_images = test_images / 255.0

# reshape for feeding into the model
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)

class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

print('\ntrain_images.shape: {}, of {}'.format(train_images.shape, train_images.dtype))
print('test_images.shape: {}, of {}'.format(test_images.shape, test_images.dtype))
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
32768/29515 [=================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26427392/26421880 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
8192/5148 [===============================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4423680/4422102 [==============================] - 0s 0us/step

train_images.shape: (60000, 28, 28, 1), of float64
test_images.shape: (10000, 28, 28, 1), of float64

Addestra e valuta il tuo modello

Usiamo la CNN più semplice possibile, poiché non siamo concentrati sulla parte di modellazione.

model = keras.Sequential([
  keras.layers.Conv2D(input_shape=(28,28,1), filters=8, kernel_size=3, 
                      strides=2, activation='relu', name='Conv1'),
  keras.layers.Flatten(),
  keras.layers.Dense(10, activation=tf.nn.softmax, name='Softmax')
])
model.summary()

testing = False
epochs = 5

model.compile(optimizer='adam', 
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(train_images, train_labels, epochs=epochs)

test_loss, test_acc = model.evaluate(test_images, test_labels)
print('\nTest accuracy: {}'.format(test_acc))
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
Conv1 (Conv2D)               (None, 13, 13, 8)         80        
_________________________________________________________________
flatten (Flatten)            (None, 1352)              0         
_________________________________________________________________
Softmax (Dense)              (None, 10)                13530     
=================================================================
Total params: 13,610
Trainable params: 13,610
Non-trainable params: 0
_________________________________________________________________
Train on 60000 samples
Epoch 1/5
60000/60000 [==============================] - 11s 185us/sample - loss: 0.5466 - accuracy: 0.8087
Epoch 2/5
60000/60000 [==============================] - 5s 79us/sample - loss: 0.4032 - accuracy: 0.8580
Epoch 3/5
60000/60000 [==============================] - 5s 76us/sample - loss: 0.3613 - accuracy: 0.8712
Epoch 4/5
60000/60000 [==============================] - 5s 75us/sample - loss: 0.3406 - accuracy: 0.8797
Epoch 5/5
60000/60000 [==============================] - 4s 75us/sample - loss: 0.3247 - accuracy: 0.8848
10000/10000 [==============================] - 1s 73us/sample - loss: 0.3510 - accuracy: 0.8747

Test accuracy: 0.8747000098228455

Salva il tuo modello

Per caricare il nostro modello addestrato in TensorFlow Serving, dobbiamo prima salvarlo nel formato SavedModel . Questo creerà un file protobuf in una gerarchia di directory ben definita e includerà un numero di versione. TensorFlow Serving ci consente di selezionare quale versione di un modello, o "servable", vogliamo utilizzare quando effettuiamo richieste di inferenza. Ogni versione verrà esportata in una sottodirectory diversa nel percorso specificato.

# Fetch the Keras session and save the model
# The signature definition is defined by the input and output tensors,
# and stored with the default serving key
import tempfile

MODEL_DIR = tempfile.gettempdir()
version = 1
export_path = os.path.join(MODEL_DIR, str(version))
print('export_path = {}\n'.format(export_path))

tf.keras.models.save_model(
    model,
    export_path,
    overwrite=True,
    include_optimizer=True,
    save_format=None,
    signatures=None,
    options=None
)

print('\nSaved model:')
!ls -l {export_path}
export_path = /tmp/1

Warning:tensorflow:From /tensorflow-2.1.0/python3.6/tensorflow_core/python/ops/resource_variable_ops.py:1786: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.
Instructions for updating:
If using Keras pass *_constraint arguments to layers.
INFO:tensorflow:Assets written to: /tmp/1/assets

Saved model:
total 84
drwxr-xr-x 2 root root  4096 Jan  7 23:15 assets
-rw-r--r-- 1 root root 74086 Jan  7 23:15 saved_model.pb
drwxr-xr-x 2 root root  4096 Jan  7 23:15 variables

Esamina il modello salvato

Useremo l'utility della riga di comando saved_model_cli per esaminare MetaGraphDefs (i modelli) e SignatureDefs (i metodi che puoi chiamare) nel nostro SavedModel. Consulta questa discussione sulla CLI di SavedModel nella Guida di TensorFlow.

saved_model_cli show --dir {export_path} --all

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['__saved_model_init_op']:
  The given SavedModel SignatureDef contains the following input(s):
  The given SavedModel SignatureDef contains the following output(s):
    outputs['__saved_model_init_op'] tensor_info:
        dtype: DT_INVALID
        shape: unknown_rank
        name: NoOp
  Method name is: 

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['Conv1_input'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 28, 28, 1)
        name: serving_default_Conv1_input:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['Softmax'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 10)
        name: StatefulPartitionedCall:0
  Method name is: tensorflow/serving/predict
WARNING:tensorflow:From /tensorflow-2.1.0/python3.6/tensorflow_core/python/ops/resource_variable_ops.py:1786: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.
Instructions for updating:
If using Keras pass *_constraint arguments to layers.

Defined Functions:
  Function Name: '__call__'
    Option #1
      Callable with:
        Argument #1
          Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #2
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #3
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #4
      Callable with:
        Argument #1
          Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None

  Function Name: '_default_save_signature'
    Option #1
      Callable with:
        Argument #1
          Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input')

  Function Name: 'call_and_return_all_conditional_losses'
    Option #1
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #2
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #3
      Callable with:
        Argument #1
          Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #4
      Callable with:
        Argument #1
          Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None

Questo ci dice molto sul nostro modello! In questo caso abbiamo appena addestrato il nostro modello, quindi conosciamo già gli input e gli output, ma se non lo facessimo questa sarebbe un'informazione importante. Non ci dice tutto, come il fatto che si tratti di dati di immagine in scala di grigi, ad esempio, ma è un ottimo inizio.

Servi il tuo modello con TensorFlow Serving

Aggiungi l'URI di distribuzione di TensorFlow Serving come origine del pacchetto:

Ci stiamo preparando a installare TensorFlow Serving utilizzando Aptitude poiché questo Colab viene eseguito in un ambiente Debian. Aggiungeremo il tensorflow-model-server all'elenco di pacchetti di cui Aptitude è a conoscenza. Nota che stiamo eseguendo come root.

# This is the same as you would do from your command line, but without the [arch=amd64], and no sudo
# You would instead do:
# echo "deb [arch=amd64] http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal" | sudo tee /etc/apt/sources.list.d/tensorflow-serving.list && \
# curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | sudo apt-key add -

!echo "deb http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal" | tee /etc/apt/sources.list.d/tensorflow-serving.list && \
curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | apt-key add -
!apt update
deb http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  2943  100  2943    0     0  11496      0 --:--:-- --:--:-- --:--:-- 11496
OK
Get:1 http://storage.googleapis.com/tensorflow-serving-apt stable InRelease [3,012 B]
Get:2 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran35/ InRelease [3,626 B]
Ign:3 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64  InRelease
Ign:4 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  InRelease
Hit:5 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64  Release
Get:6 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release [564 B]
Get:7 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release.gpg [833 B]
Hit:8 http://ppa.launchpad.net/graphics-drivers/ppa/ubuntu bionic InRelease
Hit:9 http://archive.ubuntu.com/ubuntu bionic InRelease
Get:10 http://security.ubuntu.com/ubuntu bionic-security InRelease [88.7 kB]
Get:11 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 Packages [354 B]
Get:12 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran35/ Packages [81.6 kB]
Get:13 http://archive.ubuntu.com/ubuntu bionic-updates InRelease [88.7 kB]
Get:14 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server-universal amd64 Packages [364 B]
Get:15 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic InRelease [15.4 kB]
Get:17 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Packages [30.4 kB]
Get:18 http://archive.ubuntu.com/ubuntu bionic-backports InRelease [74.6 kB]
Get:19 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic/main Sources [1,749 kB]
Get:20 http://security.ubuntu.com/ubuntu bionic-security/universe amd64 Packages [796 kB]
Get:21 http://archive.ubuntu.com/ubuntu bionic-updates/main amd64 Packages [1,073 kB]
Get:22 http://security.ubuntu.com/ubuntu bionic-security/main amd64 Packages [776 kB]
Get:23 http://security.ubuntu.com/ubuntu bionic-security/restricted amd64 Packages [21.3 kB]
Get:24 http://archive.ubuntu.com/ubuntu bionic-updates/multiverse amd64 Packages [10.8 kB]
Get:25 http://archive.ubuntu.com/ubuntu bionic-updates/universe amd64 Packages [1,324 kB]
Get:26 http://archive.ubuntu.com/ubuntu bionic-updates/restricted amd64 Packages [35.5 kB]
Get:27 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic/main amd64 Packages [844 kB]
Fetched 7,019 kB in 4s (1,913 kB/s)
Reading package lists... Done
Building dependency tree       
Reading state information... Done
21 packages can be upgraded. Run 'apt list --upgradable' to see them.

Installa TensorFlow Serving

Questo è tutto ciò di cui hai bisogno: una riga di comando!

apt-get install tensorflow-model-server
Reading package lists... Done
Building dependency tree       
Reading state information... Done
The following package was automatically installed and is no longer required:
  libnvidia-common-430
Use 'apt autoremove' to remove it.
The following NEW packages will be installed:
  tensorflow-model-server
0 upgraded, 1 newly installed, 0 to remove and 21 not upgraded.
Need to get 140 MB of archives.
After this operation, 0 B of additional disk space will be used.
Get:1 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 tensorflow-model-server all 2.0.0 [140 MB]
Fetched 140 MB in 2s (78.8 MB/s)
Selecting previously unselected package tensorflow-model-server.
(Reading database ... 145674 files and directories currently installed.)
Preparing to unpack .../tensorflow-model-server_2.0.0_all.deb ...
Unpacking tensorflow-model-server (2.0.0) ...
Setting up tensorflow-model-server (2.0.0) ...

Inizia a eseguire TensorFlow Serving

È qui che iniziamo a eseguire TensorFlow Serving e carichiamo il nostro modello. Dopo il caricamento possiamo iniziare a fare richieste di inferenza usando REST. Ci sono alcuni parametri importanti:

  • rest_api_port : la porta che utilizzerai per le richieste REST.
  • model_name : lo userai nell'URL delle richieste REST. Può essere qualsiasi cosa.
  • model_base_path : questo è il percorso della directory in cui hai salvato il tuo modello.
os.environ["MODEL_DIR"] = MODEL_DIR
%%bash --bg 
nohup tensorflow_model_server \
  --rest_api_port=8501 \
  --model_name=fashion_model \
  --model_base_path="${MODEL_DIR}" >server.log 2>&1

Starting job # 0 in a separate thread.

tail server.log
[warn] getaddrinfo: address family for nodename not supported
[evhttp_server.cc : 238] NET_LOG: Entering the event loop ...

Fai una richiesta al tuo modello in TensorFlow Serving

Per prima cosa, diamo un'occhiata a un esempio casuale dai nostri dati di test.

def show(idx, title):
  plt.figure()
  plt.imshow(test_images[idx].reshape(28,28))
  plt.axis('off')
  plt.title('\n\n{}'.format(title), fontdict={'size': 16})

import random
rando = random.randint(0,len(test_images)-1)
show(rando, 'An Example Image: {}'.format(class_names[test_labels[rando]]))

png

Ok, sembra interessante. Quanto è difficile per te riconoscerlo? Ora creiamo l'oggetto JSON per un batch di tre richieste di inferenza e vediamo quanto bene il nostro modello riconosce le cose:

import json
data = json.dumps({"signature_name": "serving_default", "instances": test_images[0:3].tolist()})
print('Data: {} ... {}'.format(data[:50], data[len(data)-52:]))
Data: {"signature_name": "serving_default", "instances": ...  [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]]]]}

Effettua richieste REST

Versione più recente del servable

Invieremo una richiesta di previsione come POST all'endpoint REST del nostro server e gli passeremo tre esempi. Chiederemo al nostro server di fornirci l'ultima versione del nostro servable non specificando una particolare versione.

!pip install -q requests

import requests
headers = {"content-type": "application/json"}
json_response = requests.post('http://localhost:8501/v1/models/fashion_model:predict', data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']

show(0, 'The model thought this was a {} (class {}), and it was actually a {} (class {})'.format(
  class_names[np.argmax(predictions[0])], np.argmax(predictions[0]), class_names[test_labels[0]], test_labels[0]))

png

Una versione particolare del servable

Ora specifichiamo una particolare versione del nostro servable. Dato che ne abbiamo solo uno, selezioniamo la versione 1. Analizzeremo anche tutti e tre i risultati.

headers = {"content-type": "application/json"}
json_response = requests.post('http://localhost:8501/v1/models/fashion_model/versions/1:predict', data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']

for i in range(0,3):
  show(i, 'The model thought this was a {} (class {}), and it was actually a {} (class {})'.format(
    class_names[np.argmax(predictions[i])], np.argmax(predictions[i]), class_names[test_labels[i]], test_labels[i]))

png

png

png

Importa il set di dati Fashion MNIST

Questa guida utilizza il dataset Fashion MNIST che contiene 70.000 immagini in scala di grigi in 10 categorie. Le immagini mostrano singoli capi di abbigliamento a bassa risoluzione (28 x 28 pixel), come si vede qui:

Sprite MNIST di moda
Figura 1. Campioni Fashion-MNIST (di Zalando, MIT License).

Fashion MNIST è inteso come un sostituto immediato del classico set di dati MNIST , spesso utilizzato come "Hello, World" dei programmi di machine learning per la visione artificiale . Puoi accedere a Fashion MNIST direttamente da TensorFlow, basta importare e caricare i dati.