docs/tutorials/transform/simple.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tghWegsjhpkt"
      },
      "source": [
        "##### Copyright 2021 The TensorFlow Authors.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "rSGJWC5biBiG"
      },
      "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": "j-Iyf5gv5oBq"
      },
      "source": [
        "# Preprocess data with TensorFlow Transform\n",
        "***The Feature Engineering Component of TensorFlow Extended (TFX)***"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S5ST8dI25wbA"
      },
      "source": [
    "Note: We recommend running this tutorial in a Colab notebook, with no setup required!  Just click \"Run in Google Colab\".\n",
    "\n",
    "<div class=\"buttons-wrapper\">\n",
    "  <a class=\"md-button\" target=\"_blank\" href=\n",
    "     \"https://www.tensorflow.org/tfx/tutorials/transform/simple\">\n",
    "    <div class=\"buttons-content\">\n",
    "      <img src=\"https://www.tensorflow.org/images/tf_logo_32px.png\">\n",
    "      View on TensorFlow.org\n",
    "    </div>\n",
    "  </a>\n",
    "  <a class=\"md-button\" target=\"_blank\" href=\n",
    "     \"https://colab.research.google.com/github/tensorflow/tfx/blob/master/docs/tutorials/transform/simple.ipynb\">\n",
    "    <div class=\"buttons-content\">\n",
    "      <img src=\n",
    "\t   \"https://www.tensorflow.org/images/colab_logo_32px.png\">\n",
    "      Run in Google Colab\n",
    "    </div>\n",
    "  </a>\n",
    "  <a class=\"md-button\" target=\"_blank\" href=\n",
    "     \"https://github.com/tensorflow/tfx/tree/master/docs/tutorials/transform/simple.ipynb\">\n",
    "    <div class=\"buttons-content\">\n",
    "      <img width=\"32px\" src=\n",
    "\t   \"https://www.tensorflow.org/images/GitHub-Mark-32px.png\">\n",
    "      View source on GitHub\n",
    "    </div>\n",
    "  </a>\n",
    "  <a class=\"md-button\" href=\n",
    "     \"https://storage.googleapis.com/tensorflow_docs/tfx/docs/tutorials/transform/simple.ipynb\">\n",
    "    <div class=\"buttons-content\">\n",
    "      <img src=\n",
    "\t   \"https://www.tensorflow.org/images/download_logo_32px.png\">\n",
    "      Download notebook\n",
    "    </div>\n",
    "  </a>\n",
    "</div>"
   ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mPt5BHTwy_0F"
      },
      "source": [
        "This example colab notebook provides a very simple example of how \u003ca target='_blank' href='https://www.tensorflow.org/tfx/transform/get_started/'\u003eTensorFlow Transform (\u003ccode\u003etf.Transform\u003c/code\u003e)\u003c/a\u003e can be used to preprocess data using exactly the same code for both training a model and serving inferences in production.\n",
        "\n",
        "TensorFlow Transform is a library for preprocessing input data for TensorFlow, including creating features that require a full pass over the training dataset.  For example, using TensorFlow Transform you could:\n",
        "\n",
        "* Normalize an input value by using the mean and standard deviation\n",
        "* Convert strings to integers by generating a vocabulary over all of the input values\n",
        "* Convert floats to integers by assigning them to buckets, based on the observed data distribution\n",
        "\n",
        "TensorFlow has built-in support for manipulations on a single example or a batch of examples. `tf.Transform` extends these capabilities to support full passes over the entire training dataset.\n",
        "\n",
        "The output of `tf.Transform` is exported as a TensorFlow graph which you can use for both training and serving. Using the same graph for both training and serving can prevent skew, since the same transformations are applied in both stages."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6c8lD3uQm8m5"
      },
      "source": [
        "### Upgrade Pip\n",
        "\n",
        "To avoid upgrading Pip in a system when running locally, check to make sure that we're running in Colab.  Local systems can of course be upgraded separately."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "EmiQXNLZm8z-"
      },
      "outputs": [],
      "source": [
        "try:\n",
        "  import colab\n",
        "  !pip install --upgrade pip\n",
        "except:\n",
        "  pass"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hiBxgnc-m8-X"
      },
      "source": [
        "### Install TensorFlow Transform"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "j2CTKbMNm9I4"
      },
      "outputs": [],
      "source": [
        "!pip install -q -U tensorflow_transform"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "R0mXLOJR_-dv"
      },
      "outputs": [],
      "source": [
        "# This cell is only necessary because packages were installed while python was\n",
        "# running. It avoids the need to restart the runtime when running in Colab.\n",
        "import pkg_resources\n",
        "import importlib\n",
        "\n",
        "importlib.reload(pkg_resources)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RptgLn2RYuK3"
      },
      "source": [
        "## Imports"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "K4QXVIM7iglN"
      },
      "outputs": [],
      "source": [
        "import pathlib\n",
        "import pprint\n",
        "import tempfile\n",
        "\n",
        "import tensorflow as tf\n",
        "import tensorflow_transform as tft\n",
        "\n",
        "import tensorflow_transform.beam as tft_beam\n",
        "from tensorflow_transform.tf_metadata import dataset_metadata\n",
        "from tensorflow_transform.tf_metadata import schema_utils\n",
        "from tensorflow_transform.keras_lib import tf_keras"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CxOxaaOYRfl7"
      },
      "source": [
        "## Data: Create some dummy data\n",
        "We'll create some simple dummy data for our simple example:\n",
        "\n",
        "* `raw_data` is the initial raw data that we're going to preprocess\n",
        "* `raw_data_metadata` contains the schema that tells us the types of each of the columns in `raw_data`.  In this case, it's very simple."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-R236Tkf_ON3"
      },
      "outputs": [],
      "source": [
        "raw_data = [\n",
        "      {'x': 1, 'y': 1, 's': 'hello'},\n",
        "      {'x': 2, 'y': 2, 's': 'world'},\n",
        "      {'x': 3, 'y': 3, 's': 'hello'}\n",
        "  ]\n",
        "\n",
        "raw_data_metadata = dataset_metadata.DatasetMetadata(\n",
        "    schema_utils.schema_from_feature_spec({\n",
        "        'y': tf.io.FixedLenFeature([], tf.float32),\n",
        "        'x': tf.io.FixedLenFeature([], tf.float32),\n",
        "        's': tf.io.FixedLenFeature([], tf.string),\n",
        "    }))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xexbWmQEBUBZ"
      },
      "source": [
        "## Transform: Create a preprocessing function\n",
        "\n",
        "The _preprocessing function_ is the most important concept of tf.Transform. A preprocessing function is where the transformation of the dataset really happens. It accepts and returns a dictionary of tensors, where a tensor means a \u003ca target='_blank' href='https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/Tensor'\u003e\u003ccode\u003eTensor\u003c/code\u003e\u003c/a\u003e or \u003ca target='_blank' href='https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/SparseTensor'\u003e\u003ccode\u003eSparseTensor\u003c/code\u003e\u003c/a\u003e. There are two main groups of API calls that typically form the heart of a preprocessing function:"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Zadh6MXLS3eD"
      },
      "source": [
        "1. **TensorFlow Ops:** Any function that accepts and returns tensors, which usually means TensorFlow ops. These add TensorFlow operations to the graph that transforms raw data into transformed data one feature vector at a time.  These will run for every example, during both training and serving.\n",
        "2. **Tensorflow Transform Analyzers/Mappers:** Any of the analyzers/mappers provided by tf.Transform. These also accept and return tensors, and typically contain a combination of Tensorflow ops and Beam computation, but unlike TensorFlow ops they only run in the Beam pipeline during analysis requiring a full pass over the entire training dataset. The Beam computation runs only once, (prior to training, during analysis), and typically make a full pass over the entire training dataset. They create `tf.constant` tensors, which are added to your graph. For example, `tft.min` computes the minimum of a tensor over the training dataset.\n",
        "\n",
        "Caution: When you apply your preprocessing function to serving inferences, the constants that were created by analyzers during training do not change.  If your data has trend or seasonality components, plan accordingly.\n",
        "\n",
        "Note: The `preprocessing_fn` is not directly callable. This means that\n",
        "calling `preprocessing_fn(raw_data)` will not work. Instead, it must\n",
        "be passed to the Transform Beam API as shown in the following cells."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "H2wANNF_2dCR"
      },
      "outputs": [],
      "source": [
        "def preprocessing_fn(inputs):\n",
        "    \"\"\"Preprocess input columns into transformed columns.\"\"\"\n",
        "    x = inputs['x']\n",
        "    y = inputs['y']\n",
        "    s = inputs['s']\n",
        "    x_centered = x - tft.mean(x)\n",
        "    y_normalized = tft.scale_to_0_1(y)\n",
        "    s_integerized = tft.compute_and_apply_vocabulary(s)\n",
        "    x_centered_times_y_normalized = (x_centered * y_normalized)\n",
        "    return {\n",
        "        'x_centered': x_centered,\n",
        "        'y_normalized': y_normalized,\n",
        "        's_integerized': s_integerized,\n",
        "        'x_centered_times_y_normalized': x_centered_times_y_normalized,\n",
        "    }"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cSl9qyTCbBKR"
      },
      "source": [
        "## Syntax\n",
        "\n",
        "You're almost ready to put everything together and use \u003ca target='_blank' href='https://beam.apache.org/'\u003eApache Beam\u003c/a\u003e to run it.\n",
        "\n",
        "Apache Beam uses a \u003ca target='_blank' href='https://beam.apache.org/documentation/programming-guide/#applying-transforms'\u003especial syntax to define and invoke transforms\u003c/a\u003e.  For example, in this line:\n",
        "\n",
        "```\n",
        "result = pass_this | 'name this step' \u003e\u003e to_this_call\n",
        "```\n",
        "\n",
        "The method `to_this_call` is being invoked and passed the object called `pass_this`, and \u003ca target='_blank' href='https://stackoverflow.com/questions/50519662/what-does-the-redirection-mean-in-apache-beam-python'\u003ethis operation will be referred to as `name this step` in a stack trace\u003c/a\u003e.  The result of the call to `to_this_call` is returned in `result`.  You will often see stages of a pipeline chained together like this:\n",
        "\n",
        "```\n",
        "result = apache_beam.Pipeline() | 'first step' \u003e\u003e do_this_first() | 'second step' \u003e\u003e do_this_last()\n",
        "```\n",
        "\n",
        "and since that started with a new pipeline, you can continue like this:\n",
        "\n",
        "```\n",
        "next_result = result | 'doing more stuff' \u003e\u003e another_function()\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5kLDSxOQ8xgg"
      },
      "source": [
        "## Putting it all together\n",
        "Now we're ready to transform our data. We'll use Apache Beam with a direct runner, and supply three inputs:\n",
        "\n",
        "1. `raw_data` - The raw input data that we created above\n",
        "2. `raw_data_metadata` - The schema for the raw data\n",
        "3. `preprocessing_fn` - The function that we created to do our transformation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "mAF9w7RTZU7c"
      },
      "outputs": [],
      "source": [
        "def main(output_dir):\n",
        "  # Ignore the warnings\n",
        "  with tft_beam.Context(temp_dir=tempfile.mkdtemp()):\n",
        "    transformed_dataset, transform_fn = (  # pylint: disable=unused-variable\n",
        "        (raw_data, raw_data_metadata) | tft_beam.AnalyzeAndTransformDataset(\n",
        "            preprocessing_fn))\n",
        "\n",
        "  transformed_data, transformed_metadata = transformed_dataset  # pylint: disable=unused-variable\n",
        "\n",
        "  # Save the transform_fn to the output_dir\n",
        "  _ = (\n",
        "      transform_fn\n",
        "      | 'WriteTransformFn' \u003e\u003e tft_beam.WriteTransformFn(output_dir))\n",
        "\n",
        "  return transformed_data, transformed_metadata"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "zZPQl0X19ni2"
      },
      "outputs": [],
      "source": [
        "output_dir = pathlib.Path(tempfile.mkdtemp())\n",
        "\n",
        "transformed_data, transformed_metadata = main(str(output_dir))\n",
        "\n",
        "print('\\nRaw data:\\n{}\\n'.format(pprint.pformat(raw_data)))\n",
        "print('Transformed data:\\n{}'.format(pprint.pformat(transformed_data)))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NO6LyTneNndy"
      },
      "source": [
        "## Is this the right answer?\n",
        "Previously, we used `tf.Transform` to do this:\n",
        "```\n",
        "x_centered = x - tft.mean(x)\n",
        "y_normalized = tft.scale_to_0_1(y)\n",
        "s_integerized = tft.compute_and_apply_vocabulary(s)\n",
        "x_centered_times_y_normalized = (x_centered * y_normalized)\n",
        "```\n",
        "\n",
        "* **x_centered** - With input of `[1, 2, 3]` the mean of x is 2, and we subtract it from x to center our x values at 0.  So our result of `[-1.0, 0.0, 1.0]` is correct.\n",
        "* **y_normalized** - We wanted to scale our y values between 0 and 1.  Our input was `[1, 2, 3]` so our result of `[0.0, 0.5, 1.0]` is correct.\n",
        "* **s_integerized** - We wanted to map our strings to indexes in a vocabulary, and there were only 2 words in our vocabulary (\"hello\" and \"world\").  So with input of `[\"hello\", \"world\", \"hello\"]` our result of `[0, 1, 0]` is correct. Since \"hello\" occurs most frequently in this data, it will be the first entry in the vocabulary.\n",
        "* **x_centered_times_y_normalized** - We wanted to create a new feature by crossing `x_centered` and `y_normalized` using multiplication.  Note that this multiplies the results, not the original values, and our new result of `[-0.0, 0.0, 1.0]` is correct."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dXw790Sr8Jws"
      },
      "source": [
        "## Use the resulting `transform_fn`"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "We4Mafrq8id6"
      },
      "outputs": [],
      "source": [
        "!ls -l {output_dir}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SoaaAXxk_vWP"
      },
      "source": [
        "The `transform_fn/` directory contains a `tf.saved_model` implementing with all the constants tensorflow-transform analysis results built into the graph. \n",
        "\n",
        "It is possible to load this directly with `tf.saved_model.load`, but this not easy to use:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "cz8dqFW6ANJQ"
      },
      "outputs": [],
      "source": [
        "loaded = tf.saved_model.load(str(output_dir/'transform_fn'))\n",
        "loaded.signatures['serving_default']"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZCugaxMiBosA"
      },
      "source": [
        "A better approach is to load it using `tft.TFTransformOutput`. The `TFTransformOutput.transform_features_layer` method returns a `tft.TransformFeaturesLayer` object that can be used to apply the transformation:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "HNd4r2gJ75nx"
      },
      "outputs": [],
      "source": [
        "tf_transform_output = tft.TFTransformOutput(output_dir)\n",
        "\n",
        "tft_layer = tf_transform_output.transform_features_layer()\n",
        "tft_layer"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "se-M1zx49kTY"
      },
      "source": [
        "This `tft.TransformFeaturesLayer` expects a dictionary of batched features. So create a `Dict[str, tf.Tensor]` from the `List[Dict[str, Any]]` in `raw_data`:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "2nyE1fVj82Gp"
      },
      "outputs": [],
      "source": [
        "raw_data_batch = {\n",
        "    's': tf.constant([ex['s'] for ex in raw_data]),\n",
        "    'x': tf.constant([ex['x'] for ex in raw_data], dtype=tf.float32),\n",
        "    'y': tf.constant([ex['y'] for ex in raw_data], dtype=tf.float32),\n",
        "}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "016sJ_cD_gVC"
      },
      "source": [
        "You can use the `tft.TransformFeaturesLayer` on its own:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "fIXJYE0Z9Mrs"
      },
      "outputs": [],
      "source": [
        "transformed_batch = tft_layer(raw_data_batch)\n",
        "\n",
        "{key: value.numpy() for key, value in transformed_batch.items()}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FBfO5cp-8pqb"
      },
      "source": [
        "## Export"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "B3SN7D-FzrZ3"
      },
      "source": [
        "A more typical use case would use `tf.Transform` to apply the transformation to the training and evaluation datasets (see the [next tutorial](census.ipynb) for an example). Then, after training, before exporting the model attach the `tft.TransformFeaturesLayer` as the first layer so that you can export it as part of your `tf.saved_model`. For a concrete example, keep reading."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lYV_oy5s9Dn9"
      },
      "source": [
        "### An example training model\n",
        "\n",
        "Below is a model that:\n",
        "\n",
        "1. takes the transformed batch,\n",
        "2. stacks them all together into a simple `(batch, features)` matrix,\n",
        "3. runs them through a few dense layers, and\n",
        "4. produces 10 linear outputs.\n",
        "\n",
        "In a real use case you would apply a one-hot to the `s_integerized` feature.\n",
        "\n",
        "You could train this model on a dataset transformed by `tf.Transform`:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "xWiEo1ZUzp4x"
      },
      "outputs": [],
      "source": [
        "class StackDict(tf_keras.layers.Layer):\n",
        "  def call(self, inputs):\n",
        "    values = [\n",
        "        tf.cast(v, tf.float32)\n",
        "        for k,v in sorted(inputs.items(), key=lambda kv: kv[0])]\n",
        "    return tf.stack(values, axis=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "A0QJpoWT1aUD"
      },
      "outputs": [],
      "source": [
        "class TrainedModel(tf_keras.Model):\n",
        "  def __init__(self):\n",
        "    super().__init__(self)\n",
        "    self.concat = StackDict()\n",
        "    self.body = tf_keras.Sequential([\n",
        "        tf_keras.layers.Dense(64, activation='relu'),\n",
        "        tf_keras.layers.Dense(64, activation='relu'),\n",
        "        tf_keras.layers.Dense(10),\n",
        "    ])\n",
        "\n",
        "  def call(self, inputs, training=None):\n",
        "    x = self.concat(inputs)\n",
        "    return self.body(x, training)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "DkMwREIx2fkD"
      },
      "outputs": [],
      "source": [
        "trained_model = TrainedModel()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uBwnbh1Q-TBK"
      },
      "source": [
        "Imagine we trained the model.\n",
        "\n",
        "```\n",
        "trained_model.compile(loss=..., optimizer='adam')\n",
        "trained_model.fit(...)\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "notFxUC0AFs6"
      },
      "source": [
        "This model runs on the transformed inputs"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "d2KJ8nGt228O"
      },
      "outputs": [],
      "source": [
        "trained_model_output = trained_model(transformed_batch)\n",
        "trained_model_output.shape"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fzWs35Ki6M5c"
      },
      "source": [
        "### An example export wrapper\n",
        "\n",
        "Imagine you've trained the above model and want to export it.\n",
        "\n",
        "You'll want to include the transform function in the exported model:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Pe-nbN123qUt"
      },
      "outputs": [],
      "source": [
        "class ExportModel(tf.Module):\n",
        "  def __init__(self, trained_model, input_transform):\n",
        "    self.trained_model = trained_model\n",
        "    self.input_transform = input_transform\n",
        "\n",
        "  @tf.function\n",
        "  def __call__(self, inputs, training=None):\n",
        "    x = self.input_transform(inputs)\n",
        "    return self.trained_model(x)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "iLUIO-Y87AC0"
      },
      "outputs": [],
      "source": [
        "export_model = ExportModel(trained_model=trained_model,\n",
        "                           input_transform=tft_layer)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sFDYQDgU7ozE"
      },
      "source": [
        "This combined model works on the raw data, and produces exactly the same results as calling the trained model directly:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "AqwHTex27ILk"
      },
      "outputs": [],
      "source": [
        "export_model_output = export_model(raw_data_batch)\n",
        "export_model_output.shape"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "AZQ6_Dfd7xws"
      },
      "outputs": [],
      "source": [
        "tf.reduce_max(abs(export_model_output - trained_model_output)).numpy()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5r-lH_nh8PM-"
      },
      "source": [
        "This `export_model` includes the `tft.TransformFeaturesLayer` and is entirely self-contained. You can save it and restore it in another environment and still get exactly the same result:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "VK17CShl8F7s"
      },
      "outputs": [],
      "source": [
        "import tempfile\n",
        "model_dir = tempfile.mkdtemp(suffix='tft')\n",
        "\n",
        "tf.saved_model.save(export_model, model_dir)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "RTF-yRnA9yrL"
      },
      "outputs": [],
      "source": [
        "reloaded = tf.saved_model.load(model_dir)\n",
        "\n",
        "reloaded_model_output = reloaded(raw_data_batch)\n",
        "reloaded_model_output.shape"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "tFx1I6FQ9_mj"
      },
      "outputs": [],
      "source": [
        "tf.reduce_max(abs(export_model_output - reloaded_model_output)).numpy()"
      ]
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [
        "tghWegsjhpkt",
        "cSl9qyTCbBKR",
        "NO6LyTneNndy"
      ],
      "name": "simple.ipynb",
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}