Pomoc chronić Wielkiej Rafy Koralowej z TensorFlow na Kaggle Dołącz Wyzwanie

Pisanie własnych callbacków

Zobacz na TensorFlow.org Uruchom w Google Colab Wyświetl źródło na GitHub Pobierz notatnik

Wstęp

Wywołanie zwrotne to potężne narzędzie do dostosowywania zachowania modelu Keras podczas uczenia, oceny lub wnioskowania. Przykłady obejmują tf.keras.callbacks.TensorBoard wizualizację postępów szkolenia i wyniki z TensorBoard lub tf.keras.callbacks.ModelCheckpoint okresowo Zapisz modelu podczas treningu.

Z tego przewodnika dowiesz się, czym jest wywołanie zwrotne Keras, co może zrobić i jak możesz zbudować własne. Na początek udostępniamy kilka wersji demonstracyjnych prostych aplikacji oddzwaniających.

Ustawiać

import tensorflow as tf
from tensorflow import keras

Przegląd wywołań zwrotnych Keras

Wszystkie callbacks podklasy keras.callbacks.Callback klasę, i zastąpić zestaw metod zwanych na różnych etapach kształcenia, i przewidywania. Wywołania zwrotne są przydatne do uzyskania wglądu w stany wewnętrzne i statystyki modelu podczas uczenia.

Można przekazać listę wywołań zwrotnych (jako kluczowych argumentów callbacks ) do następujących metod modelowych:

Przegląd metod zwrotnych

Metody globalne

on_(train|test|predict)_begin(self, logs=None)

Powołani na początku fit / evaluate / predict .

on_(train|test|predict)_end(self, logs=None)

Nazywane na końcu fit / evaluate / predict .

Metody szkolenia/testowania/przewidywania na poziomie partii

on_(train|test|predict)_batch_begin(self, batch, logs=None)

Wywoływane tuż przed przetwarzaniem partii podczas uczenia/testowania/przewidywania.

on_(train|test|predict)_batch_end(self, batch, logs=None)

Wywoływane na końcu szkolenia/testowania/przewidywania partii. W ramach tej metody, logs to dict zawierające wyniki metryki.

Metody na poziomie epoki (tylko szkolenie)

on_epoch_begin(self, epoch, logs=None)

Nazywany na początku epoki podczas treningu.

on_epoch_end(self, epoch, logs=None)

Nazywany u schyłku epoki podczas treningu.

Podstawowy przykład

Spójrzmy na konkretny przykład. Na początek zaimportujmy tensorflow i zdefiniujmy prosty model Sequential Keras:

# Define the Keras model to add callbacks to
def get_model():
    model = keras.Sequential()
    model.add(keras.layers.Dense(1, input_dim=784))
    model.compile(
        optimizer=keras.optimizers.RMSprop(learning_rate=0.1),
        loss="mean_squared_error",
        metrics=["mean_absolute_error"],
    )
    return model

Następnie załaduj dane MNIST do szkolenia i testowania z interfejsu API zestawów danych Keras:

# Load example MNIST data and pre-process it
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 784).astype("float32") / 255.0
x_test = x_test.reshape(-1, 784).astype("float32") / 255.0

# Limit the data to 1000 samples
x_train = x_train[:1000]
y_train = y_train[:1000]
x_test = x_test[:1000]
y_test = y_test[:1000]

Teraz zdefiniuj proste niestandardowe wywołanie zwrotne, które rejestruje:

  • Gdy fit / evaluate / predict starty i końce
  • Kiedy zaczyna się i kończy każda epoka
  • Kiedy zaczyna się i kończy każda partia treningowa
  • Kiedy rozpoczyna się i kończy każda partia ewaluacyjna (testowa)
  • Kiedy rozpoczyna się i kończy każda partia wnioskowania (prognozy)
class CustomCallback(keras.callbacks.Callback):
    def on_train_begin(self, logs=None):
        keys = list(logs.keys())
        print("Starting training; got log keys: {}".format(keys))

    def on_train_end(self, logs=None):
        keys = list(logs.keys())
        print("Stop training; got log keys: {}".format(keys))

    def on_epoch_begin(self, epoch, logs=None):
        keys = list(logs.keys())
        print("Start epoch {} of training; got log keys: {}".format(epoch, keys))

    def on_epoch_end(self, epoch, logs=None):
        keys = list(logs.keys())
        print("End epoch {} of training; got log keys: {}".format(epoch, keys))

    def on_test_begin(self, logs=None):
        keys = list(logs.keys())
        print("Start testing; got log keys: {}".format(keys))

    def on_test_end(self, logs=None):
        keys = list(logs.keys())
        print("Stop testing; got log keys: {}".format(keys))

    def on_predict_begin(self, logs=None):
        keys = list(logs.keys())
        print("Start predicting; got log keys: {}".format(keys))

    def on_predict_end(self, logs=None):
        keys = list(logs.keys())
        print("Stop predicting; got log keys: {}".format(keys))

    def on_train_batch_begin(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Training: start of batch {}; got log keys: {}".format(batch, keys))

    def on_train_batch_end(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Training: end of batch {}; got log keys: {}".format(batch, keys))

    def on_test_batch_begin(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Evaluating: start of batch {}; got log keys: {}".format(batch, keys))

    def on_test_batch_end(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Evaluating: end of batch {}; got log keys: {}".format(batch, keys))

    def on_predict_batch_begin(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Predicting: start of batch {}; got log keys: {}".format(batch, keys))

    def on_predict_batch_end(self, batch, logs=None):
        keys = list(logs.keys())
        print("...Predicting: end of batch {}; got log keys: {}".format(batch, keys))

Wypróbujmy to:

model = get_model()
model.fit(
    x_train,
    y_train,
    batch_size=128,
    epochs=1,
    verbose=0,
    validation_split=0.5,
    callbacks=[CustomCallback()],
)

res = model.evaluate(
    x_test, y_test, batch_size=128, verbose=0, callbacks=[CustomCallback()]
)

res = model.predict(x_test, batch_size=128, callbacks=[CustomCallback()])
Starting training; got log keys: []
Start epoch 0 of training; got log keys: []
...Training: start of batch 0; got log keys: []
...Training: end of batch 0; got log keys: ['loss', 'mean_absolute_error']
...Training: start of batch 1; got log keys: []
...Training: end of batch 1; got log keys: ['loss', 'mean_absolute_error']
...Training: start of batch 2; got log keys: []
...Training: end of batch 2; got log keys: ['loss', 'mean_absolute_error']
...Training: start of batch 3; got log keys: []
...Training: end of batch 3; got log keys: ['loss', 'mean_absolute_error']
Start testing; got log keys: []
...Evaluating: start of batch 0; got log keys: []
...Evaluating: end of batch 0; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 1; got log keys: []
...Evaluating: end of batch 1; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 2; got log keys: []
...Evaluating: end of batch 2; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 3; got log keys: []
...Evaluating: end of batch 3; got log keys: ['loss', 'mean_absolute_error']
Stop testing; got log keys: ['loss', 'mean_absolute_error']
End epoch 0 of training; got log keys: ['loss', 'mean_absolute_error', 'val_loss', 'val_mean_absolute_error']
Stop training; got log keys: ['loss', 'mean_absolute_error', 'val_loss', 'val_mean_absolute_error']
Start testing; got log keys: []
...Evaluating: start of batch 0; got log keys: []
...Evaluating: end of batch 0; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 1; got log keys: []
...Evaluating: end of batch 1; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 2; got log keys: []
...Evaluating: end of batch 2; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 3; got log keys: []
...Evaluating: end of batch 3; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 4; got log keys: []
...Evaluating: end of batch 4; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 5; got log keys: []
...Evaluating: end of batch 5; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 6; got log keys: []
...Evaluating: end of batch 6; got log keys: ['loss', 'mean_absolute_error']
...Evaluating: start of batch 7; got log keys: []
...Evaluating: end of batch 7; got log keys: ['loss', 'mean_absolute_error']
Stop testing; got log keys: ['loss', 'mean_absolute_error']
Start predicting; got log keys: []
...Predicting: start of batch 0; got log keys: []
...Predicting: end of batch 0; got log keys: ['outputs']
...Predicting: start of batch 1; got log keys: []
...Predicting: end of batch 1; got log keys: ['outputs']
...Predicting: start of batch 2; got log keys: []
...Predicting: end of batch 2; got log keys: ['outputs']
...Predicting: start of batch 3; got log keys: []
...Predicting: end of batch 3; got log keys: ['outputs']
...Predicting: start of batch 4; got log keys: []
...Predicting: end of batch 4; got log keys: ['outputs']
...Predicting: start of batch 5; got log keys: []
...Predicting: end of batch 5; got log keys: ['outputs']
...Predicting: start of batch 6; got log keys: []
...Predicting: end of batch 6; got log keys: ['outputs']
...Predicting: start of batch 7; got log keys: []
...Predicting: end of batch 7; got log keys: ['outputs']
Stop predicting; got log keys: []

Korzystanie z logs dict

logs DICT zawiera wartość strat i wszystkie wskaźniki na koniec partii lub epoki. Przykład zawiera stratę i średni błąd bezwzględny.

class LossAndErrorPrintingCallback(keras.callbacks.Callback):
    def on_train_batch_end(self, batch, logs=None):
        print(
            "Up to batch {}, the average loss is {:7.2f}.".format(batch, logs["loss"])
        )

    def on_test_batch_end(self, batch, logs=None):
        print(
            "Up to batch {}, the average loss is {:7.2f}.".format(batch, logs["loss"])
        )

    def on_epoch_end(self, epoch, logs=None):
        print(
            "The average loss for epoch {} is {:7.2f} "
            "and mean absolute error is {:7.2f}.".format(
                epoch, logs["loss"], logs["mean_absolute_error"]
            )
        )


model = get_model()
model.fit(
    x_train,
    y_train,
    batch_size=128,
    epochs=2,
    verbose=0,
    callbacks=[LossAndErrorPrintingCallback()],
)

res = model.evaluate(
    x_test,
    y_test,
    batch_size=128,
    verbose=0,
    callbacks=[LossAndErrorPrintingCallback()],
)
Up to batch 0, the average loss is   30.79.
Up to batch 1, the average loss is  459.11.
Up to batch 2, the average loss is  314.68.
Up to batch 3, the average loss is  237.97.
Up to batch 4, the average loss is  191.76.
Up to batch 5, the average loss is  160.95.
Up to batch 6, the average loss is  138.74.
Up to batch 7, the average loss is  124.85.
The average loss for epoch 0 is  124.85 and mean absolute error is    6.00.
Up to batch 0, the average loss is    5.13.
Up to batch 1, the average loss is    4.66.
Up to batch 2, the average loss is    4.71.
Up to batch 3, the average loss is    4.66.
Up to batch 4, the average loss is    4.69.
Up to batch 5, the average loss is    4.56.
Up to batch 6, the average loss is    4.77.
Up to batch 7, the average loss is    4.77.
The average loss for epoch 1 is    4.77 and mean absolute error is    1.75.
Up to batch 0, the average loss is    5.73.
Up to batch 1, the average loss is    5.04.
Up to batch 2, the average loss is    5.10.
Up to batch 3, the average loss is    5.14.
Up to batch 4, the average loss is    5.37.
Up to batch 5, the average loss is    5.24.
Up to batch 6, the average loss is    5.22.
Up to batch 7, the average loss is    5.16.

Wykorzystanie self.model atrybutu

Oprócz odbierania informacji dziennika, gdy jedna z metod ich nazywa, callbacks mieć dostęp do modelu związanego z obecnej rundy treningowej / oceny / wnioskowania: self.model .

Oto z niewielu rzeczy, które możesz zrobić z self.model w zwrotnego:

  • Zestaw self.model.stop_training = True szkolenia natychmiast przerwać.
  • Mutacji hiperparametrów optymalizatora (dostępny jako self.model.optimizer ), takie jak self.model.optimizer.learning_rate .
  • Zapisz model w odstępach czasu.
  • Nagraj wyjście model.predict() na kilka próbek testowych pod koniec każdej epoki, do wykorzystania jako kontrola poprawności podczas treningu.
  • Wyodrębnij wizualizacje cech pośrednich na koniec każdej epoki, aby monitorować, czego model uczy się w czasie.
  • itp.

Zobaczmy to w akcji na kilku przykładach.

Przykłady aplikacji oddzwaniania Keras

Wczesne zatrzymanie przy minimalnej stracie

Ten pierwszy przykład ilustruje utworzenie Callback , który zatrzymuje szkolenia po osiągnięciu minimum straty, ustawiając atrybutu self.model.stop_training (typ logiczny). Opcjonalnie można dostarczyć argumentów patience , aby określić, ile epok powinniśmy poczekać przed zatrzymaniem po osiągnięciu minimum lokalne.

tf.keras.callbacks.EarlyStopping zapewnia pełniejszy i ogólną realizację.

import numpy as np


class EarlyStoppingAtMinLoss(keras.callbacks.Callback):
    """Stop training when the loss is at its min, i.e. the loss stops decreasing.

  Arguments:
      patience: Number of epochs to wait after min has been hit. After this
      number of no improvement, training stops.
  """

    def __init__(self, patience=0):
        super(EarlyStoppingAtMinLoss, self).__init__()
        self.patience = patience
        # best_weights to store the weights at which the minimum loss occurs.
        self.best_weights = None

    def on_train_begin(self, logs=None):
        # The number of epoch it has waited when loss is no longer minimum.
        self.wait = 0
        # The epoch the training stops at.
        self.stopped_epoch = 0
        # Initialize the best as infinity.
        self.best = np.Inf

    def on_epoch_end(self, epoch, logs=None):
        current = logs.get("loss")
        if np.less(current, self.best):
            self.best = current
            self.wait = 0
            # Record the best weights if current results is better (less).
            self.best_weights = self.model.get_weights()
        else:
            self.wait += 1
            if self.wait >= self.patience:
                self.stopped_epoch = epoch
                self.model.stop_training = True
                print("Restoring model weights from the end of the best epoch.")
                self.model.set_weights(self.best_weights)

    def on_train_end(self, logs=None):
        if self.stopped_epoch > 0:
            print("Epoch %05d: early stopping" % (self.stopped_epoch + 1))


model = get_model()
model.fit(
    x_train,
    y_train,
    batch_size=64,
    steps_per_epoch=5,
    epochs=30,
    verbose=0,
    callbacks=[LossAndErrorPrintingCallback(), EarlyStoppingAtMinLoss()],
)
Up to batch 0, the average loss is   34.62.
Up to batch 1, the average loss is  405.62.
Up to batch 2, the average loss is  282.27.
Up to batch 3, the average loss is  215.95.
Up to batch 4, the average loss is  175.32.
The average loss for epoch 0 is  175.32 and mean absolute error is    8.59.
Up to batch 0, the average loss is    8.86.
Up to batch 1, the average loss is    7.31.
Up to batch 2, the average loss is    6.51.
Up to batch 3, the average loss is    6.71.
Up to batch 4, the average loss is    6.24.
The average loss for epoch 1 is    6.24 and mean absolute error is    2.06.
Up to batch 0, the average loss is    4.83.
Up to batch 1, the average loss is    5.05.
Up to batch 2, the average loss is    4.71.
Up to batch 3, the average loss is    4.41.
Up to batch 4, the average loss is    4.48.
The average loss for epoch 2 is    4.48 and mean absolute error is    1.68.
Up to batch 0, the average loss is    5.84.
Up to batch 1, the average loss is    5.73.
Up to batch 2, the average loss is    7.24.
Up to batch 3, the average loss is   10.34.
Up to batch 4, the average loss is   15.53.
The average loss for epoch 3 is   15.53 and mean absolute error is    3.20.
Restoring model weights from the end of the best epoch.
Epoch 00004: early stopping
<keras.callbacks.History at 0x7fd0843bf510>

Planowanie tempa nauki

W tym przykładzie pokazujemy, jak można użyć niestandardowego wywołania zwrotnego do dynamicznej zmiany szybkości uczenia się optymalizatora w trakcie szkolenia.

Zobacz callbacks.LearningRateScheduler do bardziej ogólnych wdrożeń.

class CustomLearningRateScheduler(keras.callbacks.Callback):
    """Learning rate scheduler which sets the learning rate according to schedule.

  Arguments:
      schedule: a function that takes an epoch index
          (integer, indexed from 0) and current learning rate
          as inputs and returns a new learning rate as output (float).
  """

    def __init__(self, schedule):
        super(CustomLearningRateScheduler, self).__init__()
        self.schedule = schedule

    def on_epoch_begin(self, epoch, logs=None):
        if not hasattr(self.model.optimizer, "lr"):
            raise ValueError('Optimizer must have a "lr" attribute.')
        # Get the current learning rate from model's optimizer.
        lr = float(tf.keras.backend.get_value(self.model.optimizer.learning_rate))
        # Call schedule function to get the scheduled learning rate.
        scheduled_lr = self.schedule(epoch, lr)
        # Set the value back to the optimizer before this epoch starts
        tf.keras.backend.set_value(self.model.optimizer.lr, scheduled_lr)
        print("\nEpoch %05d: Learning rate is %6.4f." % (epoch, scheduled_lr))


LR_SCHEDULE = [
    # (epoch to start, learning rate) tuples
    (3, 0.05),
    (6, 0.01),
    (9, 0.005),
    (12, 0.001),
]


def lr_schedule(epoch, lr):
    """Helper function to retrieve the scheduled learning rate based on epoch."""
    if epoch < LR_SCHEDULE[0][0] or epoch > LR_SCHEDULE[-1][0]:
        return lr
    for i in range(len(LR_SCHEDULE)):
        if epoch == LR_SCHEDULE[i][0]:
            return LR_SCHEDULE[i][1]
    return lr


model = get_model()
model.fit(
    x_train,
    y_train,
    batch_size=64,
    steps_per_epoch=5,
    epochs=15,
    verbose=0,
    callbacks=[
        LossAndErrorPrintingCallback(),
        CustomLearningRateScheduler(lr_schedule),
    ],
)
Epoch 00000: Learning rate is 0.1000.
Up to batch 0, the average loss is   26.55.
Up to batch 1, the average loss is  435.15.
Up to batch 2, the average loss is  298.00.
Up to batch 3, the average loss is  225.91.
Up to batch 4, the average loss is  182.66.
The average loss for epoch 0 is  182.66 and mean absolute error is    8.16.

Epoch 00001: Learning rate is 0.1000.
Up to batch 0, the average loss is    7.30.
Up to batch 1, the average loss is    6.22.
Up to batch 2, the average loss is    6.76.
Up to batch 3, the average loss is    6.37.
Up to batch 4, the average loss is    5.98.
The average loss for epoch 1 is    5.98 and mean absolute error is    2.01.

Epoch 00002: Learning rate is 0.1000.
Up to batch 0, the average loss is    4.23.
Up to batch 1, the average loss is    4.56.
Up to batch 2, the average loss is    4.81.
Up to batch 3, the average loss is    4.63.
Up to batch 4, the average loss is    4.67.
The average loss for epoch 2 is    4.67 and mean absolute error is    1.73.

Epoch 00003: Learning rate is 0.0500.
Up to batch 0, the average loss is    6.24.
Up to batch 1, the average loss is    5.62.
Up to batch 2, the average loss is    5.48.
Up to batch 3, the average loss is    5.09.
Up to batch 4, the average loss is    4.68.
The average loss for epoch 3 is    4.68 and mean absolute error is    1.77.

Epoch 00004: Learning rate is 0.0500.
Up to batch 0, the average loss is    3.38.
Up to batch 1, the average loss is    3.83.
Up to batch 2, the average loss is    3.53.
Up to batch 3, the average loss is    3.64.
Up to batch 4, the average loss is    3.76.
The average loss for epoch 4 is    3.76 and mean absolute error is    1.54.

Epoch 00005: Learning rate is 0.0500.
Up to batch 0, the average loss is    3.62.
Up to batch 1, the average loss is    3.79.
Up to batch 2, the average loss is    3.75.
Up to batch 3, the average loss is    3.83.
Up to batch 4, the average loss is    4.37.
The average loss for epoch 5 is    4.37 and mean absolute error is    1.65.

Epoch 00006: Learning rate is 0.0100.
Up to batch 0, the average loss is    6.73.
Up to batch 1, the average loss is    6.13.
Up to batch 2, the average loss is    5.11.
Up to batch 3, the average loss is    4.57.
Up to batch 4, the average loss is    4.21.
The average loss for epoch 6 is    4.21 and mean absolute error is    1.61.

Epoch 00007: Learning rate is 0.0100.
Up to batch 0, the average loss is    3.37.
Up to batch 1, the average loss is    3.83.
Up to batch 2, the average loss is    3.80.
Up to batch 3, the average loss is    3.50.
Up to batch 4, the average loss is    3.31.
The average loss for epoch 7 is    3.31 and mean absolute error is    1.42.

Epoch 00008: Learning rate is 0.0100.
Up to batch 0, the average loss is    5.33.
Up to batch 1, the average loss is    4.84.
Up to batch 2, the average loss is    4.02.
Up to batch 3, the average loss is    3.87.
Up to batch 4, the average loss is    3.85.
The average loss for epoch 8 is    3.85 and mean absolute error is    1.53.

Epoch 00009: Learning rate is 0.0050.
Up to batch 0, the average loss is    1.84.
Up to batch 1, the average loss is    2.75.
Up to batch 2, the average loss is    3.16.
Up to batch 3, the average loss is    3.52.
Up to batch 4, the average loss is    3.34.
The average loss for epoch 9 is    3.34 and mean absolute error is    1.43.

Epoch 00010: Learning rate is 0.0050.
Up to batch 0, the average loss is    2.36.
Up to batch 1, the average loss is    2.91.
Up to batch 2, the average loss is    2.63.
Up to batch 3, the average loss is    2.93.
Up to batch 4, the average loss is    3.17.
The average loss for epoch 10 is    3.17 and mean absolute error is    1.36.

Epoch 00011: Learning rate is 0.0050.
Up to batch 0, the average loss is    3.32.
Up to batch 1, the average loss is    3.02.
Up to batch 2, the average loss is    2.96.
Up to batch 3, the average loss is    2.80.
Up to batch 4, the average loss is    2.92.
The average loss for epoch 11 is    2.92 and mean absolute error is    1.32.

Epoch 00012: Learning rate is 0.0010.
Up to batch 0, the average loss is    4.11.
Up to batch 1, the average loss is    3.70.
Up to batch 2, the average loss is    3.89.
Up to batch 3, the average loss is    3.76.
Up to batch 4, the average loss is    3.45.
The average loss for epoch 12 is    3.45 and mean absolute error is    1.44.

Epoch 00013: Learning rate is 0.0010.
Up to batch 0, the average loss is    3.38.
Up to batch 1, the average loss is    3.34.
Up to batch 2, the average loss is    3.26.
Up to batch 3, the average loss is    3.56.
Up to batch 4, the average loss is    3.62.
The average loss for epoch 13 is    3.62 and mean absolute error is    1.44.

Epoch 00014: Learning rate is 0.0010.
Up to batch 0, the average loss is    2.48.
Up to batch 1, the average loss is    2.38.
Up to batch 2, the average loss is    2.76.
Up to batch 3, the average loss is    2.63.
Up to batch 4, the average loss is    2.66.
The average loss for epoch 14 is    2.66 and mean absolute error is    1.29.
<keras.callbacks.History at 0x7fd08446c290>

Wbudowane wywołania zwrotne Keras

Pamiętaj, aby sprawdzić istniejące callbacków Keras czytając docs API . Aplikacje obejmują logowanie do CSV, zapisywanie modelu, wizualizację metryk w TensorBoard i wiele więcej!