{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "wJcYs_ERTnnI"
      },
      "outputs": [],
      "source": [
        "##### Copyright 2021 The TensorFlow Authors."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "HMUDt0CiUJk9",
        "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": "77z2OchJTk0l"
      },
      "source": [
        "# 从 Estimator 迁移到 Keras API\n",
        "\n",
        "<table class=\"tfo-notebook-buttons\" align=\"left\">\n",
        "  <td>     <a target=\"_blank\" href=\"https://tensorflow.google.cn/guide/migrate/migrating_estimator\"><img src=\"https://tensorflow.google.cn/images/tf_logo_32px.png\">在 TensorFlow.org 上查看</a>   </td>\n",
        "  <td>     <a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/docs-l10n/blob/master/site/zh-cn/guide/migrate/migrating_estimator.ipynb\"><img src=\"https://tensorflow.google.cn/images/colab_logo_32px.png\">在 Google Colab 运行</a>\n",
        "</td>\n",
        "  <td>     <a target=\"_blank\" href=\"https://github.com/tensorflow/docs-l10n/blob/master/site/zh-cn/guide/migrate/migrating_estimator.ipynb\"><img src=\"https://tensorflow.google.cn/images/GitHub-Mark-32px.png\">在 Github 上查看源代码</a>\n",
        "</td>\n",
        "  <td>     <a href=\"https://storage.googleapis.com/tensorflow_docs/docs-l10n/site/zh-cn/guide/migrate/migrating_estimator.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": "meUTrR4I6m1C"
      },
      "source": [
        "本指南演示了如何从 TensorFlow 1 的 `tf.estimator.Estimator` API 迁移到 TensorFlow 2 的 `tf.keras` API。首先，您将使用 `tf.estimator.Estimator` 设置并运行一个用于训练和评估的基本模型。然后，您将使用 `tf.keras` API 在 TensorFlow 2 中执行对应步骤。此外，您还将了解如何通过子类化 `tf.keras.Model` 和使用 `tf.GradientTape` 来自定义训练步骤。\n",
        "\n",
        "- 在 TensorFlow 1 中，可以使用高级 `tf.estimator.Estimator` API 训练和评估模型，以及执行推断和保存模型（用于提供）。\n",
        "- 在 TensorFlow 2 中，使用 Keras API 执行上述任务，例如[模型构建](https://tensorflow.google.cn/guide/keras/custom_layers_and_models)、梯度应用、[训练](https://tensorflow.google.cn/guide/keras/customizing_what_happens_in_fit)、评估和预测。\n",
        "\n",
        "（要将模型/检查点保存工作流迁移到 TensorFlow 2，请查看 [SavedModel](saved_model.ipynb) 和[检查点](checkpoint_saved.ipynb)迁移指南。）"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YdZSoIXEbhg-"
      },
      "source": [
        "## 安装\n",
        "\n",
        "从导入和一个简单的数据集开始："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "iE0vSfMXumKI",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "import tensorflow.compat.v1 as tf1"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "m7rnGxsXtDkV",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "features = [[1., 1.5], [2., 2.5], [3., 3.5]]\n",
        "labels = [[0.3], [0.5], [0.7]]\n",
        "eval_features = [[4., 4.5], [5., 5.5], [6., 6.5]]\n",
        "eval_labels = [[0.8], [0.9], [1.]]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4uXff1BEssdE"
      },
      "source": [
        "## TensorFlow 1：使用 tf.estimator.Estimator 进行训练和评估\n",
        "\n",
        "此示例展示了如何在 TensorFlow 1 中使用 `tf.estimator.Estimator` 执行训练和评估。\n",
        "\n",
        "首先定义几个函数：训练数据的输入函数，评估数据的评估输入函数，以及告知 `Estimator` 如何使用特征和标签定义训练运算的模型函数："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lqe9obf7suIj",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "def _input_fn():\n",
        "  return tf1.data.Dataset.from_tensor_slices((features, labels)).batch(1)\n",
        "\n",
        "def _eval_input_fn():\n",
        "  return tf1.data.Dataset.from_tensor_slices(\n",
        "      (eval_features, eval_labels)).batch(1)\n",
        "\n",
        "def _model_fn(features, labels, mode):\n",
        "  logits = tf1.layers.Dense(1)(features)\n",
        "  loss = tf1.losses.mean_squared_error(labels=labels, predictions=logits)\n",
        "  optimizer = tf1.train.AdagradOptimizer(0.05)\n",
        "  train_op = optimizer.minimize(loss, global_step=tf1.train.get_global_step())\n",
        "  return tf1.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "44bf417bf9c0"
      },
      "source": [
        "实例化您的 `Estimator`，并训练模型："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "922720812527",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "estimator = tf1.estimator.Estimator(model_fn=_model_fn)\n",
        "estimator.train(_input_fn)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "17c9933c2d89"
      },
      "source": [
        "使用评估集评估程序："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "HsOpjW5plH9Q",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "estimator.evaluate(_eval_input_fn)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KEmzBjfnsxwT"
      },
      "source": [
        "## TensorFlow 2：使用内置 Keras 方法进行训练和评估\n",
        "\n",
        "此示例演示了如何在 TensorFlow 2 中使用 `Model.fit` 和 `Model.evaluate` 执行训练和评估。（可以在[使用内置方法进行训练和评估](https://tensorflow.google.cn/guide/keras/train_and_evaluate)指南中了解详情。）\n",
        "\n",
        "- 首先使用 `tf.data.Dataset` API 准备数据集流水线。\n",
        "- 使用一个线性 (`tf.keras.layers.Dense`) 层定义一个简单的 Keras [序贯](https://tensorflow.google.cn/guide/keras/sequential_model)模型。\n",
        "- 实例化一个 Adagrad 优化器 (`tf.keras.optimizers.Adagrad`)。\n",
        "- 通过将 `optimizer` 变量和均方差（`\"mse\"`）损失传递给 `Model.compile` 来配置模型进行训练。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "atVciNgPs0fw",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "dataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(1)\n",
        "eval_dataset = tf.data.Dataset.from_tensor_slices(\n",
        "      (eval_features, eval_labels)).batch(1)\n",
        "\n",
        "model = tf.keras.models.Sequential([tf.keras.layers.Dense(1)])\n",
        "optimizer = tf.keras.optimizers.Adagrad(learning_rate=0.05)\n",
        "\n",
        "model.compile(optimizer=optimizer, loss=\"mse\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ed17a6291959"
      },
      "source": [
        "这样，您就可以通过调用 `Model.fit` 来训练模型了："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "a0b732534501",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.fit(dataset)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "74767288a2ea"
      },
      "source": [
        "最后，使用 `Model.evaluate` 评估模型："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Kip65sYBlKiu",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.evaluate(eval_dataset, return_dict=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BuVYN0CHs5sD"
      },
      "source": [
        "## TensorFlow 2：使用自定义训练步骤和内置 Keras 方法进行训练和评估"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gHx_RUL8xcJ3"
      },
      "source": [
        "在 TensorFlow 2 中，还可以使用 `tf.GradientTape` 编写自己的自定义训练步骤函数来执行前向和后向传递，同时仍然利用内置的训练支持，例如 `tf.keras.callbacks.Callback` 和 `tf.distribute.Strategy`。（在[自定义 Model.fit 的功能](https://tensorflow.google.cn/guide/keras/customizing_what_happens_in_fit)和[从头开始编写自定义训练循环](https://tensorflow.google.cn/guide/keras/writing_a_training_loop_from_scratch)中了解详情。）\n",
        "\n",
        "在此示例中，首先通过子类化重写 `Model.train_step` 的 `tf.keras.Sequential` 来创建自定义 `tf.keras.Model`。（详细了解如何[子类化 tf.keras.Model](https://tensorflow.google.cn/guide/keras/custom_layers_and_models)）。在该类中，定义一个自定义 `train_step` 函数，此函数在一个训练步骤中为每批次数据执行前向传递和后向传递。\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "rSz_y0zOs8h2",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "class CustomModel(tf.keras.Sequential):\n",
        "  \"\"\"A custom sequential model that overrides `Model.train_step`.\"\"\"\n",
        "\n",
        "  def train_step(self, data):\n",
        "    batch_data, labels = data\n",
        "\n",
        "    with tf.GradientTape() as tape:\n",
        "      predictions = self(batch_data, training=True)\n",
        "      # Compute the loss value (the loss function is configured\n",
        "      # in `Model.compile`).\n",
        "      loss = self.compiled_loss(labels, predictions)\n",
        "\n",
        "    # Compute the gradients of the parameters with respect to the loss.\n",
        "    gradients = tape.gradient(loss, self.trainable_variables)\n",
        "    # Perform gradient descent by updating the weights/parameters.\n",
        "    self.optimizer.apply_gradients(zip(gradients, self.trainable_variables))\n",
        "    # Update the metrics (includes the metric that tracks the loss).\n",
        "    self.compiled_metrics.update_state(labels, predictions)\n",
        "    # Return a dict mapping metric names to the current values.\n",
        "    return {m.name: m.result() for m in self.metrics}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ee7c4f94d69b"
      },
      "source": [
        "接下来，和之前一样：\n",
        "\n",
        "- 使用 `tf.data.Dataset` 准备数据集流水线。\n",
        "- 使用一个 `tf.keras.layers.Dense` 层定义一个简单的模型。\n",
        "- 实例化 Adagrad (`tf.keras.optimizers.Adagrad`)\n",
        "- 使用 `Model.compile` 配置用于训练的模型，同时使用均方差（`\"mse\"`）作为损失函数。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "01fcc2b1292c",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "dataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(1)\n",
        "eval_dataset = tf.data.Dataset.from_tensor_slices(\n",
        "      (eval_features, eval_labels)).batch(1)\n",
        "\n",
        "model = CustomModel([tf.keras.layers.Dense(1)])\n",
        "optimizer = tf.keras.optimizers.Adagrad(learning_rate=0.05)\n",
        "\n",
        "model.compile(optimizer=optimizer, loss=\"mse\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "844543802ff5"
      },
      "source": [
        "调用 `Model.fit` 以训练模型："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "211be3620765",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.fit(dataset)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c93b9d6fc9d7"
      },
      "source": [
        "最后，使用 `Model.evaluate` 评估程序："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nYO2wI1SlNCG",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.evaluate(eval_dataset, return_dict=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "e9b5c9a4747b"
      },
      "source": [
        "## 后续步骤\n",
        "\n",
        "您可能会发现有用的其他 Keras 资源：\n",
        "\n",
        "- 指南：[使用内置方法进行训练和评估](https://tensorflow.google.cn/guide/keras/train_and_evaluate)\n",
        "- 指南：[自定义 Model.fit 的功能](https://tensorflow.google.cn/guide/keras/customizing_what_happens_in_fit)\n",
        "- 指南：[从头开始编写训练循环](https://tensorflow.google.cn/guide/keras/writing_a_training_loop_from_scratch)\n",
        "- 指南：[通过子类化创建新的 Keras 层和模型](https://tensorflow.google.cn/guide/keras/custom_layers_and_models)\n",
        "\n",
        "以下指南有助于从 `tf.estimator` API 迁移分布策略工作流：\n",
        "\n",
        "- [从 TPUEstimator 迁移到 TPUStrategy](tpu_estimator.ipynb)\n",
        "- [迁移单工作进程多 GPU 训练](mirrored_strategy.ipynb)\n",
        "- [迁移多工作进程 CPU/GPU 训练](multi_worker_cpu_gpu_training.ipynb)"
      ]
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [],
      "name": "migrating_estimator.ipynb",
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}