Keras Tuner 简介

在 TensorFlow.org 上查看

在 Google Colab 中运行

在 GitHub 上查看源代码

下载笔记本

概述

Keras Tuner 是一个库，它可以帮助您为 TensorFlow 程序选择最佳的超参数集。为机器学习 (ML) 应用程序选择合适的超参数集的过程称为超参数调整或超参数优化。

超参数是控制训练过程和 ML 模型拓扑结构的变量。这些变量在训练过程中保持不变，并直接影响 ML 程序的性能。超参数分为两种类型

模型超参数，影响模型选择，例如隐藏层的数量和宽度
算法超参数，影响学习算法的速度和质量，例如随机梯度下降 (SGD) 的学习率和 k 近邻 (KNN) 分类器的最近邻数量

在本教程中，您将使用 Keras Tuner 为图像分类应用程序执行超参数优化。

设置

import tensorflow as tf
from tensorflow import keras

2024-07-13 04:28:31.186022: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:479] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2024-07-13 04:28:31.212189: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:10575] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2024-07-13 04:28:31.212229: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1442] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

安装并导入 Keras Tuner。

pip install -q -U keras-tuner

import keras_tuner as kt

下载并准备数据集

在本教程中，您将使用 Keras Tuner 为机器学习模型找到最佳超参数，该模型可以从Fashion MNIST 数据集中对服装图像进行分类。

加载数据。

(img_train, label_train), (img_test, label_test) = keras.datasets.fashion_mnist.load_data()

# Normalize pixel values between 0 and 1
img_train = img_train.astype('float32') / 255.0
img_test = img_test.astype('float32') / 255.0

定义模型

在构建用于超参数优化的模型时，除了模型架构之外，您还需要定义超参数搜索空间。您为超参数优化设置的模型称为超模型。

您可以通过两种方法定义超模型

使用模型构建器函数
通过子类化 Keras Tuner API 的HyperModel 类

您还可以使用两个预定义的HyperModel 类 - HyperXception 和 HyperResNet 用于计算机视觉应用程序。

在本教程中，您使用模型构建器函数来定义图像分类模型。模型构建器函数返回一个已编译的模型，并使用您在内联中定义的超参数来对模型进行超参数优化。

def model_builder(hp):
  model = keras.Sequential()
  model.add(keras.layers.Flatten(input_shape=(28, 28)))

  # Tune the number of units in the first Dense layer
  # Choose an optimal value between 32-512
  hp_units = hp.Int('units', min_value=32, max_value=512, step=32)
  model.add(keras.layers.Dense(units=hp_units, activation='relu'))
  model.add(keras.layers.Dense(10))

  # Tune the learning rate for the optimizer
  # Choose an optimal value from 0.01, 0.001, or 0.0001
  hp_learning_rate = hp.Choice('learning_rate', values=[1e-2, 1e-3, 1e-4])

  model.compile(optimizer=keras.optimizers.Adam(learning_rate=hp_learning_rate),
                loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                metrics=['accuracy'])

  return model

实例化调优器并执行超参数优化

实例化调优器以执行超参数优化。Keras Tuner 提供了四个调优器 - RandomSearch、Hyperband、BayesianOptimization 和 Sklearn。在本教程中，您将使用Hyperband 调优器。

要实例化 Hyperband 调优器，您必须指定超模型、要优化的目标以及训练的最大时期数 (max_epochs)。

tuner = kt.Hyperband(model_builder,
                     objective='val_accuracy',
                     max_epochs=10,
                     factor=3,
                     directory='my_dir',
                     project_name='intro_to_kt')

/tmpfs/src/tf_docs_env/lib/python3.9/site-packages/keras/src/layers/reshaping/flatten.py:37: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(**kwargs)

Hyperband 调整算法使用自适应资源分配和提前停止来快速收敛到高性能模型。这是通过类似体育锦标赛的支架来实现的。该算法对大量模型进行少量时期的训练，并将仅前一半性能最佳的模型带到下一轮。Hyperband 通过计算 1 + log_factor(max_epochs) 并将其四舍五入到最接近的整数来确定支架中要训练的模型数量。

创建一个回调，在验证损失达到某个值后提前停止训练。

stop_early = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=5)

运行超参数搜索。搜索方法的参数与tf.keras.model.fit 中使用的参数相同，此外还有上面的回调。

tuner.search(img_train, label_train, epochs=50, validation_split=0.2, callbacks=[stop_early])

# Get the optimal hyperparameters
best_hps=tuner.get_best_hyperparameters(num_trials=1)[0]

print(f"""
The hyperparameter search is complete. The optimal number of units in the first densely-connected
layer is {best_hps.get('units')} and the optimal learning rate for the optimizer
is {best_hps.get('learning_rate')}.
""")

Trial 30 Complete [00h 00m 25s]
val_accuracy: 0.8756666779518127

Best val_accuracy So Far: 0.8918333053588867
Total elapsed time: 00h 05m 43s

The hyperparameter search is complete. The optimal number of units in the first densely-connected
layer is 416 and the optimal learning rate for the optimizer
is 0.001.

训练模型

找到使用从搜索中获得的超参数训练模型的最佳时期数。

# Build the model with the optimal hyperparameters and train it on the data for 50 epochs
model = tuner.hypermodel.build(best_hps)
history = model.fit(img_train, label_train, epochs=50, validation_split=0.2)

val_acc_per_epoch = history.history['val_accuracy']
best_epoch = val_acc_per_epoch.index(max(val_acc_per_epoch)) + 1
print('Best epoch: %d' % (best_epoch,))

Epoch 1/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 4s 2ms/step - accuracy: 0.7834 - loss: 0.6144 - val_accuracy: 0.8597 - val_loss: 0.3901
Epoch 2/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8655 - loss: 0.3748 - val_accuracy: 0.8707 - val_loss: 0.3614
Epoch 3/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8805 - loss: 0.3306 - val_accuracy: 0.8768 - val_loss: 0.3432
Epoch 4/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8869 - loss: 0.3040 - val_accuracy: 0.8749 - val_loss: 0.3520
Epoch 5/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8938 - loss: 0.2865 - val_accuracy: 0.8823 - val_loss: 0.3322
Epoch 6/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8966 - loss: 0.2744 - val_accuracy: 0.8836 - val_loss: 0.3270
Epoch 7/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9048 - loss: 0.2525 - val_accuracy: 0.8704 - val_loss: 0.3599
Epoch 8/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9060 - loss: 0.2467 - val_accuracy: 0.8904 - val_loss: 0.3158
Epoch 9/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9122 - loss: 0.2340 - val_accuracy: 0.8894 - val_loss: 0.3205
Epoch 10/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9169 - loss: 0.2218 - val_accuracy: 0.8905 - val_loss: 0.3211
Epoch 11/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9194 - loss: 0.2123 - val_accuracy: 0.8921 - val_loss: 0.3132
Epoch 12/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9241 - loss: 0.2038 - val_accuracy: 0.8876 - val_loss: 0.3280
Epoch 13/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9245 - loss: 0.2025 - val_accuracy: 0.8917 - val_loss: 0.3190
Epoch 14/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9316 - loss: 0.1865 - val_accuracy: 0.8958 - val_loss: 0.3175
Epoch 15/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9313 - loss: 0.1841 - val_accuracy: 0.8967 - val_loss: 0.3103
Epoch 16/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9345 - loss: 0.1770 - val_accuracy: 0.8913 - val_loss: 0.3378
Epoch 17/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9378 - loss: 0.1691 - val_accuracy: 0.8930 - val_loss: 0.3432
Epoch 18/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9373 - loss: 0.1640 - val_accuracy: 0.8878 - val_loss: 0.3555
Epoch 19/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9394 - loss: 0.1596 - val_accuracy: 0.8908 - val_loss: 0.3525
Epoch 20/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9427 - loss: 0.1527 - val_accuracy: 0.8913 - val_loss: 0.3433
Epoch 21/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9431 - loss: 0.1507 - val_accuracy: 0.8912 - val_loss: 0.3503
Epoch 22/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9442 - loss: 0.1472 - val_accuracy: 0.8891 - val_loss: 0.3593
Epoch 23/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9468 - loss: 0.1425 - val_accuracy: 0.8949 - val_loss: 0.3600
Epoch 24/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9501 - loss: 0.1343 - val_accuracy: 0.8923 - val_loss: 0.3566
Epoch 25/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9496 - loss: 0.1340 - val_accuracy: 0.8878 - val_loss: 0.3949
Epoch 26/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9539 - loss: 0.1262 - val_accuracy: 0.8907 - val_loss: 0.3792
Epoch 27/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9543 - loss: 0.1209 - val_accuracy: 0.8942 - val_loss: 0.3780
Epoch 28/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9529 - loss: 0.1213 - val_accuracy: 0.8953 - val_loss: 0.3919
Epoch 29/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9576 - loss: 0.1135 - val_accuracy: 0.8949 - val_loss: 0.3889
Epoch 30/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9577 - loss: 0.1130 - val_accuracy: 0.8932 - val_loss: 0.3988
Epoch 31/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9578 - loss: 0.1093 - val_accuracy: 0.8938 - val_loss: 0.4114
Epoch 32/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9579 - loss: 0.1102 - val_accuracy: 0.8950 - val_loss: 0.4106
Epoch 33/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9595 - loss: 0.1077 - val_accuracy: 0.8942 - val_loss: 0.4095
Epoch 34/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9584 - loss: 0.1085 - val_accuracy: 0.8950 - val_loss: 0.4217
Epoch 35/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9617 - loss: 0.1033 - val_accuracy: 0.8909 - val_loss: 0.4860
Epoch 36/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9635 - loss: 0.0971 - val_accuracy: 0.8944 - val_loss: 0.4451
Epoch 37/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9646 - loss: 0.0926 - val_accuracy: 0.8857 - val_loss: 0.5176
Epoch 38/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9659 - loss: 0.0909 - val_accuracy: 0.8892 - val_loss: 0.4487
Epoch 39/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9649 - loss: 0.0969 - val_accuracy: 0.8912 - val_loss: 0.4505
Epoch 40/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9645 - loss: 0.0945 - val_accuracy: 0.8950 - val_loss: 0.4620
Epoch 41/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9669 - loss: 0.0877 - val_accuracy: 0.8902 - val_loss: 0.4589
Epoch 42/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9689 - loss: 0.0817 - val_accuracy: 0.8925 - val_loss: 0.4819
Epoch 43/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9674 - loss: 0.0841 - val_accuracy: 0.8925 - val_loss: 0.4590
Epoch 44/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9708 - loss: 0.0764 - val_accuracy: 0.8966 - val_loss: 0.4496
Epoch 45/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9706 - loss: 0.0812 - val_accuracy: 0.8899 - val_loss: 0.5248
Epoch 46/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9698 - loss: 0.0817 - val_accuracy: 0.8972 - val_loss: 0.4896
Epoch 47/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9716 - loss: 0.0757 - val_accuracy: 0.8912 - val_loss: 0.5208
Epoch 48/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9712 - loss: 0.0777 - val_accuracy: 0.8955 - val_loss: 0.4960
Epoch 49/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9711 - loss: 0.0754 - val_accuracy: 0.8902 - val_loss: 0.5836
Epoch 50/50
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9738 - loss: 0.0712 - val_accuracy: 0.8942 - val_loss: 0.5111
Best epoch: 46

重新实例化超模型，并使用上面获得的最佳时期数进行训练。

hypermodel = tuner.hypermodel.build(best_hps)

# Retrain the model
hypermodel.fit(img_train, label_train, epochs=best_epoch, validation_split=0.2)

Epoch 1/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.7859 - loss: 0.6252 - val_accuracy: 0.8511 - val_loss: 0.4150
Epoch 2/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.8626 - loss: 0.3801 - val_accuracy: 0.8575 - val_loss: 0.3849
Epoch 3/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8738 - loss: 0.3377 - val_accuracy: 0.8665 - val_loss: 0.3635
Epoch 4/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8889 - loss: 0.3028 - val_accuracy: 0.8816 - val_loss: 0.3250
Epoch 5/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8951 - loss: 0.2875 - val_accuracy: 0.8880 - val_loss: 0.3142
Epoch 6/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.8996 - loss: 0.2703 - val_accuracy: 0.8840 - val_loss: 0.3205
Epoch 7/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9039 - loss: 0.2545 - val_accuracy: 0.8768 - val_loss: 0.3439
Epoch 8/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9101 - loss: 0.2409 - val_accuracy: 0.8853 - val_loss: 0.3357
Epoch 9/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9166 - loss: 0.2261 - val_accuracy: 0.8899 - val_loss: 0.3255
Epoch 10/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9178 - loss: 0.2207 - val_accuracy: 0.8848 - val_loss: 0.3313
Epoch 11/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9206 - loss: 0.2160 - val_accuracy: 0.8901 - val_loss: 0.3188
Epoch 12/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9215 - loss: 0.2069 - val_accuracy: 0.8901 - val_loss: 0.3178
Epoch 13/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9271 - loss: 0.1945 - val_accuracy: 0.8902 - val_loss: 0.3433
Epoch 14/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9283 - loss: 0.1905 - val_accuracy: 0.8913 - val_loss: 0.3185
Epoch 15/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9310 - loss: 0.1840 - val_accuracy: 0.8965 - val_loss: 0.3224
Epoch 16/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9348 - loss: 0.1747 - val_accuracy: 0.8907 - val_loss: 0.3480
Epoch 17/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9335 - loss: 0.1739 - val_accuracy: 0.8927 - val_loss: 0.3427
Epoch 18/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9381 - loss: 0.1654 - val_accuracy: 0.8959 - val_loss: 0.3328
Epoch 19/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9411 - loss: 0.1554 - val_accuracy: 0.8956 - val_loss: 0.3311
Epoch 20/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9409 - loss: 0.1536 - val_accuracy: 0.8913 - val_loss: 0.3663
Epoch 21/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9448 - loss: 0.1483 - val_accuracy: 0.8931 - val_loss: 0.3675
Epoch 22/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9453 - loss: 0.1419 - val_accuracy: 0.8933 - val_loss: 0.3555
Epoch 23/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9477 - loss: 0.1396 - val_accuracy: 0.8903 - val_loss: 0.3795
Epoch 24/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9480 - loss: 0.1405 - val_accuracy: 0.8949 - val_loss: 0.3739
Epoch 25/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.9490 - loss: 0.1344 - val_accuracy: 0.8935 - val_loss: 0.3846
Epoch 26/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9507 - loss: 0.1301 - val_accuracy: 0.8928 - val_loss: 0.3961
Epoch 27/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9535 - loss: 0.1218 - val_accuracy: 0.8982 - val_loss: 0.3585
Epoch 28/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9540 - loss: 0.1205 - val_accuracy: 0.8923 - val_loss: 0.3853
Epoch 29/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9548 - loss: 0.1213 - val_accuracy: 0.8934 - val_loss: 0.4092
Epoch 30/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9588 - loss: 0.1122 - val_accuracy: 0.8907 - val_loss: 0.4172
Epoch 31/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9572 - loss: 0.1136 - val_accuracy: 0.8975 - val_loss: 0.3871
Epoch 32/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9589 - loss: 0.1119 - val_accuracy: 0.8909 - val_loss: 0.4302
Epoch 33/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9609 - loss: 0.1059 - val_accuracy: 0.8941 - val_loss: 0.4405
Epoch 34/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9609 - loss: 0.1037 - val_accuracy: 0.8947 - val_loss: 0.4228
Epoch 35/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9635 - loss: 0.0993 - val_accuracy: 0.8956 - val_loss: 0.4306
Epoch 36/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9640 - loss: 0.0960 - val_accuracy: 0.8944 - val_loss: 0.4330
Epoch 37/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9618 - loss: 0.0983 - val_accuracy: 0.8962 - val_loss: 0.4432
Epoch 38/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9644 - loss: 0.0937 - val_accuracy: 0.8922 - val_loss: 0.4692
Epoch 39/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9653 - loss: 0.0926 - val_accuracy: 0.8934 - val_loss: 0.4716
Epoch 40/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9648 - loss: 0.0935 - val_accuracy: 0.8926 - val_loss: 0.4649
Epoch 41/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9663 - loss: 0.0886 - val_accuracy: 0.8912 - val_loss: 0.4818
Epoch 42/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9671 - loss: 0.0877 - val_accuracy: 0.8916 - val_loss: 0.5000
Epoch 43/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9682 - loss: 0.0830 - val_accuracy: 0.8877 - val_loss: 0.5475
Epoch 44/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 1ms/step - accuracy: 0.9674 - loss: 0.0860 - val_accuracy: 0.8955 - val_loss: 0.4778
Epoch 45/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9686 - loss: 0.0798 - val_accuracy: 0.8872 - val_loss: 0.5337
Epoch 46/46
1500/1500 ━━━━━━━━━━━━━━━━━━━━ 2s 2ms/step - accuracy: 0.9705 - loss: 0.0763 - val_accuracy: 0.8948 - val_loss: 0.4820
<keras.src.callbacks.history.History at 0x7f53840b6f10>

要完成本教程，请在测试数据上评估超模型。

eval_result = hypermodel.evaluate(img_test, label_test)
print("[test loss, test accuracy]:", eval_result)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step - accuracy: 0.8913 - loss: 0.5298
[test loss, test accuracy]: [0.5244643092155457, 0.8938999772071838]

my_dir/intro_to_kt 目录包含超参数搜索期间运行的每个试验（模型配置）的详细日志和检查点。如果您重新运行超参数搜索，Keras Tuner 将使用这些日志中的现有状态来恢复搜索。要禁用此行为，在实例化调优器时传递额外的overwrite=True 参数。

总结

在本教程中，您学习了如何使用 Keras Tuner 为模型调整超参数。要了解有关 Keras Tuner 的更多信息，请查看以下其他资源

还可以查看 TensorBoard 中的HParams 仪表板，以交互方式调整模型超参数。