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.

Exemplos de migração: estimadores predefinidos

Veja no TensorFlow.org

Executar no Google Colab

Ver fonte no GitHub

Baixar caderno

Os estimadores pré-fabricados (ou pré-fabricados) têm sido tradicionalmente usados no TensorFlow 1 como maneiras rápidas e fáceis de treinar modelos para uma variedade de casos de uso típicos. O TensorFlow 2 fornece substitutos aproximados diretos para vários deles por meio de modelos Keras. Para aqueles estimadores enlatados que não possuem substitutos integrados do TensorFlow 2, você ainda pode criar seu próprio substituto com bastante facilidade.

Este guia mostra alguns exemplos de equivalentes diretos e substituições personalizadas para demonstrar como os modelos derivados do tf.estimator do TensorFlow 1 podem ser migrados para o TF2 com Keras.

Ou seja, este guia inclui exemplos de migração:

Do tf.estimator , Classifier ou Regressor do LinearEstimator no TensorFlow 1 para Keras tf.compat.v1.keras.models.LinearModel no TensorFlow 2
Do tf.estimator , Classifier ou DNNEstimator do Regressor no TensorFlow 1 para um Keras DNN ModelKeras personalizado no TensorFlow 2
Do tf.estimator do DNNLinearCombinedEstimator , Classifier ou regressor no Regressor 1 para tf.compat.v1.keras.models.WideDeepModel no TensorFlow 2
Do tf.estimator , Classifier ou Regressor do BoostedTreesEstimator no TensorFlow 1 ao tf.compat.v1.keras.models.WideDeepModel no TensorFlow 2

Um precursor comum para o treinamento de um modelo é o pré-processamento de recursos, que é feito para modelos do TensorFlow 1 Estimator com tf.feature_column . Para obter mais informações sobre o pré-processamento de recursos no TensorFlow 2, consulte este guia sobre como migrar colunas de recursos .

Configurar

Comece com algumas importações necessárias do TensorFlow,

pip install tensorflow_decision_forests

import keras
import pandas as pd
import tensorflow as tf
import tensorflow.compat.v1 as tf1
import tensorflow_decision_forests as tfdf

WARNING:root:TF Parameter Server distributed training not available (this is expected for the pre-build release).

prepare alguns dados simples para demonstração do conjunto de dados padrão do Titanic,

x_train = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/train.csv')
x_eval = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/eval.csv')
x_train['sex'].replace(('male', 'female'), (0, 1), inplace=True)
x_eval['sex'].replace(('male', 'female'), (0, 1), inplace=True)

x_train['alone'].replace(('n', 'y'), (0, 1), inplace=True)
x_eval['alone'].replace(('n', 'y'), (0, 1), inplace=True)

x_train['class'].replace(('First', 'Second', 'Third'), (1, 2, 3), inplace=True)
x_eval['class'].replace(('First', 'Second', 'Third'), (1, 2, 3), inplace=True)

x_train.drop(['embark_town', 'deck'], axis=1, inplace=True)
x_eval.drop(['embark_town', 'deck'], axis=1, inplace=True)

y_train = x_train.pop('survived')
y_eval = x_eval.pop('survived')

# Data setup for TensorFlow 1 with `tf.estimator`
def _input_fn():
  return tf1.data.Dataset.from_tensor_slices((dict(x_train), y_train)).batch(32)


def _eval_input_fn():
  return tf1.data.Dataset.from_tensor_slices((dict(x_eval), y_eval)).batch(32)


FEATURE_NAMES = [
    'age', 'fare', 'sex', 'n_siblings_spouses', 'parch', 'class', 'alone'
]

feature_columns = []
for fn in FEATURE_NAMES:
  feat_col = tf1.feature_column.numeric_column(fn, dtype=tf.float32)
  feature_columns.append(feat_col)

e crie um método para instanciar um otimizador de amostra simplista para usar com nossos vários modelos TensorFlow 1 Estimator e TensorFlow 2 Keras.

def create_sample_optimizer(tf_version):
  if tf_version == 'tf1':
    optimizer = lambda: tf.keras.optimizers.Ftrl(
        l1_regularization_strength=0.001,
        learning_rate=tf1.train.exponential_decay(
            learning_rate=0.1,
            global_step=tf1.train.get_global_step(),
            decay_steps=10000,
            decay_rate=0.9))
  elif tf_version == 'tf2':
    optimizer = tf.keras.optimizers.Ftrl(
        l1_regularization_strength=0.001,
        learning_rate=tf.keras.optimizers.schedules.ExponentialDecay(
            initial_learning_rate=0.1, decay_steps=10000, decay_rate=0.9))
  return optimizer

Exemplo 1: migrando do LinearEstimator

TF1: Usando o LinearEstimator

No TensorFlow 1, você pode usar tf.estimator.LinearEstimator para criar um modelo linear de linha de base para problemas de regressão e classificação.

linear_estimator = tf.estimator.LinearEstimator(
    head=tf.estimator.BinaryClassHead(),
    feature_columns=feature_columns,
    optimizer=create_sample_optimizer('tf1'))

INFO:tensorflow:Using default config.
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpvoycvffz
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpvoycvffz
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpvoycvffz', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpvoycvffz', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}

linear_estimator.train(input_fn=_input_fn, steps=100)
linear_estimator.evaluate(input_fn=_eval_input_fn, steps=10)

WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/training_util.py:401: 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/python/training/training_util.py:401: 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.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_estimator/python/estimator/canned/linear.py:1478: UserWarning: `layer.add_variable` is deprecated and will be removed in a future version. Please use `layer.add_weight` method instead.
  getter=tf.compat.v1.get_variable)
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/ftrl.py:149: calling Constant.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Call initializer instance with the dtype argument instead of passing it to the constructor
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/ftrl.py:149: calling Constant.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Call initializer instance with the dtype argument instead of passing it to the constructor
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmpvoycvffz/model.ckpt.
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmpvoycvffz/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:loss = 0.6931472, step = 0
INFO:tensorflow:loss = 0.6931472, step = 0
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmpvoycvffz/model.ckpt.
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmpvoycvffz/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Loss for final step: 0.55268794.
INFO:tensorflow:Loss for final step: 0.55268794.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:45
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:45
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmpvoycvffz/model.ckpt-20
INFO:tensorflow:Restoring parameters from /tmp/tmpvoycvffz/model.ckpt-20
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Inference Time : 0.50224s
INFO:tensorflow:Inference Time : 0.50224s
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:45
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:45
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.70075756, accuracy_baseline = 0.625, auc = 0.75472915, auc_precision_recall = 0.65362054, average_loss = 0.5759378, global_step = 20, label/mean = 0.375, loss = 0.5704812, precision = 0.6388889, prediction/mean = 0.41331062, recall = 0.46464646
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.70075756, accuracy_baseline = 0.625, auc = 0.75472915, auc_precision_recall = 0.65362054, average_loss = 0.5759378, global_step = 20, label/mean = 0.375, loss = 0.5704812, precision = 0.6388889, prediction/mean = 0.41331062, recall = 0.46464646
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmpvoycvffz/model.ckpt-20
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmpvoycvffz/model.ckpt-20
{'accuracy': 0.70075756,
 'accuracy_baseline': 0.625,
 'auc': 0.75472915,
 'auc_precision_recall': 0.65362054,
 'average_loss': 0.5759378,
 'label/mean': 0.375,
 'loss': 0.5704812,
 'precision': 0.6388889,
 'prediction/mean': 0.41331062,
 'recall': 0.46464646,
 'global_step': 20}

TF2: Usando Keras LinearModel

No TensorFlow 2, você pode criar uma instância do Keras tf.compat.v1.keras.models.LinearModel que é o substituto do tf.estimator.LinearEstimator . O caminho tf.compat.v1.keras é usado para indicar que o modelo pré-fabricado existe para compatibilidade.

linear_model = tf.compat.v1.keras.experimental.LinearModel()
linear_model.compile(loss='mse', optimizer=create_sample_optimizer('tf2'), metrics=['accuracy'])
linear_model.fit(x_train, y_train, epochs=10)
linear_model.evaluate(x_eval, y_eval, return_dict=True)

Epoch 1/10
20/20 [==============================] - 0s 2ms/step - loss: 2.8157 - accuracy: 0.6300
Epoch 2/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2758 - accuracy: 0.6427
Epoch 3/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2470 - accuracy: 0.6699
Epoch 4/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1954 - accuracy: 0.7177
Epoch 5/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1931 - accuracy: 0.7145
Epoch 6/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1816 - accuracy: 0.7496
Epoch 7/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1766 - accuracy: 0.7751
Epoch 8/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2198 - accuracy: 0.7560
Epoch 9/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1657 - accuracy: 0.7959
Epoch 10/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1738 - accuracy: 0.7959
9/9 [==============================] - 0s 2ms/step - loss: 0.2278 - accuracy: 0.6780
{'loss': 0.22778697311878204, 'accuracy': 0.6780303120613098}

Exemplo 2: Migrando do DNNEstimator

TF1: Usando DNNEstimator

No TensorFlow 1, você pode usar tf.estimator.DNNEstimator para criar um modelo DNN de linha de base para problemas de regressão e classificação.

dnn_estimator = tf.estimator.DNNEstimator(
    head=tf.estimator.BinaryClassHead(),
    feature_columns=feature_columns,
    hidden_units=[128],
    activation_fn=tf.nn.relu,
    optimizer=create_sample_optimizer('tf1'))

INFO:tensorflow:Using default config.
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmphckb8f81
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmphckb8f81
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmphckb8f81', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmphckb8f81', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}

dnn_estimator.train(input_fn=_input_fn, steps=100)
dnn_estimator.evaluate(input_fn=_eval_input_fn, steps=10)

INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmphckb8f81/model.ckpt.
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmphckb8f81/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:loss = 2.1811047, step = 0
INFO:tensorflow:loss = 2.1811047, step = 0
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmphckb8f81/model.ckpt.
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmphckb8f81/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Loss for final step: 0.5881681.
INFO:tensorflow:Loss for final step: 0.5881681.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:48
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:48
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmphckb8f81/model.ckpt-20
INFO:tensorflow:Restoring parameters from /tmp/tmphckb8f81/model.ckpt-20
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Inference Time : 0.47075s
INFO:tensorflow:Inference Time : 0.47075s
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:49
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:49
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.7083333, accuracy_baseline = 0.625, auc = 0.70716256, auc_precision_recall = 0.6146256, average_loss = 0.60399944, global_step = 20, label/mean = 0.375, loss = 0.5986442, precision = 0.6486486, prediction/mean = 0.41256863, recall = 0.4848485
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.7083333, accuracy_baseline = 0.625, auc = 0.70716256, auc_precision_recall = 0.6146256, average_loss = 0.60399944, global_step = 20, label/mean = 0.375, loss = 0.5986442, precision = 0.6486486, prediction/mean = 0.41256863, recall = 0.4848485
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmphckb8f81/model.ckpt-20
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmphckb8f81/model.ckpt-20
{'accuracy': 0.7083333,
 'accuracy_baseline': 0.625,
 'auc': 0.70716256,
 'auc_precision_recall': 0.6146256,
 'average_loss': 0.60399944,
 'label/mean': 0.375,
 'loss': 0.5986442,
 'precision': 0.6486486,
 'prediction/mean': 0.41256863,
 'recall': 0.4848485,
 'global_step': 20}

TF2: Usando Keras para criar um modelo DNN personalizado

No TensorFlow 2, você pode criar um modelo DNN personalizado para substituir um gerado por tf.estimator.DNNEstimator , com níveis semelhantes de personalização especificada pelo usuário (por exemplo, como no exemplo anterior, a capacidade de personalizar um otimizador de modelo escolhido) .

Um fluxo de trabalho semelhante pode ser usado para substituir tf.estimator.experimental.RNNEstimator por um modelo Keras RNN. O Keras fornece várias opções personalizáveis integradas por meio de tf.keras.layers.RNN , tf.keras.layers.LSTM e tf.keras.layers.GRU - veja aqui para mais detalhes.

dnn_model = tf.keras.models.Sequential(
    [tf.keras.layers.Dense(128, activation='relu'),
     tf.keras.layers.Dense(1)])

dnn_model.compile(loss='mse', optimizer=create_sample_optimizer('tf2'), metrics=['accuracy'])

dnn_model.fit(x_train, y_train, epochs=10)
dnn_model.evaluate(x_eval, y_eval, return_dict=True)

Epoch 1/10
20/20 [==============================] - 0s 2ms/step - loss: 551.2993 - accuracy: 0.5997
Epoch 2/10
20/20 [==============================] - 0s 2ms/step - loss: 16.8562 - accuracy: 0.6427
Epoch 3/10
20/20 [==============================] - 0s 2ms/step - loss: 0.3048 - accuracy: 0.7161
Epoch 4/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2475 - accuracy: 0.7416
Epoch 5/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2334 - accuracy: 0.7512
Epoch 6/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2200 - accuracy: 0.7416
Epoch 7/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2012 - accuracy: 0.7656
Epoch 8/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2025 - accuracy: 0.7624
Epoch 9/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2185 - accuracy: 0.7703
Epoch 10/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2046 - accuracy: 0.7687
9/9 [==============================] - 0s 2ms/step - loss: 0.2227 - accuracy: 0.6856
{'loss': 0.2227054387331009, 'accuracy': 0.685606062412262}

Exemplo 3: Migrando do DNNLinearCombinedEstimator

TF1: Usando DNNLinearCombinedEstimator

No TensorFlow 1, você pode usar tf.estimator.DNNLinearCombinedEstimator para criar um modelo combinado de linha de base para problemas de regressão e classificação com capacidade de personalização para seus componentes lineares e DNN.

optimizer = create_sample_optimizer('tf1')

combined_estimator = tf.estimator.DNNLinearCombinedEstimator(
    head=tf.estimator.BinaryClassHead(),
    # Wide settings
    linear_feature_columns=feature_columns,
    linear_optimizer=optimizer,
    # Deep settings
    dnn_feature_columns=feature_columns,
    dnn_hidden_units=[128],
    dnn_optimizer=optimizer)

INFO:tensorflow:Using default config.
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpwl5e5eaq
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpwl5e5eaq
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpwl5e5eaq', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpwl5e5eaq', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_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}

combined_estimator.train(input_fn=_input_fn, steps=100)
combined_estimator.evaluate(input_fn=_eval_input_fn, steps=10)

INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_estimator/python/estimator/canned/linear.py:1478: UserWarning: `layer.add_variable` is deprecated and will be removed in a future version. Please use `layer.add_weight` method instead.
  getter=tf.compat.v1.get_variable)
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmpwl5e5eaq/model.ckpt.
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmpwl5e5eaq/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:loss = 2.5475807, step = 0
INFO:tensorflow:loss = 2.5475807, step = 0
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 20...
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmpwl5e5eaq/model.ckpt.
INFO:tensorflow:Saving checkpoints for 20 into /tmp/tmpwl5e5eaq/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 20...
INFO:tensorflow:Loss for final step: 0.58060575.
INFO:tensorflow:Loss for final step: 0.58060575.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:53
INFO:tensorflow:Starting evaluation at 2022-01-29T02:21:53
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tmpwl5e5eaq/model.ckpt-20
INFO:tensorflow:Restoring parameters from /tmp/tmpwl5e5eaq/model.ckpt-20
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [1/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [2/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [3/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [4/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [5/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [6/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [7/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [8/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Evaluation [9/10]
INFO:tensorflow:Inference Time : 0.54029s
INFO:tensorflow:Inference Time : 0.54029s
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:53
INFO:tensorflow:Finished evaluation at 2022-01-29-02:21:53
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.6931818, accuracy_baseline = 0.625, auc = 0.73532283, auc_precision_recall = 0.630229, average_loss = 0.65179086, global_step = 20, label/mean = 0.375, loss = 0.63768697, precision = 0.60714287, prediction/mean = 0.4162652, recall = 0.5151515
INFO:tensorflow:Saving dict for global step 20: accuracy = 0.6931818, accuracy_baseline = 0.625, auc = 0.73532283, auc_precision_recall = 0.630229, average_loss = 0.65179086, global_step = 20, label/mean = 0.375, loss = 0.63768697, precision = 0.60714287, prediction/mean = 0.4162652, recall = 0.5151515
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmpwl5e5eaq/model.ckpt-20
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 20: /tmp/tmpwl5e5eaq/model.ckpt-20
{'accuracy': 0.6931818,
 'accuracy_baseline': 0.625,
 'auc': 0.73532283,
 'auc_precision_recall': 0.630229,
 'average_loss': 0.65179086,
 'label/mean': 0.375,
 'loss': 0.63768697,
 'precision': 0.60714287,
 'prediction/mean': 0.4162652,
 'recall': 0.5151515,
 'global_step': 20}

TF2: Usando Keras WideDeepModel

No TensorFlow 2, você pode criar uma instância do Keras tf.compat.v1.keras.models.WideDeepModel para substituir uma gerada por tf.estimator.DNNLinearCombinedEstimator , com níveis semelhantes de personalização especificada pelo usuário (por exemplo, como no exemplo anterior, a capacidade de personalizar um otimizador de modelo escolhido).

Este WideDeepModel é construído com base em um LinearModel constituinte e um modelo DNN personalizado, ambos discutidos nos dois exemplos anteriores. Um modelo linear personalizado também pode ser usado no lugar do Keras LinearModel , se desejado.

Se você gostaria de construir seu próprio modelo em vez de um estimador enlatado, veja como construir um modelo keras.Sequential . Para obter mais informações sobre treinamento personalizado e otimizadores, você também pode conferir este guia .

# Create LinearModel and DNN Model as in Examples 1 and 2
optimizer = create_sample_optimizer('tf2')

linear_model = tf.compat.v1.keras.experimental.LinearModel()
linear_model.compile(loss='mse', optimizer=optimizer, metrics=['accuracy'])
linear_model.fit(x_train, y_train, epochs=10, verbose=0)

dnn_model = tf.keras.models.Sequential(
    [tf.keras.layers.Dense(128, activation='relu'),
     tf.keras.layers.Dense(1)])
dnn_model.compile(loss='mse', optimizer=optimizer, metrics=['accuracy'])

combined_model = tf.compat.v1.keras.experimental.WideDeepModel(linear_model,
                                                               dnn_model)
combined_model.compile(
    optimizer=[optimizer, optimizer], loss='mse', metrics=['accuracy'])
combined_model.fit([x_train, x_train], y_train, epochs=10)
combined_model.evaluate(x_eval, y_eval, return_dict=True)

Epoch 1/10
20/20 [==============================] - 0s 2ms/step - loss: 1118.0448 - accuracy: 0.6715
Epoch 2/10
20/20 [==============================] - 0s 2ms/step - loss: 0.5682 - accuracy: 0.7305
Epoch 3/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2719 - accuracy: 0.7671
Epoch 4/10
20/20 [==============================] - 0s 2ms/step - loss: 0.2032 - accuracy: 0.7831
Epoch 5/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1911 - accuracy: 0.7783
Epoch 6/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1895 - accuracy: 0.7863
Epoch 7/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1882 - accuracy: 0.7863
Epoch 8/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1717 - accuracy: 0.7974
Epoch 9/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1701 - accuracy: 0.7927
Epoch 10/10
20/20 [==============================] - 0s 2ms/step - loss: 0.1684 - accuracy: 0.7990
9/9 [==============================] - 0s 2ms/step - loss: 0.1930 - accuracy: 0.7424
{'loss': 0.19299836456775665, 'accuracy': 0.7424242496490479}

Exemplo 4: Migrando do BoostedTreesEstimator

TF1: Usando BoostedTreesEstimator

No TensorFlow 1, você pode usar tf.estimator.BoostedTreesEstimator para criar uma linha de base para criar um modelo de base de gradiente usando um conjunto de árvores de decisão para problemas de regressão e classificação. Essa funcionalidade não está mais incluída no TensorFlow 2.

bt_estimator = tf1.estimator.BoostedTreesEstimator(
    head=tf.estimator.BinaryClassHead(),
    n_batches_per_layer=1,
    max_depth=10,
    n_trees=1000,
    feature_columns=feature_columns)

bt_estimator.train(input_fn=_input_fn, steps=1000)
bt_estimator.evaluate(input_fn=_eval_input_fn, steps=100)

TF2: usando florestas de decisão do TensorFlow

No TensorFlow 2, o substituto pré-empacotado mais próximo para um modelo gerado por tf.estimator.BoostedTreesEstimator é aquele criado usando tfdf.keras.GradientBoostedTreesModel , que cria uma sequência treinada sequencialmente de árvores de decisão rasas, cada uma projetada para "aprender" com os erros feitas por seus antecessores na seqüência.

GradientBoostedTreesModel oferece mais opções de personalização, permitindo a especificação de tudo, desde restrições básicas de profundidade até condições de parada antecipada. Veja aqui para obter mais detalhes do atributo GradientBoostedTreesModel .

gbt_model = tfdf.keras.GradientBoostedTreesModel(
    task=tfdf.keras.Task.CLASSIFICATION)
gbt_model.compile(metrics=['mse', 'accuracy'])

Use /tmp/tmpbr1acn2_ as temporary training directory

train_df, eval_df = x_train.copy(), x_eval.copy()
train_df['survived'], eval_df['survived'] = y_train, y_eval

train_dataset = tfdf.keras.pd_dataframe_to_tf_dataset(train_df, label='survived')
eval_dataset = tfdf.keras.pd_dataframe_to_tf_dataset(eval_df, label='survived')

gbt_model.fit(train_dataset)
gbt_model.evaluate(eval_dataset, return_dict=True)

Starting reading the dataset
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_decision_forests/keras/core.py:2036: FutureWarning: In a future version of pandas all arguments of DataFrame.drop except for the argument 'labels' will be keyword-only
  features_dataframe = dataframe.drop(label, 1)
1/1 [==============================] - ETA: 0s
Dataset read in 0:00:03.161776
Training model
Model trained in 0:00:00.102649
Compiling model
1/1 [==============================] - 3s 3s/step
[INFO kernel.cc:1153] Loading model from path
[INFO abstract_model.cc:1063] Engine "GradientBoostedTreesQuickScorerExtended" built
[INFO kernel.cc:1001] Use fast generic engine
WARNING:tensorflow:AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f95e9db4e60> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING:tensorflow:AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f95e9db4e60> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING: AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f95e9db4e60> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
1/1 [==============================] - 0s 388ms/step - loss: 0.0000e+00 - mse: 0.1308 - accuracy: 0.8144
{'loss': 0.0, 'mse': 0.13076548278331757, 'accuracy': 0.814393937587738}

No TensorFlow 2, há também outro substituto TFDF disponível para um modelo gerado por tf.estimator.BoostedTreesEstimator - tfdf.keras.RandomForestModel . RandomForestModel cria um aprendiz robusto e resistente a overfitting que consiste em uma população votante de árvores de decisão profundas, cada uma treinada em subconjuntos aleatórios do conjunto de dados de treinamento de entrada.

RandomForestModel e GradientBoostedTreesModel fornecem níveis igualmente extensos de personalização. A escolha entre eles é específica do problema e depende da sua tarefa ou aplicação.

Verifique os documentos da API para obter mais informações sobre o atributo RandomForestModel e GradientBoostedTreesModel .

rf_model = tfdf.keras.RandomForestModel(
    task=tfdf.keras.Task.CLASSIFICATION)
rf_model.compile(metrics=['mse', 'accuracy'])

Use /tmp/tmpluh2ebcj as temporary training directory

rf_model.fit(train_dataset)
rf_model.evaluate(eval_dataset, return_dict=True)

Starting reading the dataset
1/1 [==============================] - ETA: 0s
Dataset read in 0:00:00.094262
Training model
Model trained in 0:00:00.083656
Compiling model
1/1 [==============================] - 0s 260ms/step
[INFO kernel.cc:1153] Loading model from path
[INFO kernel.cc:1001] Use fast generic engine
1/1 [==============================] - 0s 123ms/step - loss: 0.0000e+00 - mse: 0.1270 - accuracy: 0.8636
{'loss': 0.0, 'mse': 0.12698587775230408, 'accuracy': 0.8636363744735718}