► 開發者指南 / 撰寫您自己的回呼函數

撰寫您自己的回呼函數

作者： Rick Chao, Francois Chollet
建立日期 2019/03/20
上次修改日期 2023/06/25
描述： 撰寫新的 Keras 回呼函數的完整指南。

簡介

回呼函數是在訓練、評估或推論期間客製化 Keras 模型行為的強大工具。範例包括 keras.callbacks.TensorBoard，可使用 TensorBoard 可視化訓練進度和結果，或是 keras.callbacks.ModelCheckpoint，可在訓練期間定期儲存您的模型。

在本指南中，您將學習什麼是 Keras 回呼函數、它可以做什麼，以及如何建立自己的回呼函數。我們提供了一些簡單回呼應用程式的示範，讓您開始使用。

設定

import numpy as np
import keras

Keras 回呼函數概觀

所有回呼函數都是 keras.callbacks.Callback 類的子類別，並覆寫在訓練、測試和預測的各個階段呼叫的一組方法。回呼函數可用於在訓練期間檢視模型的內部狀態和統計資料。

您可以將回呼函數的列表 (作為關鍵字引數 callbacks) 傳遞至下列模型方法

keras.Model.fit()
keras.Model.evaluate()
keras.Model.predict()

回呼函數方法概觀

全域方法

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

在 fit/evaluate/predict 開始時呼叫。

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

在 fit/evaluate/predict 結束時呼叫。

用於訓練/測試/預測的批次層級方法

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

在訓練/測試/預測期間處理批次之前立即呼叫。

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

在訓練/測試/預測批次結束時呼叫。在此方法中，logs 是一個包含指標結果的字典。

週期層級方法 (僅限訓練)

`on_epoch_begin(self, epoch, logs=None)`

在訓練期間的週期開始時呼叫。

`on_epoch_end(self, epoch, logs=None)`

在訓練期間的週期結束時呼叫。

基本範例

讓我們來看一個具體的範例。首先，讓我們匯入 tensorflow 並定義一個簡單的循序 Keras 模型

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

然後，從 Keras 資料集 API 載入 MNIST 資料進行訓練和測試

# Load example MNIST data and pre-process it
(x_train, y_train), (x_test, y_test) = 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]

現在，定義一個簡單的自訂回呼函數，以記錄

當 fit/evaluate/predict 開始和結束時
當每個週期開始和結束時
當每個訓練批次開始和結束時
當每個評估 (測試) 批次開始和結束時
當每個推論 (預測) 批次開始和結束時

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))

讓我們試試看

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']
 1/8 ━━[37m━━━━━━━━━━━━━━━━━━  0s 13ms/step...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: []
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step

`logs` 字典的用法

logs 字典包含在批次或週期結束時的損失值和所有指標。範例包括損失和平均絕對誤差。

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   29.25.
Up to batch 1, the average loss is  485.36.
Up to batch 2, the average loss is  330.94.
Up to batch 3, the average loss is  250.62.
Up to batch 4, the average loss is  202.20.
Up to batch 5, the average loss is  169.51.
Up to batch 6, the average loss is  145.98.
Up to batch 7, the average loss is  128.48.
The average loss for epoch 0 is  128.48 and mean absolute error is    6.01.
Up to batch 0, the average loss is    5.10.
Up to batch 1, the average loss is    4.80.
Up to batch 2, the average loss is    4.96.
Up to batch 3, the average loss is    4.96.
Up to batch 4, the average loss is    4.82.
Up to batch 5, the average loss is    4.69.
Up to batch 6, the average loss is    4.51.
Up to batch 7, the average loss is    4.53.
The average loss for epoch 1 is    4.53 and mean absolute error is    1.72.
Up to batch 0, the average loss is    5.08.
Up to batch 1, the average loss is    4.66.
Up to batch 2, the average loss is    4.64.
Up to batch 3, the average loss is    4.72.
Up to batch 4, the average loss is    4.82.
Up to batch 5, the average loss is    4.83.
Up to batch 6, the average loss is    4.77.
Up to batch 7, the average loss is    4.72.

`self.model` 屬性的用法

除了在呼叫其中一個方法時接收記錄資訊外，回呼函數還可存取與目前訓練/評估/推論回合相關聯的模型：self.model。

以下是您可以使用回呼函數中的 self.model 執行的一些操作

設定 self.model.stop_training = True 以立即中斷訓練。
變更最佳化工具的超參數 (可作為 self.model.optimizer)，例如 self.model.optimizer.learning_rate。
定期儲存模型。
在每個週期結束時，記錄幾個測試範例的 model.predict() 輸出，以便在訓練期間用作健全性檢查。
在每個週期結束時，擷取中間特徵的可視化結果，以監控模型隨時間學習的內容。
等等。

讓我們在幾個範例中看看實際運作情況。

Keras 回呼應用程式範例

在損失達到最小值時提早停止

第一個範例顯示建立一個 Callback，當損失達到最小值時，透過設定屬性 self.model.stop_training (布林值) 來停止訓練。您也可以選擇性地提供引數 patience，以指定在達到局部最小值後我們應該等待多少個週期才停止。

keras.callbacks.EarlyStopping 提供更完整和通用的實作。

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().__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(f"Epoch {self.stopped_epoch + 1}: early stopping")


model = get_model()
model.fit(
    x_train,
    y_train,
    batch_size=64,
    epochs=30,
    verbose=0,
    callbacks=[LossAndErrorPrintingCallback(), EarlyStoppingAtMinLoss()],
)

Up to batch 0, the average loss is   25.57.
Up to batch 1, the average loss is  471.66.
Up to batch 2, the average loss is  322.55.
Up to batch 3, the average loss is  243.88.
Up to batch 4, the average loss is  196.53.
Up to batch 5, the average loss is  165.02.
Up to batch 6, the average loss is  142.34.
Up to batch 7, the average loss is  125.17.
Up to batch 8, the average loss is  111.83.
Up to batch 9, the average loss is  101.35.
Up to batch 10, the average loss is   92.60.
Up to batch 11, the average loss is   85.16.
Up to batch 12, the average loss is   79.02.
Up to batch 13, the average loss is   73.71.
Up to batch 14, the average loss is   69.23.
Up to batch 15, the average loss is   65.26.
The average loss for epoch 0 is   65.26 and mean absolute error is    3.89.
Up to batch 0, the average loss is    3.92.
Up to batch 1, the average loss is    4.34.
Up to batch 2, the average loss is    5.39.
Up to batch 3, the average loss is    6.58.
Up to batch 4, the average loss is   10.55.
Up to batch 5, the average loss is   19.29.
Up to batch 6, the average loss is   31.58.
Up to batch 7, the average loss is   38.20.
Up to batch 8, the average loss is   41.96.
Up to batch 9, the average loss is   41.30.
Up to batch 10, the average loss is   39.31.
Up to batch 11, the average loss is   37.09.
Up to batch 12, the average loss is   35.08.
Up to batch 13, the average loss is   33.27.
Up to batch 14, the average loss is   31.54.
Up to batch 15, the average loss is   30.00.
The average loss for epoch 1 is   30.00 and mean absolute error is    4.23.
Up to batch 0, the average loss is    5.70.
Up to batch 1, the average loss is    6.90.
Up to batch 2, the average loss is    7.74.
Up to batch 3, the average loss is    8.85.
Up to batch 4, the average loss is   12.53.
Up to batch 5, the average loss is   21.55.
Up to batch 6, the average loss is   35.70.
Up to batch 7, the average loss is   44.16.
Up to batch 8, the average loss is   44.82.
Up to batch 9, the average loss is   43.07.
Up to batch 10, the average loss is   40.51.
Up to batch 11, the average loss is   38.44.
Up to batch 12, the average loss is   36.69.
Up to batch 13, the average loss is   34.77.
Up to batch 14, the average loss is   32.97.
Up to batch 15, the average loss is   31.32.
The average loss for epoch 2 is   31.32 and mean absolute error is    4.39.
Restoring model weights from the end of the best epoch.
Epoch 3: early stopping

<keras.src.callbacks.history.History at 0x1187b7430>

學習率排程

在此範例中，我們示範如何使用自訂的回呼函數，在訓練過程中動態變更最佳化工具的學習率。

如需更多通用的實作，請參閱 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().__init__()
        self.schedule = schedule

    def on_epoch_begin(self, epoch, logs=None):
        if not hasattr(self.model.optimizer, "learning_rate"):
            raise ValueError('Optimizer must have a "learning_rate" attribute.')
        # Get the current learning rate from model's optimizer.
        lr = 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
        self.model.optimizer.learning_rate = scheduled_lr
        print(f"\nEpoch {epoch}: Learning rate is {float(np.array(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,
    epochs=15,
    verbose=0,
    callbacks=[
        LossAndErrorPrintingCallback(),
        CustomLearningRateScheduler(lr_schedule),
    ],
)

Epoch 0: Learning rate is 0.10000000149011612.
Up to batch 0, the average loss is   27.90.
Up to batch 1, the average loss is  439.49.
Up to batch 2, the average loss is  302.08.
Up to batch 3, the average loss is  228.83.
Up to batch 4, the average loss is  184.97.
Up to batch 5, the average loss is  155.25.
Up to batch 6, the average loss is  134.03.
Up to batch 7, the average loss is  118.29.
Up to batch 8, the average loss is  105.65.
Up to batch 9, the average loss is   95.53.
Up to batch 10, the average loss is   87.25.
Up to batch 11, the average loss is   80.33.
Up to batch 12, the average loss is   74.48.
Up to batch 13, the average loss is   69.46.
Up to batch 14, the average loss is   65.05.
Up to batch 15, the average loss is   61.31.
The average loss for epoch 0 is   61.31 and mean absolute error is    3.85.

Epoch 1: Learning rate is 0.10000000149011612.
Up to batch 0, the average loss is   57.96.
Up to batch 1, the average loss is   55.11.
Up to batch 2, the average loss is   52.81.
Up to batch 3, the average loss is   51.06.
Up to batch 4, the average loss is   50.58.
Up to batch 5, the average loss is   51.49.
Up to batch 6, the average loss is   53.24.
Up to batch 7, the average loss is   54.20.
Up to batch 8, the average loss is   54.39.
Up to batch 9, the average loss is   54.31.
Up to batch 10, the average loss is   53.83.
Up to batch 11, the average loss is   52.93.
Up to batch 12, the average loss is   51.73.
Up to batch 13, the average loss is   50.34.
Up to batch 14, the average loss is   48.94.
Up to batch 15, the average loss is   47.65.
The average loss for epoch 1 is   47.65 and mean absolute error is    4.30.

Epoch 2: Learning rate is 0.10000000149011612.
Up to batch 0, the average loss is   46.38.
Up to batch 1, the average loss is   45.16.
Up to batch 2, the average loss is   44.03.
Up to batch 3, the average loss is   43.11.
Up to batch 4, the average loss is   42.52.
Up to batch 5, the average loss is   42.32.
Up to batch 6, the average loss is   43.06.
Up to batch 7, the average loss is   44.58.
Up to batch 8, the average loss is   45.33.
Up to batch 9, the average loss is   45.15.
Up to batch 10, the average loss is   44.59.
Up to batch 11, the average loss is   43.88.
Up to batch 12, the average loss is   43.17.
Up to batch 13, the average loss is   42.40.
Up to batch 14, the average loss is   41.74.
Up to batch 15, the average loss is   41.19.
The average loss for epoch 2 is   41.19 and mean absolute error is    4.27.

Epoch 3: Learning rate is 0.05.
Up to batch 0, the average loss is   40.85.
Up to batch 1, the average loss is   40.11.
Up to batch 2, the average loss is   39.38.
Up to batch 3, the average loss is   38.69.
Up to batch 4, the average loss is   38.01.
Up to batch 5, the average loss is   37.38.
Up to batch 6, the average loss is   36.77.
Up to batch 7, the average loss is   36.18.
Up to batch 8, the average loss is   35.61.
Up to batch 9, the average loss is   35.08.
Up to batch 10, the average loss is   34.54.
Up to batch 11, the average loss is   34.04.
Up to batch 12, the average loss is   33.56.
Up to batch 13, the average loss is   33.08.
Up to batch 14, the average loss is   32.64.
Up to batch 15, the average loss is   32.25.
The average loss for epoch 3 is   32.25 and mean absolute error is    3.64.

Epoch 4: Learning rate is 0.05000000074505806.
Up to batch 0, the average loss is   31.83.
Up to batch 1, the average loss is   31.42.
Up to batch 2, the average loss is   31.05.
Up to batch 3, the average loss is   30.72.
Up to batch 4, the average loss is   30.49.
Up to batch 5, the average loss is   30.37.
Up to batch 6, the average loss is   30.15.
Up to batch 7, the average loss is   29.94.
Up to batch 8, the average loss is   29.75.
Up to batch 9, the average loss is   29.56.
Up to batch 10, the average loss is   29.27.
Up to batch 11, the average loss is   28.96.
Up to batch 12, the average loss is   28.67.
Up to batch 13, the average loss is   28.39.
Up to batch 14, the average loss is   28.11.
Up to batch 15, the average loss is   27.80.
The average loss for epoch 4 is   27.80 and mean absolute error is    3.43.

Epoch 5: Learning rate is 0.05000000074505806.
Up to batch 0, the average loss is   27.51.
Up to batch 1, the average loss is   27.25.
Up to batch 2, the average loss is   27.05.
Up to batch 3, the average loss is   26.88.
Up to batch 4, the average loss is   26.76.
Up to batch 5, the average loss is   26.60.
Up to batch 6, the average loss is   26.44.
Up to batch 7, the average loss is   26.25.
Up to batch 8, the average loss is   26.08.
Up to batch 9, the average loss is   25.89.
Up to batch 10, the average loss is   25.71.
Up to batch 11, the average loss is   25.48.
Up to batch 12, the average loss is   25.26.
Up to batch 13, the average loss is   25.03.
Up to batch 14, the average loss is   24.81.
Up to batch 15, the average loss is   24.58.
The average loss for epoch 5 is   24.58 and mean absolute error is    3.25.

Epoch 6: Learning rate is 0.01.
Up to batch 0, the average loss is   24.36.
Up to batch 1, the average loss is   24.14.
Up to batch 2, the average loss is   23.93.
Up to batch 3, the average loss is   23.71.
Up to batch 4, the average loss is   23.52.
Up to batch 5, the average loss is   23.32.
Up to batch 6, the average loss is   23.12.
Up to batch 7, the average loss is   22.93.
Up to batch 8, the average loss is   22.74.
Up to batch 9, the average loss is   22.55.
Up to batch 10, the average loss is   22.37.
Up to batch 11, the average loss is   22.19.
Up to batch 12, the average loss is   22.01.
Up to batch 13, the average loss is   21.83.
Up to batch 14, the average loss is   21.67.
Up to batch 15, the average loss is   21.50.
The average loss for epoch 6 is   21.50 and mean absolute error is    2.98.

Epoch 7: Learning rate is 0.009999999776482582.
Up to batch 0, the average loss is   21.33.
Up to batch 1, the average loss is   21.17.
Up to batch 2, the average loss is   21.01.
Up to batch 3, the average loss is   20.85.
Up to batch 4, the average loss is   20.71.
Up to batch 5, the average loss is   20.57.
Up to batch 6, the average loss is   20.41.
Up to batch 7, the average loss is   20.27.
Up to batch 8, the average loss is   20.13.
Up to batch 9, the average loss is   19.98.
Up to batch 10, the average loss is   19.83.
Up to batch 11, the average loss is   19.69.
Up to batch 12, the average loss is   19.57.
Up to batch 13, the average loss is   19.44.
Up to batch 14, the average loss is   19.32.
Up to batch 15, the average loss is   19.19.
The average loss for epoch 7 is   19.19 and mean absolute error is    2.77.

Epoch 8: Learning rate is 0.009999999776482582.
Up to batch 0, the average loss is   19.07.
Up to batch 1, the average loss is   18.95.
Up to batch 2, the average loss is   18.83.
Up to batch 3, the average loss is   18.70.
Up to batch 4, the average loss is   18.58.
Up to batch 5, the average loss is   18.46.
Up to batch 6, the average loss is   18.35.
Up to batch 7, the average loss is   18.24.
Up to batch 8, the average loss is   18.12.
Up to batch 9, the average loss is   18.01.
Up to batch 10, the average loss is   17.90.
Up to batch 11, the average loss is   17.79.
Up to batch 12, the average loss is   17.68.
Up to batch 13, the average loss is   17.58.
Up to batch 14, the average loss is   17.48.
Up to batch 15, the average loss is   17.38.
The average loss for epoch 8 is   17.38 and mean absolute error is    2.61.

Epoch 9: Learning rate is 0.005.
Up to batch 0, the average loss is   17.28.
Up to batch 1, the average loss is   17.18.
Up to batch 2, the average loss is   17.08.
Up to batch 3, the average loss is   16.99.
Up to batch 4, the average loss is   16.90.
Up to batch 5, the average loss is   16.80.
Up to batch 6, the average loss is   16.71.
Up to batch 7, the average loss is   16.62.
Up to batch 8, the average loss is   16.53.
Up to batch 9, the average loss is   16.44.
Up to batch 10, the average loss is   16.35.
Up to batch 11, the average loss is   16.26.
Up to batch 12, the average loss is   16.17.
Up to batch 13, the average loss is   16.09.
Up to batch 14, the average loss is   16.00.
Up to batch 15, the average loss is   15.92.
The average loss for epoch 9 is   15.92 and mean absolute error is    2.48.

Epoch 10: Learning rate is 0.004999999888241291.
Up to batch 0, the average loss is   15.84.
Up to batch 1, the average loss is   15.76.
Up to batch 2, the average loss is   15.68.
Up to batch 3, the average loss is   15.61.
Up to batch 4, the average loss is   15.53.
Up to batch 5, the average loss is   15.45.
Up to batch 6, the average loss is   15.37.
Up to batch 7, the average loss is   15.29.
Up to batch 8, the average loss is   15.23.
Up to batch 9, the average loss is   15.15.
Up to batch 10, the average loss is   15.08.
Up to batch 11, the average loss is   15.00.
Up to batch 12, the average loss is   14.93.
Up to batch 13, the average loss is   14.86.
Up to batch 14, the average loss is   14.79.
Up to batch 15, the average loss is   14.72.
The average loss for epoch 10 is   14.72 and mean absolute error is    2.37.

Epoch 11: Learning rate is 0.004999999888241291.
Up to batch 0, the average loss is   14.65.
Up to batch 1, the average loss is   14.58.
Up to batch 2, the average loss is   14.52.
Up to batch 3, the average loss is   14.45.
Up to batch 4, the average loss is   14.39.
Up to batch 5, the average loss is   14.33.
Up to batch 6, the average loss is   14.26.
Up to batch 7, the average loss is   14.20.
Up to batch 8, the average loss is   14.14.
Up to batch 9, the average loss is   14.08.
Up to batch 10, the average loss is   14.02.
Up to batch 11, the average loss is   13.96.
Up to batch 12, the average loss is   13.90.
Up to batch 13, the average loss is   13.84.
Up to batch 14, the average loss is   13.78.
Up to batch 15, the average loss is   13.72.
The average loss for epoch 11 is   13.72 and mean absolute error is    2.27.

Epoch 12: Learning rate is 0.001.
Up to batch 0, the average loss is   13.67.
Up to batch 1, the average loss is   13.60.
Up to batch 2, the average loss is   13.55.
Up to batch 3, the average loss is   13.49.
Up to batch 4, the average loss is   13.44.
Up to batch 5, the average loss is   13.38.
Up to batch 6, the average loss is   13.33.
Up to batch 7, the average loss is   13.28.
Up to batch 8, the average loss is   13.22.
Up to batch 9, the average loss is   13.17.
Up to batch 10, the average loss is   13.12.
Up to batch 11, the average loss is   13.07.
Up to batch 12, the average loss is   13.02.
Up to batch 13, the average loss is   12.97.
Up to batch 14, the average loss is   12.92.
Up to batch 15, the average loss is   12.87.
The average loss for epoch 12 is   12.87 and mean absolute error is    2.19.

Epoch 13: Learning rate is 0.0010000000474974513.
Up to batch 0, the average loss is   12.82.
Up to batch 1, the average loss is   12.77.
Up to batch 2, the average loss is   12.72.
Up to batch 3, the average loss is   12.68.
Up to batch 4, the average loss is   12.63.
Up to batch 5, the average loss is   12.58.
Up to batch 6, the average loss is   12.53.
Up to batch 7, the average loss is   12.49.
Up to batch 8, the average loss is   12.45.
Up to batch 9, the average loss is   12.40.
Up to batch 10, the average loss is   12.35.
Up to batch 11, the average loss is   12.30.
Up to batch 12, the average loss is   12.26.
Up to batch 13, the average loss is   12.22.
Up to batch 14, the average loss is   12.17.
Up to batch 15, the average loss is   12.13.
The average loss for epoch 13 is   12.13 and mean absolute error is    2.12.

Epoch 14: Learning rate is 0.0010000000474974513.
Up to batch 0, the average loss is   12.09.
Up to batch 1, the average loss is   12.05.
Up to batch 2, the average loss is   12.01.
Up to batch 3, the average loss is   11.97.
Up to batch 4, the average loss is   11.92.
Up to batch 5, the average loss is   11.88.
Up to batch 6, the average loss is   11.84.
Up to batch 7, the average loss is   11.80.
Up to batch 8, the average loss is   11.76.
Up to batch 9, the average loss is   11.72.
Up to batch 10, the average loss is   11.68.
Up to batch 11, the average loss is   11.64.
Up to batch 12, the average loss is   11.60.
Up to batch 13, the average loss is   11.57.
Up to batch 14, the average loss is   11.54.
Up to batch 15, the average loss is   11.50.
The average loss for epoch 14 is   11.50 and mean absolute error is    2.06.

<keras.src.callbacks.history.History at 0x168619c60>