Aprenda o que há de mais recente em aprendizado de máquina, IA generativa e muito mais no WiML Symposium 2023 Registre-se

Esta página foi traduzida pela API Cloud Translation.

Migrar salvamento de ponto de verificação

Veja no TensorFlow.org

Executar no Google Colab

Ver fonte no GitHub

Baixar caderno

Salvar continuamente o "melhor" modelo ou pesos/parâmetros do modelo traz muitos benefícios. Isso inclui a capacidade de rastrear o progresso do treinamento e carregar modelos salvos de diferentes estados salvos.

No TensorFlow 1, para configurar o salvamento do ponto de verificação durante o treinamento/validação com as APIs tf.estimator.Estimator , especifique uma programação em tf.estimator.RunConfig ou use tf.estimator.CheckpointSaverHook . Este guia demonstra como migrar desse fluxo de trabalho para as APIs do TensorFlow 2 Keras.

No TensorFlow 2, você pode configurar tf.keras.callbacks.ModelCheckpoint de várias maneiras:

Salve a "melhor" versão de acordo com uma métrica monitorada usando o parâmetro save_best_only=True , onde monitor pode ser, por exemplo, 'loss' , 'val_loss' , 'accuracy', or 'val_accuracy'`.
Salve continuamente em uma determinada frequência (usando o argumento save_freq ).
Salve os pesos/parâmetros apenas em vez de todo o modelo definindo save_weights_only como True .

Para obter mais detalhes, consulte os documentos da API tf.keras.callbacks.ModelCheckpoint e a seção Salvar pontos de verificação durante o treinamento no tutorial Salvar e carregar modelos . Saiba mais sobre o formato Checkpoint na seção Formato TF Checkpoint no guia Salvar e carregar modelos Keras . Além disso, para adicionar tolerância a falhas, você pode usar tf.keras.callbacks.BackupAndRestore ou tf.train.Checkpoint para verificação manual. Saiba mais no guia de migração de tolerância a falhas .

Os retornos de chamada Keras são objetos que são chamados em diferentes pontos durante o treinamento/avaliação/previsão nas APIs Keras Model.fit / Model.evaluate / Model.predict Model.predict . Saiba mais na seção Próximas etapas no final do guia.

Configurar

Comece com importações e um conjunto de dados simples para fins de demonstração:

import tensorflow.compat.v1 as tf1
import tensorflow as tf
import numpy as np
import tempfile

mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11493376/11490434 [==============================] - 0s 0us/step
11501568/11490434 [==============================] - 0s 0us/step

TensorFlow 1: salve pontos de verificação com APIs tf.estimator

Este exemplo do TensorFlow 1 mostra como configurar o tf.estimator.RunConfig para salvar pontos de verificação em cada etapa durante o treinamento/avaliação com as APIs tf.estimator.Estimator :

feature_columns = [tf1.feature_column.numeric_column("x", shape=[28, 28])]

config = tf1.estimator.RunConfig(save_summary_steps=1,
                                 save_checkpoints_steps=1)

path = tempfile.mkdtemp()

classifier = tf1.estimator.DNNClassifier(
    feature_columns=feature_columns,
    hidden_units=[256, 32],
    optimizer=tf1.train.AdamOptimizer(0.001),
    n_classes=10,
    dropout=0.2,
    model_dir=path,
    config = config
)

train_input_fn = tf1.estimator.inputs.numpy_input_fn(
    x={"x": x_train},
    y=y_train.astype(np.int32),
    num_epochs=10,
    batch_size=50,
    shuffle=True,
)

test_input_fn = tf1.estimator.inputs.numpy_input_fn(
    x={"x": x_test},
    y=y_test.astype(np.int32),
    num_epochs=10,
    shuffle=False
)

train_spec = tf1.estimator.TrainSpec(input_fn=train_input_fn, max_steps=10)
eval_spec = tf1.estimator.EvalSpec(input_fn=test_input_fn,
                                   steps=10,
                                   throttle_secs=0)

tf1.estimator.train_and_evaluate(estimator=classifier,
                                train_spec=train_spec,
                                eval_spec=eval_spec)

INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmplrkjo9in', '_tf_random_seed': None, '_save_summary_steps': 1, '_save_checkpoints_steps': 1, '_save_checkpoints_secs': None, '_session_config': allow_soft_placement: true
graph_options {
  rewrite_options {
    meta_optimizer_iterations: ONE
  }
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_checkpoint_save_graph_def': True, '_service': None, '_cluster_spec': ClusterSpec({}), '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
WARNING:tensorflow:From /tmp/ipykernel_20296/3980459272.py:18: The name tf.estimator.inputs is deprecated. Please use tf.compat.v1.estimator.inputs instead.

WARNING:tensorflow:From /tmp/ipykernel_20296/3980459272.py:18: The name tf.estimator.inputs.numpy_input_fn is deprecated. Please use tf.compat.v1.estimator.inputs.numpy_input_fn instead.

INFO:tensorflow:Not using Distribute Coordinator.
INFO:tensorflow:Running training and evaluation locally (non-distributed).
INFO:tensorflow:Start train and evaluate loop. The evaluate will happen after every checkpoint. Checkpoint frequency is determined based on RunConfig arguments: save_checkpoints_steps 1 or save_checkpoints_secs None.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/training_util.py:397: Variable.initialized_value (from tensorflow.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Use Variable.read_value. Variables in 2.X are initialized automatically both in eager and graph (inside tf.defun) contexts.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_estimator/python/estimator/inputs/queues/feeding_queue_runner.py:65: QueueRunner.__init__ (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_estimator/python/estimator/inputs/queues/feeding_functions.py:491: add_queue_runner (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/monitored_session.py:914: start_queue_runners (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 1...
INFO:tensorflow:Saving checkpoints for 1 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 1...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:47
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-1
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.26374s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:47
INFO:tensorflow:Saving dict for global step 1: accuracy = 0.1765625, average_loss = 2.2546134, global_step = 1, loss = 288.5905
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 1: /tmp/tmplrkjo9in/model.ckpt-1
INFO:tensorflow:loss = 118.3231, step = 0
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 2...
INFO:tensorflow:Saving checkpoints for 2 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 2...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:48
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-2
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.36662s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:48
INFO:tensorflow:Saving dict for global step 2: accuracy = 0.2859375, average_loss = 2.1868849, global_step = 2, loss = 279.92126
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 2: /tmp/tmplrkjo9in/model.ckpt-2
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 3...
INFO:tensorflow:Saving checkpoints for 3 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 3...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:48
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-3
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22792s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:48
INFO:tensorflow:Saving dict for global step 3: accuracy = 0.35078126, average_loss = 2.1220195, global_step = 3, loss = 271.6185
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 3: /tmp/tmplrkjo9in/model.ckpt-3
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 4...
INFO:tensorflow:Saving checkpoints for 4 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 4...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:49
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-4
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22387s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:49
INFO:tensorflow:Saving dict for global step 4: accuracy = 0.40234375, average_loss = 2.0655982, global_step = 4, loss = 264.39658
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 4: /tmp/tmplrkjo9in/model.ckpt-4
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 5...
INFO:tensorflow:Saving checkpoints for 5 into /tmp/tmplrkjo9in/model.ckpt.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/saver.py:1054: remove_checkpoint (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.
Instructions for updating:
Use standard file APIs to delete files with this prefix.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 5...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:49
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-5
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22548s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:49
INFO:tensorflow:Saving dict for global step 5: accuracy = 0.42421874, average_loss = 2.0072064, global_step = 5, loss = 256.92242
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 5: /tmp/tmplrkjo9in/model.ckpt-5
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 6...
INFO:tensorflow:Saving checkpoints for 6 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 6...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:50
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-6
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22806s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:50
INFO:tensorflow:Saving dict for global step 6: accuracy = 0.43984374, average_loss = 1.9473753, global_step = 6, loss = 249.26404
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 6: /tmp/tmplrkjo9in/model.ckpt-6
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 7...
INFO:tensorflow:Saving checkpoints for 7 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 7...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:50
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-7
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.23091s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:50
INFO:tensorflow:Saving dict for global step 7: accuracy = 0.44296876, average_loss = 1.8903366, global_step = 7, loss = 241.96309
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 7: /tmp/tmplrkjo9in/model.ckpt-7
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 8...
INFO:tensorflow:Saving checkpoints for 8 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 8...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:51
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-8
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22453s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:51
INFO:tensorflow:Saving dict for global step 8: accuracy = 0.44453126, average_loss = 1.8294731, global_step = 8, loss = 234.17256
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 8: /tmp/tmplrkjo9in/model.ckpt-8
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 9...
INFO:tensorflow:Saving checkpoints for 9 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 9...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:51
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-9
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.22271s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:51
INFO:tensorflow:Saving dict for global step 9: accuracy = 0.47734374, average_loss = 1.7674354, global_step = 9, loss = 226.23174
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 9: /tmp/tmplrkjo9in/model.ckpt-9
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 10...
INFO:tensorflow:Saving checkpoints for 10 into /tmp/tmplrkjo9in/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 10...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-14T02:28:52
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmplrkjo9in/model.ckpt-10
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [10/10]
INFO:tensorflow:Inference Time : 0.38483s
INFO:tensorflow:Finished evaluation at 2022-01-14-02:28:52
INFO:tensorflow:Saving dict for global step 10: accuracy = 0.5140625, average_loss = 1.7108486, global_step = 10, loss = 218.98862
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 10: /tmp/tmplrkjo9in/model.ckpt-10
INFO:tensorflow:Loss for final step: 96.2236.
({'accuracy': 0.5140625,
  'average_loss': 1.7108486,
  'loss': 218.98862,
  'global_step': 10},
 [])

%ls {classifier.model_dir}

checkpoint
eval/
events.out.tfevents.1642127326.kokoro-gcp-ubuntu-prod-837339153
graph.pbtxt
model.ckpt-10.data-00000-of-00001
model.ckpt-10.index
model.ckpt-10.meta
model.ckpt-6.data-00000-of-00001
model.ckpt-6.index
model.ckpt-6.meta
model.ckpt-7.data-00000-of-00001
model.ckpt-7.index
model.ckpt-7.meta
model.ckpt-8.data-00000-of-00001
model.ckpt-8.index
model.ckpt-8.meta
model.ckpt-9.data-00000-of-00001
model.ckpt-9.index
model.ckpt-9.meta

TensorFlow 2: salve pontos de verificação com um retorno de chamada Keras para Model.fit

No TensorFlow 2, ao usar o Keras Model.fit (ou Model.evaluate ) integrado para treinamento/avaliação, você pode configurar tf.keras.callbacks.ModelCheckpoint e passá-lo para o parâmetro callbacks de Model.fit (ou Model.evaluate ). (Saiba mais nos documentos da API e na seção Usando retornos de chamada no guia Treinamento e avaliação com o guia de métodos integrados .)

No exemplo abaixo, você usará um retorno de chamada tf.keras.callbacks.ModelCheckpoint para armazenar pontos de verificação em um diretório temporário:

def create_model():
  return tf.keras.models.Sequential([
    tf.keras.layers.Flatten(input_shape=(28, 28)),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Dense(10, activation='softmax')
  ])

model = create_model()
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'],
              steps_per_execution=10)

log_dir = tempfile.mkdtemp()

model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
    filepath=log_dir)

model.fit(x=x_train,
          y=y_train,
          epochs=10,
          validation_data=(x_test, y_test),
          callbacks=[model_checkpoint_callback])

Epoch 1/10
1840/1875 [============================>.] - ETA: 0s - loss: 0.2224 - accuracy: 0.9348
2022-01-14 02:28:56.714889: W tensorflow/python/util/util.cc:368] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.
INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 4s 2ms/step - loss: 0.2208 - accuracy: 0.9354 - val_loss: 0.1132 - val_accuracy: 0.9669
Epoch 2/10
1870/1875 [============================>.] - ETA: 0s - loss: 0.0961 - accuracy: 0.9706INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0962 - accuracy: 0.9706 - val_loss: 0.0784 - val_accuracy: 0.9753
Epoch 3/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0696 - accuracy: 0.9781INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0695 - accuracy: 0.9782 - val_loss: 0.0684 - val_accuracy: 0.9788
Epoch 4/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0529 - accuracy: 0.9826INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0531 - accuracy: 0.9826 - val_loss: 0.0671 - val_accuracy: 0.9791
Epoch 5/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0423 - accuracy: 0.9860INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0424 - accuracy: 0.9860 - val_loss: 0.0772 - val_accuracy: 0.9757
Epoch 6/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0345 - accuracy: 0.9888INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0345 - accuracy: 0.9888 - val_loss: 0.0669 - val_accuracy: 0.9811
Epoch 7/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0314 - accuracy: 0.9895INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0313 - accuracy: 0.9895 - val_loss: 0.0718 - val_accuracy: 0.9800
Epoch 8/10
1870/1875 [============================>.] - ETA: 0s - loss: 0.0298 - accuracy: 0.9899INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0298 - accuracy: 0.9899 - val_loss: 0.0632 - val_accuracy: 0.9825
Epoch 9/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0230 - accuracy: 0.9925INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0231 - accuracy: 0.9924 - val_loss: 0.0748 - val_accuracy: 0.9800
Epoch 10/10
1860/1875 [============================>.] - ETA: 0s - loss: 0.0220 - accuracy: 0.9920INFO:tensorflow:Assets written to: /tmp/tmpb85suru4/assets
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0222 - accuracy: 0.9920 - val_loss: 0.0703 - val_accuracy: 0.9825
<keras.callbacks.History at 0x7f638c204410>

%ls {model_checkpoint_callback.filepath}

assets/  keras_metadata.pb  saved_model.pb  variables/

Próximos passos

Saiba mais sobre checkpoints em:

Documentos da API: tf.keras.callbacks.ModelCheckpoint
Tutorial: Salvar e carregar modelos (a seção Salvar pontos de verificação durante o treinamento )
Guia: Salvar e carregar modelos Keras (a seção do formato TF Checkpoint )

Saiba mais sobre retornos de chamada em:

Documentos da API: tf.keras.callbacks.Callback
Guia: escrevendo seus próprios retornos de chamada
Guia: Treinamento e avaliação com os métodos integrados (a seção Usando retornos de chamada)

Você também pode achar úteis os seguintes recursos relacionados à migração:

O guia de migração de tolerância a falhas : tf.keras.callbacks.BackupAndRestore para Model.fit ou tf.train.Checkpoint e tf.train.CheckpointManager APIs para um loop de treinamento personalizado
O guia de migração de parada antecipada : tf.keras.callbacks.EarlyStopping é um retorno de chamada de parada antecipada integrado
O guia de migração do TensorBoard : O TensorBoard permite rastrear e exibir métricas
O guia de migração de retornos de chamada LoggingTensorHook e StopAtStepHook para Keras
O guia de retornos de chamada SessionRunHook para Keras

Migrar salvamento de ponto de verificação Mantenha tudo organizado com as coleções Salve e categorize o conteúdo com base nas suas preferências.

Configurar

TensorFlow 1: salve pontos de verificação com APIs tf.estimator

TensorFlow 2: salve pontos de verificação com um retorno de chamada Keras para Model.fit

Próximos passos

Migrar salvamento de ponto de verificação