{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "Tce3stUlHN0L"
},
"outputs": [],
"source": [
"##### Copyright 2020 The TensorFlow Authors."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"cellView": "form",
"id": "tuOe1ymfHZPu"
},
"outputs": [],
"source": [
"#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n",
"# you may not use this file except in compliance with the License.\n",
"# You may obtain a copy of the License at\n",
"#\n",
"# https://www.apache.org/licenses/LICENSE-2.0\n",
"#\n",
"# Unless required by applicable law or agreed to in writing, software\n",
"# distributed under the License is distributed on an \"AS IS\" BASIS,\n",
"# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
"# See the License for the specific language governing permissions and\n",
"# limitations under the License."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' # 设置日志级别为ERROR，以减少警告信息\n",
"# 禁用 Gemini 的底层库（gRPC 和 Abseil）在初始化日志警告\n",
"os.environ[\"GRPC_VERBOSITY\"] = \"ERROR\"\n",
"os.environ[\"GLOG_minloglevel\"] = \"3\" # 0: INFO, 1: WARNING, 2: ERROR, 3: FATAL\n",
"os.environ[\"GLOG_minloglevel\"] = \"true\"\n",
"import logging\n",
"import tensorflow as tf\n",
"tf.get_logger().setLevel(logging.ERROR)\n",
"tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)\n",
"!export TF_FORCE_GPU_ALLOW_GROWTH=true\n",
"from pathlib import Path\n",
"\n",
"temp_dir = Path(\".temp\")\n",
"temp_dir.mkdir(parents=True, exist_ok=True)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "qFdPvlXBOdUN"
},
"source": [
"# Keras Tuner 简介"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MfBg1C5NB3X0"
},
"source": [
""
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "xHxb-dlhMIzW"
},
"source": [
"## 概述\n",
"\n",
"Keras Tuner 是一个库，可帮助您为 TensorFlow 程序选择最佳的超参数集。为您的机器学习 (ML) 应用选择正确的超参数集，这一过程称为*超参数调节*或*超调*。\n",
"\n",
"超参数是控制训练过程和 ML 模型拓扑的变量。这些变量在训练过程中保持不变，并会直接影响 ML 程序的性能。超参数有两种类型：\n",
"\n",
"1. **模型超参数**：影响模型的选择，例如隐藏层的数量和宽度\n",
"2. **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) 的学习率以及 k 近邻 (KNN) 分类器的近邻数\n",
"\n",
"在本教程中，您将使用 Keras Tuner 对图像分类应用执行超调。"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MUXex9ctTuDB"
},
"source": [
"## 设置"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"id": "IqR2PQG4ZaZ0"
},
"outputs": [],
"source": [
"import tensorflow as tf\n",
"from tensorflow import keras"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "g83Lwsy-Aq2_"
},
"source": [
"安装并导入 Keras Tuner。"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"id": "hpMLpbt9jcO6"
},
"outputs": [],
"source": [
"!pip install -q -U keras-tuner"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"id": "_leAIdFKAxAD"
},
"outputs": [],
"source": [
"import keras_tuner as kt"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ReV_UXOgCZvx"
},
"source": [
"## 下载并准备数据集\n",
"\n",
"在本教程中，您将使用 Keras Tuner 为某个对 [Fashion MNIST 数据集](https://github.com/zalandoresearch/fashion-mnist)内的服装图像进行分类的机器学习模型找到最佳超参数。"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "HljH_ENLEdHa"
},
"source": [
"加载数据。"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"id": "OHlHs9Wj_PUM"
},
"outputs": [],
"source": [
"(img_train, label_train), (img_test, label_test) = keras.datasets.fashion_mnist.load_data()"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"id": "bLVhXs3xrUD0"
},
"outputs": [],
"source": [
"# Normalize pixel values between 0 and 1\n",
"img_train = img_train.astype('float32') / 255.0\n",
"img_test = img_test.astype('float32') / 255.0"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "K5YEL2H2Ax3e"
},
"source": [
"## 定义模型\n",
"\n",
"构建用于超调的模型时，除了模型架构之外，还要定义超参数搜索空间。您为超调设置的模型称为*超模型*。\n",
"\n",
"您可以通过两种方式定义超模型：\n",
"\n",
"- 使用模型构建工具函数\n",
"- 将 Keras Tuner API 的 `HyperModel` 类子类化\n",
"\n",
"您还可以将两个预定义的 HyperModel 类 [HyperXception](https://keras.io/api/keras_tuner/hypermodels/hyper_xception/) 和 [HyperResNet](https://keras.io/api/keras_tuner/hypermodels/hyper_resnet/) 用于计算机视觉应用。\n",
"\n",
"在本教程中，您将使用模型构建工具函数来定义图像分类模型。模型构建工具函数将返回已编译的模型，并使用您以内嵌方式定义的超参数对模型进行超调。"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"id": "ZQKodC-jtsva"
},
"outputs": [],
"source": [
"def model_builder(hp):\n",
" model = keras.Sequential()\n",
" model.add(keras.layers.Flatten())\n",
"\n",
" # Tune the number of units in the first Dense layer\n",
" # Choose an optimal value between 32-512\n",
" hp_units = hp.Int('units', min_value=32, max_value=512, step=32)\n",
" model.add(keras.layers.Dense(units=hp_units, activation='relu'))\n",
" model.add(keras.layers.Dense(10))\n",
"\n",
" # Tune the learning rate for the optimizer\n",
" # Choose an optimal value from 0.01, 0.001, or 0.0001\n",
" hp_learning_rate = hp.Choice('learning_rate', values=[1e-2, 1e-3, 1e-4])\n",
"\n",
" model.compile(optimizer=keras.optimizers.Adam(learning_rate=hp_learning_rate),\n",
" loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['accuracy'])\n",
"\n",
" return model"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "0J1VYw4q3x0b"
},
"source": [
"## 实例化调节器并执行超调\n",
"\n",
"实例化调节器以执行超调。Keras Tuner 提供了四种调节器：`RandomSearch`、`Hyperband`、`BayesianOptimization` 和 `Sklearn`。在本教程中，您将使用 [Hyperband](https://arxiv.org/pdf/1603.06560.pdf) 调节器。\n",
"\n",
"要实例化 Hyperband 调节器，必须指定超模型、要优化的 `objective` 和要训练的最大周期数 (`max_epochs`)。"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"id": "oichQFly6Y46"
},
"outputs": [],
"source": [
"tuner = kt.Hyperband(model_builder,\n",
" objective='val_accuracy',\n",
" max_epochs=10,\n",
" factor=3,\n",
" directory=temp_dir/'my_dir',\n",
" project_name='intro_to_kt')"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "VaIhhdKf9VtI"
},
"source": [
"Hyperband 调节算法使用自适应资源分配和早停法来快速收敛到高性能模型。该过程采用了体育竞技争冠模式的排除法。算法会将大量模型训练多个周期，并仅将性能最高的一半模型送入下一轮训练。Hyperband 通过计算 1 + logfactor(`max_epochs`) 并将其向上舍入到最接近的整数来确定要训练的模型的数量。"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "cwhBdXx0Ekj8"
},
"source": [
"创建回调以在验证损失达到特定值后提前停止训练。"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"id": "WT9IkS9NEjLc"
},
"outputs": [],
"source": [
"stop_early = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=5)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "UKghEo15Tduy"
},
"source": [
"运行超参数搜索。除了上面的回调外，搜索方法的参数也与 `tf.keras.model.fit` 所用参数相同。"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"id": "dSBQcTHF9cKt"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Trial 30 Complete [00h 00m 40s]\n",
"val_accuracy: 0.8653333187103271\n",
"\n",
"Best val_accuracy So Far: 0.8915833234786987\n",
"Total elapsed time: 00h 09m 06s\n",
"\n",
"The hyperparameter search is complete. The optimal number of units in the first densely-connected\n",
"layer is 480 and the optimal learning rate for the optimizer\n",
"is 0.001.\n",
"\n"
]
}
],
"source": [
"tuner.search(img_train, label_train, epochs=50, validation_split=0.2, callbacks=[stop_early])\n",
"\n",
"# Get the optimal hyperparameters\n",
"best_hps=tuner.get_best_hyperparameters(num_trials=1)[0]\n",
"\n",
"print(f\"\"\"\n",
"The hyperparameter search is complete. The optimal number of units in the first densely-connected\n",
"layer is {best_hps.get('units')} and the optimal learning rate for the optimizer\n",
"is {best_hps.get('learning_rate')}.\n",
"\"\"\")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "Lak_ylf88xBv"
},
"source": [
"## 训练模型\n",
"\n",
"使用从搜索中获得的超参数找到训练模型的最佳周期数。"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"id": "McO82AXOuxXh"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 3ms/step - accuracy: 0.7844 - loss: 0.6154 - val_accuracy: 0.8453 - val_loss: 0.4136\n",
"Epoch 2/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8609 - loss: 0.3789 - val_accuracy: 0.8729 - val_loss: 0.3500\n",
"Epoch 3/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8794 - loss: 0.3273 - val_accuracy: 0.8728 - val_loss: 0.3514\n",
"Epoch 4/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8872 - loss: 0.3048 - val_accuracy: 0.8827 - val_loss: 0.3235\n",
"Epoch 5/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.8939 - loss: 0.2821 - val_accuracy: 0.8809 - val_loss: 0.3286\n",
"Epoch 6/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9030 - loss: 0.2598 - val_accuracy: 0.8760 - val_loss: 0.3422\n",
"Epoch 7/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9061 - loss: 0.2545 - val_accuracy: 0.8867 - val_loss: 0.3147\n",
"Epoch 8/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9094 - loss: 0.2420 - val_accuracy: 0.8893 - val_loss: 0.3141\n",
"Epoch 9/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9122 - loss: 0.2288 - val_accuracy: 0.8780 - val_loss: 0.3494\n",
"Epoch 10/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9165 - loss: 0.2218 - val_accuracy: 0.8899 - val_loss: 0.3222\n",
"Epoch 11/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9201 - loss: 0.2132 - val_accuracy: 0.8898 - val_loss: 0.3182\n",
"Epoch 12/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9235 - loss: 0.2020 - val_accuracy: 0.8924 - val_loss: 0.3073\n",
"Epoch 13/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9247 - loss: 0.1993 - val_accuracy: 0.8898 - val_loss: 0.3306\n",
"Epoch 14/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9291 - loss: 0.1879 - val_accuracy: 0.8894 - val_loss: 0.3475\n",
"Epoch 15/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9332 - loss: 0.1792 - val_accuracy: 0.8904 - val_loss: 0.3315\n",
"Epoch 16/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9332 - loss: 0.1797 - val_accuracy: 0.8928 - val_loss: 0.3508\n",
"Epoch 17/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9358 - loss: 0.1720 - val_accuracy: 0.8895 - val_loss: 0.3469\n",
"Epoch 18/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9390 - loss: 0.1633 - val_accuracy: 0.8905 - val_loss: 0.3497\n",
"Epoch 19/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9394 - loss: 0.1599 - val_accuracy: 0.8880 - val_loss: 0.3483\n",
"Epoch 20/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9436 - loss: 0.1496 - val_accuracy: 0.8885 - val_loss: 0.3815\n",
"Epoch 21/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9437 - loss: 0.1494 - val_accuracy: 0.8947 - val_loss: 0.3472\n",
"Epoch 22/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9457 - loss: 0.1457 - val_accuracy: 0.8883 - val_loss: 0.3934\n",
"Epoch 23/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9466 - loss: 0.1400 - val_accuracy: 0.8943 - val_loss: 0.3530\n",
"Epoch 24/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9487 - loss: 0.1352 - val_accuracy: 0.8901 - val_loss: 0.3810\n",
"Epoch 25/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9512 - loss: 0.1296 - val_accuracy: 0.8928 - val_loss: 0.3857\n",
"Epoch 26/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9515 - loss: 0.1283 - val_accuracy: 0.8956 - val_loss: 0.3769\n",
"Epoch 27/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9552 - loss: 0.1199 - val_accuracy: 0.8915 - val_loss: 0.3791\n",
"Epoch 28/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9553 - loss: 0.1173 - val_accuracy: 0.8946 - val_loss: 0.3851\n",
"Epoch 29/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9537 - loss: 0.1231 - val_accuracy: 0.8941 - val_loss: 0.4013\n",
"Epoch 30/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9576 - loss: 0.1152 - val_accuracy: 0.8942 - val_loss: 0.4023\n",
"Epoch 31/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9605 - loss: 0.1080 - val_accuracy: 0.8950 - val_loss: 0.4240\n",
"Epoch 32/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9596 - loss: 0.1069 - val_accuracy: 0.8918 - val_loss: 0.4392\n",
"Epoch 33/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9597 - loss: 0.1055 - val_accuracy: 0.8918 - val_loss: 0.4087\n",
"Epoch 34/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9605 - loss: 0.1034 - val_accuracy: 0.8744 - val_loss: 0.5429\n",
"Epoch 35/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9623 - loss: 0.1027 - val_accuracy: 0.8938 - val_loss: 0.4789\n",
"Epoch 36/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9618 - loss: 0.0986 - val_accuracy: 0.8942 - val_loss: 0.4359\n",
"Epoch 37/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9657 - loss: 0.0934 - val_accuracy: 0.8948 - val_loss: 0.4627\n",
"Epoch 38/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9629 - loss: 0.0959 - val_accuracy: 0.8932 - val_loss: 0.4700\n",
"Epoch 39/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9639 - loss: 0.0953 - val_accuracy: 0.8938 - val_loss: 0.4628\n",
"Epoch 40/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9677 - loss: 0.0863 - val_accuracy: 0.8943 - val_loss: 0.4916\n",
"Epoch 41/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9695 - loss: 0.0828 - val_accuracy: 0.8948 - val_loss: 0.4793\n",
"Epoch 42/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9702 - loss: 0.0814 - val_accuracy: 0.8954 - val_loss: 0.4605\n",
"Epoch 43/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9699 - loss: 0.0805 - val_accuracy: 0.8959 - val_loss: 0.4710\n",
"Epoch 44/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9714 - loss: 0.0769 - val_accuracy: 0.8903 - val_loss: 0.4923\n",
"Epoch 45/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9699 - loss: 0.0777 - val_accuracy: 0.8861 - val_loss: 0.5372\n",
"Epoch 46/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9692 - loss: 0.0820 - val_accuracy: 0.8960 - val_loss: 0.5028\n",
"Epoch 47/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9710 - loss: 0.0748 - val_accuracy: 0.8913 - val_loss: 0.5318\n",
"Epoch 48/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9717 - loss: 0.0758 - val_accuracy: 0.8902 - val_loss: 0.5180\n",
"Epoch 49/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9728 - loss: 0.0718 - val_accuracy: 0.8907 - val_loss: 0.5315\n",
"Epoch 50/50\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9744 - loss: 0.0671 - val_accuracy: 0.8940 - val_loss: 0.5327\n",
"Best epoch: 46\n"
]
}
],
"source": [
"# Build the model with the optimal hyperparameters and train it on the data for 50 epochs\n",
"model = tuner.hypermodel.build(best_hps)\n",
"history = model.fit(img_train, label_train, epochs=50, validation_split=0.2)\n",
"\n",
"val_acc_per_epoch = history.history['val_accuracy']\n",
"best_epoch = val_acc_per_epoch.index(max(val_acc_per_epoch)) + 1\n",
"print('Best epoch: %d' % (best_epoch,))"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "uOTSirSTI3Gp"
},
"source": [
"重新实例化超模型并使用上面的最佳周期数对其进行训练。"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"id": "NoiPUEHmMhCe"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 3ms/step - accuracy: 0.7809 - loss: 0.6163 - val_accuracy: 0.8484 - val_loss: 0.4384\n",
"Epoch 2/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8677 - loss: 0.3708 - val_accuracy: 0.8735 - val_loss: 0.3508\n",
"Epoch 3/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8762 - loss: 0.3344 - val_accuracy: 0.8788 - val_loss: 0.3403\n",
"Epoch 4/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8859 - loss: 0.3046 - val_accuracy: 0.8776 - val_loss: 0.3412\n",
"Epoch 5/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.8953 - loss: 0.2834 - val_accuracy: 0.8808 - val_loss: 0.3291\n",
"Epoch 6/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.8992 - loss: 0.2724 - val_accuracy: 0.8758 - val_loss: 0.3492\n",
"Epoch 7/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9066 - loss: 0.2501 - val_accuracy: 0.8848 - val_loss: 0.3247\n",
"Epoch 8/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9102 - loss: 0.2437 - val_accuracy: 0.8839 - val_loss: 0.3256\n",
"Epoch 9/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9117 - loss: 0.2329 - val_accuracy: 0.8885 - val_loss: 0.3255\n",
"Epoch 10/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9188 - loss: 0.2199 - val_accuracy: 0.8921 - val_loss: 0.3090\n",
"Epoch 11/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9197 - loss: 0.2120 - val_accuracy: 0.8873 - val_loss: 0.3218\n",
"Epoch 12/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9239 - loss: 0.2056 - val_accuracy: 0.8924 - val_loss: 0.3150\n",
"Epoch 13/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9273 - loss: 0.1980 - val_accuracy: 0.8956 - val_loss: 0.3158\n",
"Epoch 14/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9293 - loss: 0.1931 - val_accuracy: 0.8917 - val_loss: 0.3301\n",
"Epoch 15/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9323 - loss: 0.1786 - val_accuracy: 0.8976 - val_loss: 0.3212\n",
"Epoch 16/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9353 - loss: 0.1746 - val_accuracy: 0.8898 - val_loss: 0.3441\n",
"Epoch 17/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9347 - loss: 0.1709 - val_accuracy: 0.8947 - val_loss: 0.3288\n",
"Epoch 18/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9386 - loss: 0.1641 - val_accuracy: 0.8929 - val_loss: 0.3327\n",
"Epoch 19/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9424 - loss: 0.1533 - val_accuracy: 0.8949 - val_loss: 0.3435\n",
"Epoch 20/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9421 - loss: 0.1555 - val_accuracy: 0.8934 - val_loss: 0.3547\n",
"Epoch 21/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9448 - loss: 0.1469 - val_accuracy: 0.8953 - val_loss: 0.3412\n",
"Epoch 22/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9452 - loss: 0.1449 - val_accuracy: 0.8937 - val_loss: 0.3451\n",
"Epoch 23/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9477 - loss: 0.1407 - val_accuracy: 0.8945 - val_loss: 0.3516\n",
"Epoch 24/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9472 - loss: 0.1392 - val_accuracy: 0.8927 - val_loss: 0.3789\n",
"Epoch 25/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9506 - loss: 0.1318 - val_accuracy: 0.8898 - val_loss: 0.3972\n",
"Epoch 26/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9517 - loss: 0.1303 - val_accuracy: 0.8828 - val_loss: 0.4100\n",
"Epoch 27/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9533 - loss: 0.1262 - val_accuracy: 0.8924 - val_loss: 0.3892\n",
"Epoch 28/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9553 - loss: 0.1200 - val_accuracy: 0.8952 - val_loss: 0.3770\n",
"Epoch 29/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9571 - loss: 0.1156 - val_accuracy: 0.8959 - val_loss: 0.4118\n",
"Epoch 30/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9559 - loss: 0.1157 - val_accuracy: 0.8954 - val_loss: 0.3937\n",
"Epoch 31/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9579 - loss: 0.1112 - val_accuracy: 0.8932 - val_loss: 0.4362\n",
"Epoch 32/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9582 - loss: 0.1114 - val_accuracy: 0.8940 - val_loss: 0.4085\n",
"Epoch 33/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9594 - loss: 0.1066 - val_accuracy: 0.8959 - val_loss: 0.4060\n",
"Epoch 34/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 3ms/step - accuracy: 0.9607 - loss: 0.1041 - val_accuracy: 0.8966 - val_loss: 0.4171\n",
"Epoch 35/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9609 - loss: 0.1024 - val_accuracy: 0.8908 - val_loss: 0.4622\n",
"Epoch 36/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9613 - loss: 0.1018 - val_accuracy: 0.8754 - val_loss: 0.4987\n",
"Epoch 37/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9633 - loss: 0.0964 - val_accuracy: 0.8890 - val_loss: 0.4426\n",
"Epoch 38/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9644 - loss: 0.0931 - val_accuracy: 0.8934 - val_loss: 0.4596\n",
"Epoch 39/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9656 - loss: 0.0923 - val_accuracy: 0.8958 - val_loss: 0.4529\n",
"Epoch 40/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9687 - loss: 0.0821 - val_accuracy: 0.8915 - val_loss: 0.4697\n",
"Epoch 41/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.9685 - loss: 0.0824 - val_accuracy: 0.8928 - val_loss: 0.4751\n",
"Epoch 42/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9662 - loss: 0.0861 - val_accuracy: 0.8842 - val_loss: 0.5023\n",
"Epoch 43/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9690 - loss: 0.0812 - val_accuracy: 0.8896 - val_loss: 0.4934\n",
"Epoch 44/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9684 - loss: 0.0836 - val_accuracy: 0.8953 - val_loss: 0.4849\n",
"Epoch 45/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9703 - loss: 0.0785 - val_accuracy: 0.8932 - val_loss: 0.5106\n",
"Epoch 46/46\n",
"\u001b[1m1500/1500\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9717 - loss: 0.0782 - val_accuracy: 0.8932 - val_loss: 0.4991\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"hypermodel = tuner.hypermodel.build(best_hps)\n",
"\n",
"# Retrain the model\n",
"hypermodel.fit(img_train, label_train, epochs=best_epoch, validation_split=0.2)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MqU5ZVAaag2v"
},
"source": [
"要完成本教程，请在测试数据上评估超模型。"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"id": "9E0BTp9Ealjb"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[1m313/313\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.8867 - loss: 0.5731\n",
"[test loss, test accuracy]: [0.5481381416320801, 0.8877999782562256]\n"
]
}
],
"source": [
"eval_result = hypermodel.evaluate(img_test, label_test)\n",
"print(\"[test loss, test accuracy]:\", eval_result)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "EQRpPHZsz-eC"
},
"source": [
"`my_dir/intro_to_kt` 目录中包含了在超参数搜索期间每次试验（模型配置）运行的详细日志和检查点。如果重新运行超参数搜索，Keras Tuner 将使用这些日志中记录的现有状态来继续搜索。要停用此行为，请在实例化调节器时传递一个附加的 `overwrite = True` 参数。"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "sKwLOzKpFGAj"
},
"source": [
"## 总结\n",
"\n",
"在本教程中，您学习了如何使用 Keras Tuner 调节模型的超参数。要详细了解 Keras Tuner，请查看以下其他资源：\n",
"\n",
"- [TensorFlow 博客上的 Keras Tuner](https://blog.tensorflow.org/2020/01/hyperparameter-tuning-with-keras-tuner.html)\n",
"- [Keras Tuner 网站](https://keras-team.github.io/keras-tuner/)\n",
"\n",
"另请查看 TensorBoard 中的 [HParams Dashboard](https://tensorflow.google.cn/tensorboard/hyperparameter_tuning_with_hparams)，以交互方式调节模型超参数。"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"collapsed_sections": [
"Tce3stUlHN0L"
],
"name": "keras_tuner.ipynb",
"toc_visible": true
},
"kernelspec": {
"display_name": "xxx",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.2"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
![](\"https://tensorflow.google.cn/images/tf_logo_32px.png\"){kind=logo}
在 TensorFlow.org 上查看 ![](\"https://tensorflow.google.cn/images/colab_logo_32px.png\"){kind=logo}
在 Google Colab 中运行![](\"https://tensorflow.google.cn/images/GitHub-Mark-32px.png\")
在 GitHub 上查看源代码![](\"https://tensorflow.google.cn/images/download_logo_32px.png\"){kind=logo}
下载笔记本