{
  "cells": [
    {
      "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"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "cellView": "form",
        "id": "BZSlp3DAjdYf"
      },
      "source": [
        "````{admonition} Copyright 2019 The TensorFlow Authors.\n",
        "```\n",
        "#@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.\n",
        "```\n",
        "````"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3wF5wszaj97Y"
      },
      "source": [
        "# 初学者的 TensorFlow 2.0 教程"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DUNzJc4jTj6G"
      },
      "source": [
        "<table class=\"tfo-notebook-buttons\" align=\"left\">\n",
        "  <td>     <a target=\"_blank\" href=\"https://tensorflow.google.cn/tutorials/quickstart/beginner\"><img src=\"https://tensorflow.google.cn/images/tf_logo_32px.png\">在 TensorFlow.org 观看</a>   </td>\n",
        "  <td><a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/docs-l10n/blob/master/site/zh-cn/tutorials/quickstart/beginner.ipynb\"><img src=\"https://tensorflow.google.cn/images/colab_logo_32px.png\">在 Google Colab 中运行 </a></td>\n",
        "  <td>     <a target=\"_blank\" href=\"https://github.com/tensorflow/docs-l10n/blob/master/site/zh-cn/tutorials/quickstart/beginner.ipynb\"><img src=\"https://tensorflow.google.cn/images/GitHub-Mark-32px.png\">在 GitHub 查看源代码</a>   </td>\n",
        "  <td>     <a href=\"https://storage.googleapis.com/tensorflow_docs/docs-l10n/site/zh-cn/tutorials/quickstart/beginner.ipynb\"><img src=\"https://tensorflow.google.cn/images/download_logo_32px.png\">下载笔记本</a>   </td>\n",
        "</table>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "04QgGZc9bF5D"
      },
      "source": [
        "```{topic} 导航\n",
        "\n",
        "此简短介绍使用 [Keras](https://tensorflow.google.cn/guide/keras/overview) 进行以下操作：\n",
        "\n",
        "1. 加载预构建的数据集。\n",
        "2. 构建对图像进行分类的神经网络机器学习模型。\n",
        "3. 训练此神经网络。\n",
        "4. 评估模型的准确率。\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nnrWf3PCEzXL"
      },
      "source": [
        "## 设置 TensorFlow\n",
        "\n",
        "首先将 TensorFlow 导入到您的程序："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "0trJmd6DjqBZ"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7NAbSZiaoJ4z"
      },
      "source": [
        "如果您在自己的开发环境而不是 [Colab](https://colab.research.google.com/github/tensorflow/docs/blob/master/site/en/tutorials/quickstart/beginner.ipynb) 中操作，请参阅设置 TensorFlow 以进行开发的[安装指南](https://tensorflow.google.cn/install)。\n",
        "\n",
        "注：如果您使用自己的开发环境，请确保您已升级到最新的 `pip` 以安装 TensorFlow 2 软件包。有关详情，请参阅[安装指南](https://tensorflow.google.cn/install)。\n",
        "\n",
        "## 加载数据集\n",
        "\n",
        "加载并准备 [MNIST 数据集](http://yann.lecun.com/exdb/mnist/)。图像的像素值范围从 0 到 255。通过将值除以 `255.0` 来将这些值缩放到 0 到 1 的范围。这还会将样本数据从整数转换为浮点数："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "7FP5258xjs-v"
      },
      "outputs": [],
      "source": [
        "mnist = tf.keras.datasets.mnist\n",
        "\n",
        "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
        "x_train, x_test = x_train / 255.0, x_test / 255.0"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BPZ68wASog_I"
      },
      "source": [
        "## 构建机器学习模型\n",
        "\n",
        "构建 `tf.keras.Sequential` 模型："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "h3IKyzTCDNGo"
      },
      "outputs": [],
      "source": [
        "model = tf.keras.models.Sequential([\n",
        "  tf.keras.layers.Flatten(),\n",
        "  tf.keras.layers.Dense(128, activation='relu'),\n",
        "  tf.keras.layers.Dropout(0.2),\n",
        "  tf.keras.layers.Dense(10)\n",
        "])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "l2hiez2eIUz8"
      },
      "source": [
        "[`Sequential`](https://tensorflow.google.cn/guide/keras/sequential_model) 对于堆叠层很有用，其中每层都有输入[张量](https://tensorflow.google.cn/guide/tensor)和输出张量。层是具有已知数学结构的函数，可以重复使用并具有可训练的变量。大多数 TensorFlow 模型都由层组成。此模型使用 [`Flatten`](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/Flatten)、[`Dense`](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/Dense) 和 [`Dropout`](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/Dropout) 层。\n",
        "\n",
        "对于每个样本，模型都会返回包含 [logits](https://developers.google.com/machine-learning/glossary#logits) 或 [log-odds](https://developers.google.com/machine-learning/glossary#log-odds) 分数的向量，每个类一个。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "id": "OeOrNdnkEEcR"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "array([[-0.29356095, -0.18187018, -0.09060116,  0.02143997,  0.24739322,\n",
              "         0.02414675, -0.34286305,  0.02412909,  0.39885187,  0.39236996]],\n",
              "      dtype=float32)"
            ]
          },
          "execution_count": 5,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "predictions = model(x_train[:1]).numpy()\n",
        "predictions"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tgjhDQGcIniO"
      },
      "source": [
        "{func}`tensorflow.nn.softmax` 函数将这些 logits 转换为每个类的*概率*： "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "id": "zWSRnQ0WI5eq"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "array([[0.07084628, 0.07921798, 0.08678835, 0.09707788, 0.12168877,\n",
              "        0.09734099, 0.06743812, 0.09733927, 0.14158855, 0.14067376]],\n",
              "      dtype=float32)"
            ]
          },
          "execution_count": 6,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "tf.nn.softmax(predictions).numpy()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "he5u_okAYS4a"
      },
      "source": [
        "注：可以将 {func}`tensorflow.nn.softmax` 烘焙到网络最后一层的激活函数中。虽然这可以使模型输出更易解释，但不建议使用这种方式，因为在使用 `softmax` 输出时不可能为所有模型提供精确且数值稳定的损失计算。 "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hQyugpgRIyrA"
      },
      "source": [
        "使用 {class}`~tensorflow.keras.losses.SparseCategoricalCrossentropy` 定义训练的损失函数："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "id": "RSkzdv8MD0tT"
      },
      "outputs": [],
      "source": [
        "loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SfR4MsSDU880"
      },
      "source": [
        "损失函数采用真实值向量和逻辑向量，并返回每个样本的标量损失。此损失等于真实类的负对数概率：如果模型确定类正确，则损失为零。\n",
        "\n",
        "这个未经训练的模型给出的概率接近随机（每个类为 1/10），因此初始损失应该接近 `-tf.math.log(1/10) ~= 2.3`。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "NJWqEVrrJ7ZB"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "2.329535"
            ]
          },
          "execution_count": 8,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "loss_fn(y_train[:1], predictions).numpy()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ada44eb947d4"
      },
      "source": [
        "在开始训练之前，使用 Keras {meth}`~tensorflow.models.model.Model.compile` 配置和编译模型。将 [`optimizer`](https://tensorflow.google.cn/api_docs/python/tf/keras/optimizers) 类设置为 `adam`，将 `loss` 设置为您之前定义的 `loss_fn` 函数，并通过将 `metrics` 参数设置为 `accuracy` 来指定要为模型评估的指标。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "id": "9foNKHzTD2Vo"
      },
      "outputs": [],
      "source": [
        "model.compile(optimizer='adam',\n",
        "              loss=loss_fn,\n",
        "              metrics=['accuracy'])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ix4mEL65on-w"
      },
      "source": [
        "## 训练并评估模型\n",
        "\n",
        "使用 `Model.fit` 方法调整您的模型参数并最小化损失： "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "y7suUbJXVLqP"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/5\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "WARNING: All log messages before absl::InitializeLog() is called are written to STDERR\n",
            "I0000 00:00:1729733723.232793 3171586 service.cc:146] XLA service 0x7f3d74006ef0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:\n",
            "I0000 00:00:1729733723.232815 3171586 service.cc:154]   StreamExecutor device (0): NVIDIA GeForce RTX 3090, Compute Capability 8.6\n",
            "I0000 00:00:1729733723.232833 3171586 service.cc:154]   StreamExecutor device (1): NVIDIA GeForce RTX 2080 Ti, Compute Capability 7.5\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\u001b[1m  78/1875\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 2ms/step - accuracy: 0.5239 - loss: 1.5746"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "I0000 00:00:1729733725.890061 3171586 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 2ms/step - accuracy: 0.8585 - loss: 0.4892\n",
            "Epoch 2/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9557 - loss: 0.1527\n",
            "Epoch 3/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9675 - loss: 0.1089\n",
            "Epoch 4/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9736 - loss: 0.0847\n",
            "Epoch 5/5\n",
            "\u001b[1m1875/1875\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 2ms/step - accuracy: 0.9762 - loss: 0.0727\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "<keras.src.callbacks.history.History at 0x7f40f4d03710>"
            ]
          },
          "execution_count": 10,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "model.fit(x_train, y_train, epochs=5)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4mDAAPFqVVgn"
      },
      "source": [
        "`Model.evaluate` 方法会检查模型的性能（通常是在[验证集](https://developers.google.com/machine-learning/glossary#validation-set)或[测试集](https://developers.google.com/machine-learning/glossary#test-set)上）。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "id": "F7dTAzgHDUh7"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "313/313 - 2s - 8ms/step - accuracy: 0.9769 - loss: 0.0729\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "[0.07293855398893356, 0.9768999814987183]"
            ]
          },
          "execution_count": 11,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "model.evaluate(x_test,  y_test, verbose=2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "T4JfEh7kvx6m"
      },
      "source": [
        "现在，这个照片分类器的准确度已经达到 98%。想要了解更多，请阅读 [TensorFlow 教程](https://tensorflow.google.cn/tutorials/)。"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Aj8NrlzlJqDG"
      },
      "source": [
        "如果您想让模型返回概率，可以封装经过训练的模型，并将 softmax 附加到该模型："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {
        "id": "rYb6DrEH0GMv"
      },
      "outputs": [],
      "source": [
        "probability_model = tf.keras.Sequential([\n",
        "  model,\n",
        "  tf.keras.layers.Softmax()\n",
        "])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "cnqOZtUp1YR_"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<tf.Tensor: shape=(5, 10), dtype=float32, numpy=\n",
              "array([[5.70330599e-07, 1.30929301e-09, 3.31053479e-06, 6.53220995e-05,\n",
              "        5.35139099e-13, 1.03895630e-07, 7.61280822e-14, 9.99930143e-01,\n",
              "        1.48227429e-07, 3.17864817e-07],\n",
              "       [2.60567834e-08, 1.02092403e-04, 9.99895215e-01, 2.47421417e-06,\n",
              "        2.45008007e-14, 1.18589675e-07, 7.92316790e-09, 1.31756047e-11,\n",
              "        1.22220749e-08, 6.07167094e-13],\n",
              "       [7.65437704e-08, 9.98419642e-01, 1.11594715e-03, 8.62039724e-06,\n",
              "        1.14041950e-05, 3.91538015e-06, 4.68364533e-07, 3.88111861e-04,\n",
              "        4.93279695e-05, 2.40418944e-06],\n",
              "       [9.99776661e-01, 5.56393265e-10, 1.60305080e-05, 3.42057410e-06,\n",
              "        1.33386193e-05, 5.57819885e-06, 1.63603676e-04, 1.89078055e-05,\n",
              "        5.46506328e-07, 1.99484361e-06],\n",
              "       [3.93524680e-08, 6.11409900e-09, 1.89196712e-07, 8.14441758e-09,\n",
              "        9.99578297e-01, 1.13444706e-07, 4.50963626e-06, 2.70816599e-06,\n",
              "        2.53426805e-07, 4.13792412e-04]], dtype=float32)>"
            ]
          },
          "execution_count": 13,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "probability_model(x_test[:5])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-47O6_GLdRuT"
      },
      "source": [
        "## 结论\n",
        "\n",
        "恭喜！您已经利用 [Keras](https://tensorflow.google.cn/guide/keras/overview) API 借助预构建数据集训练了一个机器学习模型。\n",
        "\n",
        "有关使用 Keras 的更多示例，请查阅[教程](https://tensorflow.google.cn/tutorials/keras/)。要详细了解如何使用 Keras 构建模型，请阅读[指南](https://tensorflow.google.cn/guide/keras)。如果您想详细了解如何加载和准备数据，请参阅有关[图像数据加载](https://tensorflow.google.cn/tutorials/load_data/images)或 [CSV 数据加载](https://tensorflow.google.cn/tutorials/load_data/csv)的教程。\n"
      ]
    }
  ],
  "metadata": {
    "colab": {
      "name": "beginner.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
}