Diese Seite wurde von der Cloud Translation API übersetzt.
Switch to English

Schreiben Sie Ihre eigenen Rückrufe

Ansicht auf TensorFlow.org Führen Sie in Google Colab aus Quelle auf GitHub anzeigen Notizbuch herunterladen

Einführung

Ein Rückruf ist ein leistungsstarkes Tool zum Anpassen des Verhaltens eines Keras-Modells während des Trainings, der Bewertung oder der Inferenz. Beispiele hierfür sind tf.keras.callbacks.TensorBoard zur Visualisierung des Trainingsfortschritts und der Trainingsergebnisse mit TensorBoard oder tf.keras.callbacks.ModelCheckpoint , um Ihr Modell während des Trainings regelmäßig zu speichern.

In diesem Handbuch erfahren Sie, was ein Keras-Rückruf ist, was er kann und wie Sie Ihren eigenen erstellen können. Wir bieten einige Demos einfacher Rückrufanwendungen, um Ihnen den Einstieg zu erleichtern.

Installieren

import tensorflow as tf
from tensorflow import keras

Übersicht über Keras-Rückrufe

Alle Rückrufe keras.callbacks.Callback Klasse keras.callbacks.Callback und überschreiben eine Reihe von Methoden, die in verschiedenen Phasen des Trainings, Testens und Vorhersagens aufgerufen werden. Rückrufe sind nützlich, um während des Trainings einen Überblick über interne Zustände und Statistiken des Modells zu erhalten.

Sie können eine Liste von Rückrufen (als Schlüsselwortargument- callbacks ) an die folgenden Modellmethoden übergeben:

Eine Übersicht über Rückrufmethoden

Globale Methoden

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

Wird zu Beginn der fit / evaluate / predict aufgerufen.

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

Wird am Ende der fit / evaluate / predict aufgerufen.

Batch-Level-Methoden zum Trainieren / Testen / Vorhersagen

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

Wird direkt vor der Verarbeitung einer Charge während des Trainings / Testens / Vorhersagens aufgerufen.

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

Wird am Ende des Trainings / Testens / Vorhersagen einer Charge aufgerufen. Bei dieser Methode sind logs ein Diktat, das die Metrikergebnisse enthält.

Methoden auf Epochenebene (nur Training)

on_epoch_begin(self, epoch, logs=None)

Wird zu Beginn einer Epoche während des Trainings aufgerufen.

on_epoch_end(self, epoch, logs=None)

Wird am Ende einer Epoche während des Trainings aufgerufen.

Ein einfaches Beispiel

Schauen wir uns ein konkretes Beispiel an. Importieren Sie zunächst den Tensorflow und definieren Sie ein einfaches sequentielles Keras-Modell:

# 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

Laden Sie dann die MNIST-Daten zum Trainieren und Testen aus der Keras-Datasets-API:

# 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]

Definieren Sie nun einen einfachen benutzerdefinierten Rückruf, der protokolliert:

  • Wenn fit / evaluate / predict beginnt und endet
  • Wenn jede Epoche beginnt und endet
  • Wenn jeder Trainingscharge beginnt und endet
  • Wenn jede Evaluierungs- (Test-) Charge beginnt und endet
  • Wenn jeder Inferenz- (Vorhersage-) Stapel beginnt und endet
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))

Probieren wir es aus:

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: []

Verwendung von logs diktieren

Das logs enthält den Verlustwert und alle Metriken am Ende eines Stapels oder einer Epoche. Beispiel beinhaltet den Verlust und den mittleren absoluten Fehler.

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

    def on_test_batch_end(self, batch, logs=None):
        print("For batch {}, 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()],
)
For batch 0, loss is   25.20.
For batch 1, loss is  433.41.
For batch 2, loss is  296.65.
For batch 3, loss is  225.30.
For batch 4, loss is  181.68.
For batch 5, loss is  152.45.
For batch 6, loss is  131.45.
For batch 7, loss is  118.27.
The average loss for epoch 0 is  118.27 and mean absolute error is    5.86.
For batch 0, loss is    4.57.
For batch 1, loss is    4.56.
For batch 2, loss is    4.62.
For batch 3, loss is    4.57.
For batch 4, loss is    4.56.
For batch 5, loss is    4.63.
For batch 6, loss is    4.52.
For batch 7, loss is    4.44.
The average loss for epoch 1 is    4.44 and mean absolute error is    1.70.
For batch 0, loss is    5.00.
For batch 1, loss is    4.46.
For batch 2, loss is    4.60.
For batch 3, loss is    4.53.
For batch 4, loss is    4.64.
For batch 5, loss is    4.60.
For batch 6, loss is    4.52.
For batch 7, loss is    4.45.

Verwendung des Attributs self.model

Rückrufe erhalten nicht nur Protokollinformationen, wenn eine ihrer Methoden aufgerufen wird, sondern auch Zugriff auf das Modell, das der aktuellen Trainings- / Evaluierungs- / Inferenzrunde zugeordnet ist: self.model .

Hier sind einige der Dinge, die Sie mit self.model in einem Rückruf tun können:

  • Setzen Sie self.model.stop_training = True um das Training sofort zu unterbrechen.
  • Mutieren Sie Hyperparameter des Optimierers (verfügbar als self.model.optimizer ), z. B. self.model.optimizer.learning_rate .
  • Speichern Sie das Modell in regelmäßigen Abständen.
  • model.predict() die Ausgabe von model.predict() am Ende jeder Epoche auf einigen Testmustern, um sie während des Trainings als Überprüfung der model.predict() zu verwenden.
  • Extrahieren Sie am Ende jeder Epoche Visualisierungen von Zwischenmerkmalen, um zu überwachen, was das Modell im Laufe der Zeit lernt.
  • etc.

Lassen Sie uns dies anhand einiger Beispiele in Aktion sehen.

Beispiele für Keras-Rückrufanwendungen

Frühes Anhalten bei minimalem Verlust

Dieses erste Beispiel zeigt die Erstellung eines Callback , der das Training beendet, wenn das Minimum an Verlust erreicht ist, indem das Attribut self.model.stop_training (boolean) festgelegt wird. Optional können Sie ein Argument patience um anzugeben, wie viele Epochen wir warten sollen, bevor wir anhalten, nachdem wir ein lokales Minimum erreicht haben.

tf.keras.callbacks.EarlyStopping bietet eine vollständigere und allgemeinere Implementierung.

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()],
)
For batch 0, loss is   23.44.
For batch 1, loss is  406.69.
For batch 2, loss is  279.72.
For batch 3, loss is  212.43.
For batch 4, loss is  171.98.
The average loss for epoch 0 is  171.98 and mean absolute error is    7.90.
For batch 0, loss is    4.90.
For batch 1, loss is    5.80.
For batch 2, loss is    6.08.
For batch 3, loss is    5.92.
For batch 4, loss is    5.71.
The average loss for epoch 1 is    5.71 and mean absolute error is    1.94.
For batch 0, loss is    5.28.
For batch 1, loss is    4.79.
For batch 2, loss is    4.87.
For batch 3, loss is    5.29.
For batch 4, loss is    5.65.
The average loss for epoch 2 is    5.65 and mean absolute error is    1.95.
For batch 0, loss is    8.66.
For batch 1, loss is   12.04.
For batch 2, loss is   15.36.
For batch 3, loss is   23.19.
For batch 4, loss is   37.54.
The average loss for epoch 3 is   37.54 and mean absolute error is    5.09.
Restoring model weights from the end of the best epoch.
Epoch 00004: early stopping

<tensorflow.python.keras.callbacks.History at 0x7fa82a016ac8>

Planung der Lernrate

In diesem Beispiel zeigen wir, wie ein benutzerdefinierter Rückruf verwendet werden kann, um die Lernrate des Optimierers während des Trainings dynamisch zu ändern.

Weitere allgemeine Implementierungen finden Sie unter callbacks.LearningRateScheduler .

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.
For batch 0, loss is   28.12.
For batch 1, loss is  486.39.
For batch 2, loss is  333.25.
For batch 3, loss is  251.94.
For batch 4, loss is  202.90.
The average loss for epoch 0 is  202.90 and mean absolute error is    8.33.

Epoch 00001: Learning rate is 0.1000.
For batch 0, loss is    6.74.
For batch 1, loss is    5.65.
For batch 2, loss is    5.89.
For batch 3, loss is    5.58.
For batch 4, loss is    5.61.
The average loss for epoch 1 is    5.61 and mean absolute error is    1.93.

Epoch 00002: Learning rate is 0.1000.
For batch 0, loss is    4.10.
For batch 1, loss is    3.94.
For batch 2, loss is    4.34.
For batch 3, loss is    4.33.
For batch 4, loss is    4.79.
The average loss for epoch 2 is    4.79 and mean absolute error is    1.74.

Epoch 00003: Learning rate is 0.0500.
For batch 0, loss is    6.13.
For batch 1, loss is    4.77.
For batch 2, loss is    5.06.
For batch 3, loss is    4.60.
For batch 4, loss is    4.41.
The average loss for epoch 3 is    4.41 and mean absolute error is    1.67.

Epoch 00004: Learning rate is 0.0500.
For batch 0, loss is    4.81.
For batch 1, loss is    4.73.
For batch 2, loss is    4.55.
For batch 3, loss is    4.63.
For batch 4, loss is    4.58.
The average loss for epoch 4 is    4.58 and mean absolute error is    1.70.

Epoch 00005: Learning rate is 0.0500.
For batch 0, loss is    4.45.
For batch 1, loss is    4.62.
For batch 2, loss is    4.30.
For batch 3, loss is    4.67.
For batch 4, loss is    5.11.
The average loss for epoch 5 is    5.11 and mean absolute error is    1.85.

Epoch 00006: Learning rate is 0.0100.
For batch 0, loss is   11.94.
For batch 1, loss is    9.66.
For batch 2, loss is    7.27.
For batch 3, loss is    5.80.
For batch 4, loss is    5.47.
The average loss for epoch 6 is    5.47 and mean absolute error is    1.85.

Epoch 00007: Learning rate is 0.0100.
For batch 0, loss is    4.25.
For batch 1, loss is    3.60.
For batch 2, loss is    4.19.
For batch 3, loss is    3.94.
For batch 4, loss is    3.74.
The average loss for epoch 7 is    3.74 and mean absolute error is    1.46.

Epoch 00008: Learning rate is 0.0100.
For batch 0, loss is    3.72.
For batch 1, loss is    3.55.
For batch 2, loss is    3.54.
For batch 3, loss is    3.60.
For batch 4, loss is    3.57.
The average loss for epoch 8 is    3.57 and mean absolute error is    1.49.

Epoch 00009: Learning rate is 0.0050.
For batch 0, loss is    3.55.
For batch 1, loss is    3.74.
For batch 2, loss is    3.68.
For batch 3, loss is    3.76.
For batch 4, loss is    3.57.
The average loss for epoch 9 is    3.57 and mean absolute error is    1.46.

Epoch 00010: Learning rate is 0.0050.
For batch 0, loss is    4.07.
For batch 1, loss is    3.84.
For batch 2, loss is    3.73.
For batch 3, loss is    3.46.
For batch 4, loss is    3.53.
The average loss for epoch 10 is    3.53 and mean absolute error is    1.43.

Epoch 00011: Learning rate is 0.0050.
For batch 0, loss is    3.49.
For batch 1, loss is    2.75.
For batch 2, loss is    2.67.
For batch 3, loss is    3.18.
For batch 4, loss is    3.47.
The average loss for epoch 11 is    3.47 and mean absolute error is    1.41.

Epoch 00012: Learning rate is 0.0010.
For batch 0, loss is    2.99.
For batch 1, loss is    3.01.
For batch 2, loss is    3.12.
For batch 3, loss is    3.07.
For batch 4, loss is    2.87.
The average loss for epoch 12 is    2.87 and mean absolute error is    1.35.

Epoch 00013: Learning rate is 0.0010.
For batch 0, loss is    2.93.
For batch 1, loss is    3.15.
For batch 2, loss is    3.41.
For batch 3, loss is    3.44.
For batch 4, loss is    3.44.
The average loss for epoch 13 is    3.44 and mean absolute error is    1.44.

Epoch 00014: Learning rate is 0.0010.
For batch 0, loss is    3.07.
For batch 1, loss is    3.00.
For batch 2, loss is    2.68.
For batch 3, loss is    3.21.
For batch 4, loss is    3.16.
The average loss for epoch 14 is    3.16 and mean absolute error is    1.36.

<tensorflow.python.keras.callbacks.History at 0x7fa829f018d0>

Integrierte Keras-Rückrufe

Überprüfen Sie die vorhandenen Keras-Rückrufe, indem Sie die API-Dokumente lesen. Zu den Anwendungen gehören die Protokollierung in CSV, das Speichern des Modells, die Visualisierung von Metriken in TensorBoard und vieles mehr!