{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": { "id": "SB93Ge748VQs" }, "outputs": [], "source": [ "##### Copyright 2019 The TensorFlow Authors." ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "cellView": "form", "id": "0sK8X2O9bTlz", "vscode": { "languageId": "python" } }, "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": "markdown", "metadata": { "id": "HEYuO5NFwDK9" }, "source": [ "# 检查 TensorFlow 图\n", "\n", "\n", " \n", " \n", " \n", " \n", "
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
" ] }, { "cell_type": "markdown", "metadata": { "id": "56V5oun18ZdZ" }, "source": [ "## 概述\n", "\n", "TensorBoard 的 **Graphs 信息中心**是检查 TensorFlow 模型的强大工具。您可以快速查看模型结构的概念计算图，并确保其符合您的预期设计。您还可以查看运算级计算图以了解 TensorFlow 如何理解您的程序。检查运算级计算图可以使您深入了解如何更改模型。例如，如果训练进度比预期的慢，则可以重新设计模型。" ] }, { "cell_type": "markdown", "metadata": { "id": "TOSJ-4nteBYG" }, "source": [ "本教程简要概述了如何生成计算图诊断数据并将其在 TensorBoard 的 Graphs 信息中心中呈现。您将为 Fashion-MNIST 数据集定义和训练一个简单的 Keras 序贯模型，并学习如何记录和检查模型计算图。您还将使用跟踪 API 为使用新 `tf.function` 注解创建的函数生成计算图数据。" ] }, { "cell_type": "markdown", "metadata": { "id": "zNI1-dflrAo0" }, "source": [ "## 设置" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "id": "6B95Hb6YVgPZ", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "# Load the TensorBoard notebook extension.\n", "%load_ext tensorboard" ] }, { "cell_type": "code", "execution_count": 3, "metadata": { "id": "_wqSAZExy6xV", "vscode": { "languageId": "python" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "TensorFlow version: 2.2.0\n" ] } ], "source": [ "from datetime import datetime\n", "from packaging import version\n", "\n", "import tensorflow as tf\n", "from tensorflow import keras\n", "\n", "print(\"TensorFlow version: \", tf.__version__)\n", "assert version.parse(tf.__version__).release[0] >= 2, \\\n", " \"This notebook requires TensorFlow 2.0 or above.\"" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "id": "qRZlYIEcJ56Z", "vscode": { "languageId": "python" } }, "outputs": [ { "data": { "text/plain": [ "'2.2.1'" ] }, "execution_count": 4, "metadata": {}, "output_type": "execute_result" } ], "source": [ "import tensorboard\n", "tensorboard.__version__" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "id": "Ao7fJW1Pyiza", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "# Clear any logs from previous runs\n", "!rm -rf ./logs/ " ] }, { "cell_type": "markdown", "metadata": { "id": "e25E37vd1xEW" }, "source": [ "## 定义一个 Keras 模型\n", "\n", "在此示例中，分类器是一个简单的四层顺序模型。" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "id": "skqORzvE3Egy", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "# Define the model.\n", "model = keras.models.Sequential([\n", " keras.layers.Flatten(input_shape=(28, 28)),\n", " keras.layers.Dense(32, activation='relu'),\n", " keras.layers.Dropout(0.2),\n", " keras.layers.Dense(10, activation='softmax')\n", "])\n", "\n", "model.compile(\n", " optimizer='adam',\n", " loss='sparse_categorical_crossentropy',\n", " metrics=['accuracy'])" ] }, { "cell_type": "markdown", "metadata": { "id": "qbjuoz9E3VC_" }, "source": [ "下载并准备训练数据" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "id": "6TDmc41z3g38", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "(train_images, train_labels), _ = keras.datasets.fashion_mnist.load_data()\n", "train_images = train_images / 255.0" ] }, { "cell_type": "markdown", "metadata": { "id": "8DV0xibO3bRC" }, "source": [ "## 训练模型并记录数据\n", "\n", "训练之前，请定义 [Keras TensorBoard callback](https://tensorflow.google.cn/api_docs/python/tf/keras/callbacks/TensorBoard), 并指定日志目录。通过将此回调传递给 Model.fit(), 可以确保在 TensorBoard 中记录图形数据以进行可视化。" ] }, { "cell_type": "code", "execution_count": 8, "metadata": { "id": "TU_L_u9SqQdH", "vscode": { "languageId": "python" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/5\n", "938/938 [==============================] - 2s 2ms/step - loss: 0.6955 - accuracy: 0.7618\n", "Epoch 2/5\n", "938/938 [==============================] - 2s 2ms/step - loss: 0.4877 - accuracy: 0.8296\n", "Epoch 3/5\n", "938/938 [==============================] - 2s 2ms/step - loss: 0.4458 - accuracy: 0.8414\n", "Epoch 4/5\n", "938/938 [==============================] - 2s 2ms/step - loss: 0.4246 - accuracy: 0.8476\n", "Epoch 5/5\n", "938/938 [==============================] - 2s 2ms/step - loss: 0.4117 - accuracy: 0.8508\n" ] }, { "data": { "text/plain": [ "" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# Define the Keras TensorBoard callback.\n", "logdir=\"logs/fit/\" + datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n", "tensorboard_callback = keras.callbacks.TensorBoard(log_dir=logdir)\n", "\n", "# Train the model.\n", "model.fit(\n", " train_images,\n", " train_labels, \n", " batch_size=64,\n", " epochs=5, \n", " callbacks=[tensorboard_callback])" ] }, { "cell_type": "markdown", "metadata": { "id": "IRX5OIsi4TTV" }, "source": [ "## op-level graph\n", "\n", "启动 TensorBoard，然后等待几秒钟以加载 UI。通过点击顶部的 “graph” 来选择图形仪表板。 " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "PFgFjlPEqXb9", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "%tensorboard --logdir logs" ] }, { "cell_type": "markdown", "metadata": { "id": "9PFgFjlPEqXb" }, "source": [ "您还可以选择使用 TensorBoard.dev 来创建托管的可共享实验。 " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "b9PFgFjlPEqX", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "!tensorboard dev upload \\\n", " --logdir logs \\\n", " --name \"Sample op-level graph\" \\\n", " --one_shot" ] }, { "cell_type": "markdown", "metadata": { "id": "EGlOqRp54ufD" }, "source": [ "默认情况下，TensorBoard 显示 **op-level图**。（在左侧，您可以看到已选择 “Default” 标签。）请注意，图是倒置的。 数据从下到上流动，因此与代码相比是上下颠倒的。 但是，您可以看到该图与 Keras 模型定义紧密匹配，并具有其他计算节点的额外边缘。\n", "\n", "图通常很大，因此您可以操纵图的可视化效果：\n", "\n", "- 滚动到 **zoom** 来放大和缩小\n", "- 拖到 **pan** 平移\n", "- 双击切换 **node expansion** 来进行节点扩展（一个节点可以是其他节点的容器）\n", "\n", "您还可以通过单击节点来查看元数据。这使您可以查看输入，输出，形状和其他详细信息。\n" ] }, { "cell_type": "markdown", "metadata": { "id": "F-2yw5qd7OpK" }, "source": [ "" ] }, { "cell_type": "markdown", "metadata": { "id": "jDRynpVw53SJ" }, "source": [ "" ] }, { "cell_type": "markdown", "metadata": { "id": "Oj9FSPdz6SO2" }, "source": [ "## 概念图\n", "\n", "除了执行图，TensorBoard 还显示一个“概念图”。 这只是 Keras 模型的视图。 如果您要重新使用保存的模型并且想要检查或验证其结构，这可能会很有用。\n", "\n", "要查看概念计算图，请选择“keras”标签。在此示例中，您将看到一个收起的 **Sequential** 节点。双击该节点可以查看模型的结构：" ] }, { "cell_type": "markdown", "metadata": { "id": "Qw9rbEcE6eZB" }, "source": [ "
\n", "" ] }, { "cell_type": "markdown", "metadata": { "id": "zVuaKBifu-qF" }, "source": [ "## tf.functions 的图\n", "\n", "目前为止的示例描述了 Keras 模型的计算图，其中这些计算图是通过定义 Keras 层并调用 Model.fit() 创建的。\n", "\n", "您可能会遇到需要使用 `tf.function` 注释来[[autograph]](https://tensorflow.google.cn/guide/function)的情况，即将 Python 计算函数转换为高性能 TensorFlow 图。对于这些情况，您可以使用 TensorBoard 中的 **TensorFlow Summary Trace API** 记录签名函数以进行可视化。" ] }, { "cell_type": "markdown", "metadata": { "id": "JIuhJnQ8w-dT" }, "source": [ "要使用 Summary Trace API ，请执行以下操作:\n", "\n", "- 使用 `tf.function` 定义和注释功能\n", "- 在函数调用站点之前立即使用 `tf.summary.trace_on()`\n", "- 通过传递 `profiler=True` 将性能剖析信息（内存、CPU 时间）添加到计算图中\n", "- 使用摘要文件编写器，调用 `tf.summary.trace_export()` 保存日志数据\n", "\n", "然后，您可以使用 TensorBoard 查看函数的行为。\n" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "id": "woI67Stgv_uY", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "# The function to be traced.\n", "@tf.function\n", "def my_func(x, y):\n", " # A simple hand-rolled layer.\n", " return tf.nn.relu(tf.matmul(x, y))\n", "\n", "# Set up logging.\n", "stamp = datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n", "logdir = 'logs/func/%s' % stamp\n", "writer = tf.summary.create_file_writer(logdir)\n", "\n", "# Sample data for your function.\n", "x = tf.random.uniform((3, 3))\n", "y = tf.random.uniform((3, 3))\n", "\n", "# Bracket the function call with\n", "# tf.summary.trace_on() and tf.summary.trace_export().\n", "tf.summary.trace_on(graph=True, profiler=True)\n", "# Call only one tf.function when tracing.\n", "z = my_func(x, y)\n", "with writer.as_default():\n", " tf.summary.trace_export(\n", " name=\"my_func_trace\",\n", " step=0,\n", " profiler_outdir=logdir)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "zCArnWzP0VuZ", "vscode": { "languageId": "python" } }, "outputs": [], "source": [ "%tensorboard --logdir logs/func" ] }, { "cell_type": "markdown", "metadata": { "id": "WDl1PBFQ64xi" }, "source": [ "" ] }, { "cell_type": "markdown", "metadata": { "id": "1pLRaf3q6Nku" }, "source": [ "现在，您可以看到 TensorBoard 所理解的函数结构。 单击 “Profile” 按钮以查看 CPU 和内存统计信息。" ] } ], "metadata": { "colab": { "collapsed_sections": [ "SB93Ge748VQs" ], "name": "graphs.ipynb", "toc_visible": true }, "kernelspec": { "display_name": "Python 3", "name": "python3" } }, "nbformat": 4, "nbformat_minor": 0 }