{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "Tce3stUlHN0L"
},
"source": [
"##### Copyright 2020 The TensorFlow Authors."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"cellView": "form",
"execution": {
"iopub.execute_input": "2021-02-13T03:02:53.247295Z",
"iopub.status.busy": "2021-02-13T03:02:53.246506Z",
"iopub.status.idle": "2021-02-13T03:02:53.249004Z",
"shell.execute_reply": "2021-02-13T03:02:53.248485Z"
},
"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": "markdown",
"metadata": {
"id": "qFdPvlXBOdUN"
},
"source": [
"# Keras の例による量子化認識トレーニング"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MfBg1C5NB3X0"
},
"source": [
""
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "Bjmi3qZeu_xk"
},
"source": [
"## 概要\n",
"\n",
"*量子化認識トレーニング*のエンドツーエンドの例へようこそ。\n",
"\n",
"### その他のページ\n",
"\n",
"量子化認識トレーニングの紹介、および認識トレーニングを使用すべきかどうかの判定（サポート情報も含む）については、[概要ページ](https://www.tensorflow.org/model_optimization/guide/quantization/training.md)をご覧ください。\n",
"\n",
"ユースケースに合った API を素早く特定するには（8 ビットのモデルの完全量子化を超えるユースケース）、[総合ガイド](https://www.tensorflow.org/model_optimization/guide/quantization/training_comprehensive_guide.md)をご覧ください。\n",
"\n",
"### 要約\n",
"\n",
"このチュートリアルでは、次について説明しています。\n",
"\n",
"1. MNIST の `tf.keras` モデルを最初からトレーニングする。\n",
"2. 量子化認識トレーニング API を適用してモデルをファインチューニングし、精度を確認して量子化認識モデルをエクスポートする。\n",
"3. このモデルを使用して、TFLite バックエンドのために実際に量子化されたモデルを作成する。\n",
"4. TFLite および 1/4 のモデルの精度の永続性を確認する。モバイルでのレイテンシーのメリットを確認するには、[TFLite アプリリポジトリ内の](https://www.tensorflow.org/lite/models) TFLite の例を試してみてください。"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "yEAZYXvZU_XG"
},
"source": [
"## セットアップ"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:02:53.260043Z",
"iopub.status.busy": "2021-02-13T03:02:53.259328Z",
"iopub.status.idle": "2021-02-13T03:03:19.366045Z",
"shell.execute_reply": "2021-02-13T03:03:19.365425Z"
},
"id": "zN4yVFK5-0Bf"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Found existing installation: tensorflow 2.4.1\r\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Uninstalling tensorflow-2.4.1:\r\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
" Successfully uninstalled tensorflow-2.4.1\r\n"
]
}
],
"source": [
"! pip uninstall -y tensorflow\n",
"! pip install -q tf-nightly\n",
"! pip install -q tensorflow-model-optimization\n"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:19.371559Z",
"iopub.status.busy": "2021-02-13T03:03:19.370847Z",
"iopub.status.idle": "2021-02-13T03:03:25.956453Z",
"shell.execute_reply": "2021-02-13T03:03:25.956935Z"
},
"id": "yJwIonXEVJo6"
},
"outputs": [],
"source": [
"import tempfile\n",
"import os\n",
"\n",
"import tensorflow as tf\n",
"\n",
"from tensorflow import keras"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "psViY5PRDurp"
},
"source": [
"## 量子化認識トレーニングを使用せずに、MNIST のモデルをトレーニングする"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:25.964690Z",
"iopub.status.busy": "2021-02-13T03:03:25.963989Z",
"iopub.status.idle": "2021-02-13T03:03:34.754140Z",
"shell.execute_reply": "2021-02-13T03:03:34.754563Z"
},
"id": "pbY-KGMPvbW9"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\r",
" 1/1688 [..............................] - ETA: 57:15 - loss: 2.3546 - accuracy: 0.0000e+00"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 23/1688 [..............................] - ETA: 3s - loss: 2.1957 - accuracy: 0.2001 "
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 46/1688 [..............................] - ETA: 3s - loss: 2.0404 - accuracy: 0.3254"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 71/1688 [>.............................] - ETA: 3s - loss: 1.8700 - accuracy: 0.4154"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 96/1688 [>.............................] - ETA: 3s - loss: 1.7193 - accuracy: 0.4782"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 121/1688 [=>............................] - ETA: 3s - loss: 1.5970 - accuracy: 0.5232"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 145/1688 [=>............................] - ETA: 3s - loss: 1.5011 - accuracy: 0.5566"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 170/1688 [==>...........................] - ETA: 3s - loss: 1.4166 - accuracy: 0.5850"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 195/1688 [==>...........................] - ETA: 3s - loss: 1.3448 - accuracy: 0.6084"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 220/1688 [==>...........................] - ETA: 3s - loss: 1.2831 - accuracy: 0.6279"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 245/1688 [===>..........................] - ETA: 3s - loss: 1.2294 - accuracy: 0.6446"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 270/1688 [===>..........................] - ETA: 2s - loss: 1.1820 - accuracy: 0.6592"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 295/1688 [====>.........................] - ETA: 2s - loss: 1.1399 - accuracy: 0.6720"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 320/1688 [====>.........................] - ETA: 2s - loss: 1.1022 - accuracy: 0.6834"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 345/1688 [=====>........................] - ETA: 2s - loss: 1.0682 - accuracy: 0.6936"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 369/1688 [=====>........................] - ETA: 2s - loss: 1.0384 - accuracy: 0.7026"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 393/1688 [=====>........................] - ETA: 2s - loss: 1.0111 - accuracy: 0.7107"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 416/1688 [======>.......................] - ETA: 2s - loss: 0.9868 - accuracy: 0.7179"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 441/1688 [======>.......................] - ETA: 2s - loss: 0.9625 - accuracy: 0.7251"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 466/1688 [=======>......................] - ETA: 2s - loss: 0.9400 - accuracy: 0.7318"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 491/1688 [=======>......................] - ETA: 2s - loss: 0.9192 - accuracy: 0.7379"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 515/1688 [========>.....................] - ETA: 2s - loss: 0.9006 - accuracy: 0.7433"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 539/1688 [========>.....................] - ETA: 2s - loss: 0.8832 - accuracy: 0.7484"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 565/1688 [=========>....................] - ETA: 2s - loss: 0.8654 - accuracy: 0.7535"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 590/1688 [=========>....................] - ETA: 2s - loss: 0.8494 - accuracy: 0.7582"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 615/1688 [=========>....................] - ETA: 2s - loss: 0.8343 - accuracy: 0.7626"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 640/1688 [==========>...................] - ETA: 2s - loss: 0.8201 - accuracy: 0.7667"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 665/1688 [==========>...................] - ETA: 2s - loss: 0.8065 - accuracy: 0.7706"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 690/1688 [===========>..................] - ETA: 2s - loss: 0.7937 - accuracy: 0.7743"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 715/1688 [===========>..................] - ETA: 2s - loss: 0.7815 - accuracy: 0.7778"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 740/1688 [============>.................] - ETA: 1s - loss: 0.7698 - accuracy: 0.7812"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 765/1688 [============>.................] - ETA: 1s - loss: 0.7587 - accuracy: 0.7844"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 790/1688 [=============>................] - ETA: 1s - loss: 0.7481 - accuracy: 0.7875"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 815/1688 [=============>................] - ETA: 1s - loss: 0.7379 - accuracy: 0.7904"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 839/1688 [=============>................] - ETA: 1s - loss: 0.7285 - accuracy: 0.7931"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 863/1688 [==============>...............] - ETA: 1s - loss: 0.7195 - accuracy: 0.7957"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 888/1688 [==============>...............] - ETA: 1s - loss: 0.7104 - accuracy: 0.7983"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 913/1688 [===============>..............] - ETA: 1s - loss: 0.7017 - accuracy: 0.8008"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 938/1688 [===============>..............] - ETA: 1s - loss: 0.6933 - accuracy: 0.8033"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 963/1688 [================>.............] - ETA: 1s - loss: 0.6851 - accuracy: 0.8056"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
" 986/1688 [================>.............] - ETA: 1s - loss: 0.6779 - accuracy: 0.8077"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1011/1688 [================>.............] - ETA: 1s - loss: 0.6703 - accuracy: 0.8098"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1036/1688 [=================>............] - ETA: 1s - loss: 0.6630 - accuracy: 0.8119"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1061/1688 [=================>............] - ETA: 1s - loss: 0.6560 - accuracy: 0.8139"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1085/1688 [==================>...........] - ETA: 1s - loss: 0.6494 - accuracy: 0.8158"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1110/1688 [==================>...........] - ETA: 1s - loss: 0.6428 - accuracy: 0.8177"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1135/1688 [===================>..........] - ETA: 1s - loss: 0.6364 - accuracy: 0.8196"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1160/1688 [===================>..........] - ETA: 1s - loss: 0.6301 - accuracy: 0.8213"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1185/1688 [====================>.........] - ETA: 1s - loss: 0.6241 - accuracy: 0.8231"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1210/1688 [====================>.........] - ETA: 0s - loss: 0.6182 - accuracy: 0.8247"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1235/1688 [====================>.........] - ETA: 0s - loss: 0.6125 - accuracy: 0.8264"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1260/1688 [=====================>........] - ETA: 0s - loss: 0.6069 - accuracy: 0.8280"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1285/1688 [=====================>........] - ETA: 0s - loss: 0.6015 - accuracy: 0.8295"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1310/1688 [======================>.......] - ETA: 0s - loss: 0.5962 - accuracy: 0.8310"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1335/1688 [======================>.......] - ETA: 0s - loss: 0.5911 - accuracy: 0.8325"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1359/1688 [=======================>......] - ETA: 0s - loss: 0.5863 - accuracy: 0.8339"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1383/1688 [=======================>......] - ETA: 0s - loss: 0.5816 - accuracy: 0.8352"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1407/1688 [========================>.....] - ETA: 0s - loss: 0.5770 - accuracy: 0.8365"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1432/1688 [========================>.....] - ETA: 0s - loss: 0.5724 - accuracy: 0.8379"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1457/1688 [========================>.....] - ETA: 0s - loss: 0.5678 - accuracy: 0.8391"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1482/1688 [=========================>....] - ETA: 0s - loss: 0.5634 - accuracy: 0.8404"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1506/1688 [=========================>....] - ETA: 0s - loss: 0.5592 - accuracy: 0.8416"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1531/1688 [==========================>...] - ETA: 0s - loss: 0.5550 - accuracy: 0.8428"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1556/1688 [==========================>...] - ETA: 0s - loss: 0.5509 - accuracy: 0.8440"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1581/1688 [===========================>..] - ETA: 0s - loss: 0.5468 - accuracy: 0.8451"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1606/1688 [===========================>..] - ETA: 0s - loss: 0.5429 - accuracy: 0.8463"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1631/1688 [===========================>..] - ETA: 0s - loss: 0.5390 - accuracy: 0.8474"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1656/1688 [============================>.] - ETA: 0s - loss: 0.5352 - accuracy: 0.8484"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1681/1688 [============================>.] - ETA: 0s - loss: 0.5315 - accuracy: 0.8495"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"1688/1688 [==============================] - 6s 2ms/step - loss: 0.5304 - accuracy: 0.8498 - val_loss: 0.1238 - val_accuracy: 0.9680\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 1,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Load MNIST dataset\n",
"mnist = keras.datasets.mnist\n",
"(train_images, train_labels), (test_images, test_labels) = mnist.load_data()\n",
"\n",
"# Normalize the input image so that each pixel value is between 0 to 1.\n",
"train_images = train_images / 255.0\n",
"test_images = test_images / 255.0\n",
"\n",
"# Define the model architecture.\n",
"model = keras.Sequential([\n",
" keras.layers.InputLayer(input_shape=(28, 28)),\n",
" keras.layers.Reshape(target_shape=(28, 28, 1)),\n",
" keras.layers.Conv2D(filters=12, kernel_size=(3, 3), activation='relu'),\n",
" keras.layers.MaxPooling2D(pool_size=(2, 2)),\n",
" keras.layers.Flatten(),\n",
" keras.layers.Dense(10)\n",
"])\n",
"\n",
"# Train the digit classification model\n",
"model.compile(optimizer='adam',\n",
" loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['accuracy'])\n",
"\n",
"model.fit(\n",
" train_images,\n",
" train_labels,\n",
" epochs=1,\n",
" validation_split=0.1,\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "K8747K9OE72P"
},
"source": [
"## 量子化認識トレーニングを使用して、事前トレーニング済みモデルをクローンおよびファインチューニングする\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "F19k7ExXF_h2"
},
"source": [
"### モデルを定義する"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "JsZROpNYMWQ0"
},
"source": [
"量子化認識トレーニングをモデル全体に適用し、これをモデルの要約で確認します。すべてのレイヤーにプレフィックス \"quant\" が付いているはずです。\n",
"\n",
"結果のモデルは量子化認識モデルですが、量子化はされていないので注意してください（例えば、重みは int8 ではなく float32 です）。次のセクションでは、量子化認識モデルから量子化モデルを作成する方法を示します。\n",
"\n",
"[総合ガイド](https://www.tensorflow.org/model_optimization/guide/quantization/training_comprehensive_guide.md)では、モデルの精度を改善するために、一部のレイヤーを量子化する方法をご覧いただけます。"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:34.760223Z",
"iopub.status.busy": "2021-02-13T03:03:34.759531Z",
"iopub.status.idle": "2021-02-13T03:03:36.232377Z",
"shell.execute_reply": "2021-02-13T03:03:36.231883Z"
},
"id": "oq6blGjgFDCW"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model: \"sequential\"\n",
"_________________________________________________________________\n",
"Layer (type) Output Shape Param # \n",
"=================================================================\n",
"quantize_layer (QuantizeLaye (None, 28, 28) 3 \n",
"_________________________________________________________________\n",
"quant_reshape (QuantizeWrapp (None, 28, 28, 1) 1 \n",
"_________________________________________________________________\n",
"quant_conv2d (QuantizeWrappe (None, 26, 26, 12) 147 \n",
"_________________________________________________________________\n",
"quant_max_pooling2d (Quantiz (None, 13, 13, 12) 1 \n",
"_________________________________________________________________\n",
"quant_flatten (QuantizeWrapp (None, 2028) 1 \n",
"_________________________________________________________________\n",
"quant_dense (QuantizeWrapper (None, 10) 20295 \n",
"=================================================================\n",
"Total params: 20,448\n",
"Trainable params: 20,410\n",
"Non-trainable params: 38\n",
"_________________________________________________________________\n"
]
}
],
"source": [
"import tensorflow_model_optimization as tfmot\n",
"\n",
"quantize_model = tfmot.quantization.keras.quantize_model\n",
"\n",
"# q_aware stands for for quantization aware.\n",
"q_aware_model = quantize_model(model)\n",
"\n",
"# `quantize_model` requires a recompile.\n",
"q_aware_model.compile(optimizer='adam',\n",
" loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['accuracy'])\n",
"\n",
"q_aware_model.summary()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "uDr2ijwpGCI-"
},
"source": [
"### モデルをベースラインに対してトレーニングおよび評価する"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "XUBEn94hXYB1"
},
"source": [
"モデルを 1 エポックだけトレーニングした後のファインチューニングを実証するために、トレーニングデータのサブセットに対して、量子化認識トレーニングを使用してファインチューニングを行います。"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:36.237791Z",
"iopub.status.busy": "2021-02-13T03:03:36.237115Z",
"iopub.status.idle": "2021-02-13T03:03:37.030206Z",
"shell.execute_reply": "2021-02-13T03:03:37.029654Z"
},
"id": "_PHDGJryE31X"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\r",
"1/2 [==============>...............] - ETA: 0s - loss: 0.1671 - accuracy: 0.9520"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r",
"2/2 [==============================] - 1s 210ms/step - loss: 0.1565 - accuracy: 0.9544 - val_loss: 0.1604 - val_accuracy: 0.9700\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 1,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"train_images_subset = train_images[0:1000] # out of 60000\n",
"train_labels_subset = train_labels[0:1000]\n",
"\n",
"q_aware_model.fit(train_images_subset, train_labels_subset,\n",
" batch_size=500, epochs=1, validation_split=0.1)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-byC2lYlMkfN"
},
"source": [
"この例では、ベースラインと比較し、量子化認識トレーニング後のテスト精度の損失は、最小限あるいはゼロです。"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:37.035495Z",
"iopub.status.busy": "2021-02-13T03:03:37.034806Z",
"iopub.status.idle": "2021-02-13T03:03:38.107098Z",
"shell.execute_reply": "2021-02-13T03:03:38.106528Z"
},
"id": "6bMFTKSSHyyZ"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Baseline test accuracy: 0.9642000198364258\n",
"Quant test accuracy: 0.9656000137329102\n"
]
}
],
"source": [
"_, baseline_model_accuracy = model.evaluate(\n",
" test_images, test_labels, verbose=0)\n",
"\n",
"_, q_aware_model_accuracy = q_aware_model.evaluate(\n",
" test_images, test_labels, verbose=0)\n",
"\n",
"print('Baseline test accuracy:', baseline_model_accuracy)\n",
"print('Quant test accuracy:', q_aware_model_accuracy)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "2IepmUPSITn6"
},
"source": [
"## TFLite バックエンドの量子化モデルを作成する"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "1FgNP4rbOLH8"
},
"source": [
"この後に、重み int8 と活性化関数 uint8 を持つ、実際に量子化されたモデルが出来上がります。"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:38.114343Z",
"iopub.status.busy": "2021-02-13T03:03:38.111524Z",
"iopub.status.idle": "2021-02-13T03:03:39.719196Z",
"shell.execute_reply": "2021-02-13T03:03:39.718509Z"
},
"id": "w7fztWsAOHTz"
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING:absl:Found untraced functions such as reshape_layer_call_fn, reshape_layer_call_and_return_conditional_losses, conv2d_layer_call_fn, conv2d_layer_call_and_return_conditional_losses, max_pooling2d_layer_call_fn while saving (showing 5 of 25). These functions will not be directly callable after loading.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: /tmp/tmpupf9tsyx/assets\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: /tmp/tmpupf9tsyx/assets\n"
]
}
],
"source": [
"converter = tf.lite.TFLiteConverter.from_keras_model(q_aware_model)\n",
"converter.optimizations = [tf.lite.Optimize.DEFAULT]\n",
"\n",
"quantized_tflite_model = converter.convert()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "BEYsyYVqNgeY"
},
"source": [
"## TF から TFLite への精度の永続性を確認する"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "saadXD4JQsBK"
},
"source": [
"テストデータセットで TFLite モデルを評価するヘルパー関数を定義します。"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:39.727587Z",
"iopub.status.busy": "2021-02-13T03:03:39.726869Z",
"iopub.status.idle": "2021-02-13T03:03:39.728677Z",
"shell.execute_reply": "2021-02-13T03:03:39.729093Z"
},
"id": "b8yBouuGNqls"
},
"outputs": [],
"source": [
"import numpy as np\n",
"\n",
"def evaluate_model(interpreter):\n",
" input_index = interpreter.get_input_details()[0][\"index\"]\n",
" output_index = interpreter.get_output_details()[0][\"index\"]\n",
"\n",
" # Run predictions on every image in the \"test\" dataset.\n",
" prediction_digits = []\n",
" for i, test_image in enumerate(test_images):\n",
" if i % 1000 == 0:\n",
" print('Evaluated on {n} results so far.'.format(n=i))\n",
" # Pre-processing: add batch dimension and convert to float32 to match with\n",
" # the model's input data format.\n",
" test_image = np.expand_dims(test_image, axis=0).astype(np.float32)\n",
" interpreter.set_tensor(input_index, test_image)\n",
"\n",
" # Run inference.\n",
" interpreter.invoke()\n",
"\n",
" # Post-processing: remove batch dimension and find the digit with highest\n",
" # probability.\n",
" output = interpreter.tensor(output_index)\n",
" digit = np.argmax(output()[0])\n",
" prediction_digits.append(digit)\n",
"\n",
" print('\\n')\n",
" # Compare prediction results with ground truth labels to calculate accuracy.\n",
" prediction_digits = np.array(prediction_digits)\n",
" accuracy = (prediction_digits == test_labels).mean()\n",
" return accuracy"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "TuEFS4CIQvUw"
},
"source": [
"量子化されたモデルを評価し、TensorFlow の精度が TFLite バックエンドに持続されていることを確認します。"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:39.734260Z",
"iopub.status.busy": "2021-02-13T03:03:39.733611Z",
"iopub.status.idle": "2021-02-13T03:03:43.556998Z",
"shell.execute_reply": "2021-02-13T03:03:43.556484Z"
},
"id": "VqQTyqz4NsWd"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 0 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 1000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 2000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 3000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 4000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 5000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 6000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 7000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 8000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluated on 9000 results so far.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"\n",
"Quant TFLite test_accuracy: 0.9656\n",
"Quant TF test accuracy: 0.9656000137329102\n"
]
}
],
"source": [
"interpreter = tf.lite.Interpreter(model_content=quantized_tflite_model)\n",
"interpreter.allocate_tensors()\n",
"\n",
"test_accuracy = evaluate_model(interpreter)\n",
"\n",
"print('Quant TFLite test_accuracy:', test_accuracy)\n",
"print('Quant TF test accuracy:', q_aware_model_accuracy)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "z8D7WnFF5DZR"
},
"source": [
"## 量子化でモデルが 1/4 になることを確認する"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "I1c2IecBRCdQ"
},
"source": [
"浮動小数点数の TFLite モデルを作成して、量子化された TFLite モデルが 1/4 になっていることを確認します。"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"execution": {
"iopub.execute_input": "2021-02-13T03:03:43.564325Z",
"iopub.status.busy": "2021-02-13T03:03:43.563679Z",
"iopub.status.idle": "2021-02-13T03:03:44.138100Z",
"shell.execute_reply": "2021-02-13T03:03:44.137590Z"
},
"id": "jy_Lgfh8VkyX"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: /tmp/tmpx96l01vr/assets\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: /tmp/tmpx96l01vr/assets\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Float model in Mb: 0.08058547973632812\n",
"Quantized model in Mb: 0.0234527587890625\n"
]
}
],
"source": [
"# Create float TFLite model.\n",
"float_converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
"float_tflite_model = float_converter.convert()\n",
"\n",
"# Measure sizes of models.\n",
"_, float_file = tempfile.mkstemp('.tflite')\n",
"_, quant_file = tempfile.mkstemp('.tflite')\n",
"\n",
"with open(quant_file, 'wb') as f:\n",
" f.write(quantized_tflite_model)\n",
"\n",
"with open(float_file, 'wb') as f:\n",
" f.write(float_tflite_model)\n",
"\n",
"print(\"Float model in Mb:\", os.path.getsize(float_file) / float(2**20))\n",
"print(\"Quantized model in Mb:\", os.path.getsize(quant_file) / float(2**20))"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "0O5xuci-SonI"
},
"source": [
"## 結論"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "O2I7xmyMW5QY"
},
"source": [
"このチュートリアルでは、TensorFlow Model Optimization Toolkit API を使用して量子化認識モデルを作成し、TFLite バックエンドの量子化モデルを作成する方法を紹介しました。\n",
"\n",
"MNIST のモデルでは、精度の違いを最小限に抑えながらモデルサイズを 1/4 に圧縮できることを示しましたた。モバイルでのレイテンシーのメリットを確認するには、[TFLite アプリリポジトリ内の](https://www.tensorflow.org/lite/models) TFLite の例を試してみてください。\n",
"\n",
"この新しい機能をぜひお試しください。リソースが制限される環境でのデプロイにおいて、特に重要となります。\n"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"collapsed_sections": [],
"name": "training_example.ipynb",
"toc_visible": true
},
"kernelspec": {
"display_name": "Python 3",
"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": "3.6.9"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
![](\"https://www.tensorflow.org/images/tf_logo_32px.png\"){kind=logo}
TensorFlow.orgで表示![](\"https://www.tensorflow.org/images/colab_logo_32px.png\"){kind=logo}
Google Colab で実行![](\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\")
GitHub でソースを表示![](\"https://www.tensorflow.org/images/download_logo_32px.png\"){kind=logo}
ノートブックをダウンロード