Last active
September 11, 2018 04:16
-
-
Save serithemage/ee86d2d9db762e522287cebfb540e35d to your computer and use it in GitHub Desktop.
mxnet_on_colab.ipynb
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| { | |
| "nbformat": 4, | |
| "nbformat_minor": 0, | |
| "metadata": { | |
| "colab": { | |
| "name": "mxnet_on_colab.ipynb", | |
| "version": "0.3.2", | |
| "provenance": [], | |
| "collapsed_sections": [] | |
| }, | |
| "kernelspec": { | |
| "name": "python3", | |
| "display_name": "Python 3" | |
| }, | |
| "accelerator": "GPU" | |
| }, | |
| "cells": [ | |
| { | |
| "metadata": { | |
| "id": "sxTZKk9GfQAQ", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "# Google Colaboratory에서 Keras의 백엔드로서 MXNet을 설정하는 방법\n", | |
| "\n", | |
| "> 이 문서는 [Shikoan](https://blog.shikoan.com/)님이 작성한 [\"Google ColaboratoryでMXNetバックエンドの高速Kerasを体験しよう\"](https://qiita.com/koshian2/items/b688bb8ac05e1c956e9f)를 번역한 것 입니다. 번역을 허락해 주신 Shikoan님에게 감사드립니다." | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "s0g887LoDd06", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "## Keras + MXNet은 빠른다\n", | |
| "---\n", | |
| "Keras의 MXNet백엔드는, Amazon에 의해 개발이 진행되고 있습니다.\n", | |
| "\n", | |
| "[이전에 작성한 글](https://qiita.com/koshian2/items/9877ed4fb3716eac0c37)에서 Keras는 느리다고 했더니 MXNet를 백엔드로 사용하면 빠르다고 지적해 주신 분들이 있었기 때문에 어떻게든 Google Colab에서 사용하는 방법을 궁리해 보았습니다. Colab에서 이를 설정하는 정보가 전혀 없어기 때문에 이 글을 작성합니다.\n", | |
| "\n", | |
| "| Instance Type | GPUs | Batch Size | Keras-MXNet (img/sec) | Keras-TensorFlow (img/sec) |\n", | |
| "|---------------|------|------------|-----------------------|----------------------------|\n", | |
| "| C5.18X Large | 0 | 32 | 87 | 59 |\n", | |
| "| P3.8X Large | 1 | 32 | 831 | 509 |\n", | |
| "| P3.8X Large | 4 | 128 | 1783 | 699 |\n", | |
| "| P3.16X Large | 8 | 256 | 1680 | 435 |\n", | |
| "\n", | |
| "출전 : https://github.com/awslabs/keras-apache-mxnet/tree/master/benchmark\n", | |
| "\n", | |
| "이게 진짠가 싶을 정도 인데요, 특히 GPU수가 늘어날수록 차이가 많이 나네요. 꼭 한번 여러분도 체험 해 보시기 바랍니다." | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "LA4NIO6JMpe0", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "## 인스톨\n", | |
| "---\n", | |
| "2018년 8월 현재, Keras는 공식적으로는 MXNet을 백엔드로 대응하지 않고 있기 때문에, 포크된 Keras(keras-mxnet)을 사용합니다.\n", | |
| "\n", | |
| "Google Colab의 GPU엑셀레이터를 활성화 한다음 GPU대응의 MXNet을 인스톨 합니다.\n" | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "7b2ZcEx-f0By", | |
| "colab_type": "code", | |
| "colab": { | |
| "base_uri": "https://localhost:8080/", | |
| "height": 258 | |
| }, | |
| "outputId": "6733f639-d0b2-456b-dd5a-2e2a56f37124" | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "!pip install mxnet-cu80\n" | |
| ], | |
| "execution_count": 1, | |
| "outputs": [ | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "Collecting mxnet-cu80\n", | |
| "\u001b[?25l Downloading https://files.pythonhosted.org/packages/f6/6c/566a1d4b8b1005b7d9ccfaecd7632f6dca596246f6657827b2d4e97c72c7/mxnet_cu80-1.2.1-py2.py3-none-manylinux1_x86_64.whl (299.1MB)\n", | |
| "\u001b[K 100% |████████████████████████████████| 299.1MB 123kB/s \n", | |
| "\u001b[?25hRequirement already satisfied: requests<2.19.0,>=2.18.4 in /usr/local/lib/python3.6/dist-packages (from mxnet-cu80) (2.18.4)\n", | |
| "Collecting graphviz<0.9.0,>=0.8.1 (from mxnet-cu80)\n", | |
| " Downloading https://files.pythonhosted.org/packages/53/39/4ab213673844e0c004bed8a0781a0721a3f6bb23eb8854ee75c236428892/graphviz-0.8.4-py2.py3-none-any.whl\n", | |
| "Requirement already satisfied: numpy<1.15.0,>=1.8.2 in /usr/local/lib/python3.6/dist-packages (from mxnet-cu80) (1.14.5)\n", | |
| "Requirement already satisfied: idna<2.7,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests<2.19.0,>=2.18.4->mxnet-cu80) (2.6)\n", | |
| "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests<2.19.0,>=2.18.4->mxnet-cu80) (3.0.4)\n", | |
| "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests<2.19.0,>=2.18.4->mxnet-cu80) (2018.8.24)\n", | |
| "Requirement already satisfied: urllib3<1.23,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests<2.19.0,>=2.18.4->mxnet-cu80) (1.22)\n", | |
| "Installing collected packages: graphviz, mxnet-cu80\n", | |
| "Successfully installed graphviz-0.8.4 mxnet-cu80-1.2.1\n" | |
| ], | |
| "name": "stdout" | |
| } | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "WNZ7O2L3NnWl", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "CUDA 9.0대응의 MXNet을 인스톨 하면, import keras실행시에 libcudart.so.9.0: cannot open shared object file: No such file or directory라는 에러메세지가 표시됩니다. CUDA 8.0의 MXNet을 인스톨 하는 경우엔 문제 없었습니다. 이 문제에 대한 해결책이 있다면 알려주세요.\n", | |
| "\n", | |
| "그 다음에는 keras-mxnet도 마찬가지로 인스톨합니다." | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "KpjgOo4C8ggy", | |
| "colab_type": "code", | |
| "colab": { | |
| "base_uri": "https://localhost:8080/", | |
| "height": 309 | |
| }, | |
| "outputId": "a966343c-c59b-4f02-c96f-8bcfc365f936" | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "!pip install keras-mxnet" | |
| ], | |
| "execution_count": 2, | |
| "outputs": [ | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "Collecting keras-mxnet\n", | |
| "\u001b[?25l Downloading https://files.pythonhosted.org/packages/fc/96/9276f9f6fc8f55f8e4fdf0a481d6cf132bb0c91aa0454e1c82dd739dd720/keras_mxnet-2.2.2-py2.py3-none-any.whl (353kB)\n", | |
| "\u001b[K 100% |████████████████████████████████| 358kB 5.7MB/s \n", | |
| "\u001b[?25hRequirement already satisfied: h5py>=2.7.1 in /usr/local/lib/python3.6/dist-packages (from keras-mxnet) (2.8.0)\n", | |
| "Collecting keras-preprocessing==1.0.1 (from keras-mxnet)\n", | |
| " Downloading https://files.pythonhosted.org/packages/f8/33/275506afe1d96b221f66f95adba94d1b73f6b6087cfb6132a5655b6fe338/Keras_Preprocessing-1.0.1-py2.py3-none-any.whl\n", | |
| "Requirement already satisfied: six>=1.9.0 in /usr/local/lib/python3.6/dist-packages (from keras-mxnet) (1.11.0)\n", | |
| "Requirement already satisfied: numpy>=1.9.1 in /usr/local/lib/python3.6/dist-packages (from keras-mxnet) (1.14.5)\n", | |
| "Collecting keras-applications==1.0.4 (from keras-mxnet)\n", | |
| "\u001b[?25l Downloading https://files.pythonhosted.org/packages/54/90/8f327deaa37a71caddb59b7b4aaa9d4b3e90c0e76f8c2d1572005278ddc5/Keras_Applications-1.0.4-py2.py3-none-any.whl (43kB)\n", | |
| "\u001b[K 100% |████████████████████████████████| 51kB 5.6MB/s \n", | |
| "\u001b[?25hRequirement already satisfied: scipy>=0.14 in /usr/local/lib/python3.6/dist-packages (from keras-mxnet) (0.19.1)\n", | |
| "Requirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (from keras-mxnet) (3.13)\n", | |
| "Requirement already satisfied: keras>=2.1.6 in /usr/local/lib/python3.6/dist-packages (from keras-preprocessing==1.0.1->keras-mxnet) (2.1.6)\n", | |
| "Installing collected packages: keras-preprocessing, keras-applications, keras-mxnet\n", | |
| "Successfully installed keras-applications-1.0.4 keras-mxnet-2.2.2 keras-preprocessing-1.0.1\n" | |
| ], | |
| "name": "stdout" | |
| } | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "28hRrC5EO2Ir", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "## Channels_first로 전환\n", | |
| "---\n", | |
| "CNN에서 사용하는 경우, 이 부분만 주의할 필요가 있습니다. **MXNet은 Channel_first** 이므로, 예를 들면 CIFAR라면 (50000, 32, 32, 3)이 아니라 (50000, 3, 32, 32)가 됩니다. TensorFlow 처럼 channel_last로 작성하면 실패하기 때문이라고 생각합니다.\n", | |
| "\n", | |
| "## Keras.json의 편집\n", | |
| "\n", | |
| "디폴트를 channels_last로부터 channels_first로 전환합니다. Jupyter Notebook에서도 가능합니다. 다음의 코드를 실행해 주세요.\n" | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "UJZorQ768ttQ", | |
| "colab_type": "code", | |
| "colab": {} | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "import os\n", | |
| "keras_json='{\\n \"floatx\": \"float32\",\\n \"epsilon\": 1e-07,\\n \"backend\": \"mxnet\",\\n \"image_data_format\": \"channels_first\"\\n}'\n", | |
| "keras_json_dir=os.environ['HOME']+\"/.keras\"\n", | |
| "if not os.path.exists(keras_json_dir): os.mkdir(keras_json_dir)\n", | |
| "with open(keras_json_dir+\"/keras.json\", \"w\") as fp:\n", | |
| " fp.write(keras_json)\n", | |
| " \n" | |
| ], | |
| "execution_count": 0, | |
| "outputs": [] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "hOtU_NudQDJW", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "제데로 반영이 되었는지는 다음의 코드를 실행하는 것으로 확인 가능합니다." | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "2g9W4XJc9GWs", | |
| "colab_type": "code", | |
| "colab": { | |
| "base_uri": "https://localhost:8080/", | |
| "height": 119 | |
| }, | |
| "outputId": "95b05bee-1b40-43c4-c687-5784224f94b6" | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "!cat ~/.keras/keras.json" | |
| ], | |
| "execution_count": 4, | |
| "outputs": [ | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "{\r\n", | |
| " \"floatx\": \"float32\",\r\n", | |
| " \"epsilon\": 1e-07,\r\n", | |
| " \"backend\": \"mxnet\",\r\n", | |
| " \"image_data_format\": \"channels_first\"\r\n", | |
| "}" | |
| ], | |
| "name": "stdout" | |
| } | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "nMgorEiVRJfL", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "## MXNet이 백엔드로 지정되었는지를 확인\n", | |
| "\n", | |
| "import keras를 실행하여 Using MXNet Backend라고 표시되면 OK입니다." | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "DhrU3iOPRW1S", | |
| "colab_type": "code", | |
| "colab": { | |
| "base_uri": "https://localhost:8080/", | |
| "height": 34 | |
| }, | |
| "outputId": "d0067ac6-a6b1-42b3-c4d2-8b152df0a442" | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "import keras" | |
| ], | |
| "execution_count": 5, | |
| "outputs": [ | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "Using MXNet backend\n" | |
| ], | |
| "name": "stderr" | |
| } | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "n-V7QrjbRcIU", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "## 모델 작성시의 주의점\n", | |
| "channels_first이기 때문에 Conv2d나 Pooling과 같은 레이어를 추가할 때엔, data_format과 같이 명시적으로 선언해 줄 필요가 있습니다. 또한, BatchNormalization이라던지 Merge (Add, Concatenate등)를 실시 할 때엔, axis=1 이라고 체널 축을 지정할 필요가 있습니다.\n", | |
| "\n", | |
| "에러 메세지에도 나오지만, channels_last로 하게 되면 너무 무거워지기 때문에 주의가 필요합니다.\n", | |
| "\n", | |
| "## MNIST샘플\n", | |
| "---\n", | |
| "\n", | |
| "간단한 샘플코드로서 AlexNet을 만들어 봤습니다. 데이터는 MNIST입니다.\n", | |
| "\n" | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "ds-tGFlAToAN", | |
| "colab_type": "code", | |
| "colab": { | |
| "base_uri": "https://localhost:8080/", | |
| "height": 819 | |
| }, | |
| "outputId": "db8716c7-3998-442b-dffb-74a3b25088a8" | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "from keras.layers import Conv2D, Activation, MaxPooling2D, BatchNormalization, Input, Flatten, Dense\n", | |
| "from keras.models import Model\n", | |
| "from keras.optimizers import Adam\n", | |
| "from keras.datasets import mnist\n", | |
| "from keras.utils import to_categorical\n", | |
| "import numpy as np\n", | |
| "\n", | |
| "# Data\n", | |
| "(X_train, y_train), (X_test, y_test) = mnist.load_data()\n", | |
| "X_train = np.expand_dims(X_train, axis=1)\n", | |
| "X_test = np.expand_dims(X_test, axis=1)\n", | |
| "X_train = X_train / 255.0\n", | |
| "X_test = X_test / 255.0\n", | |
| "y_train = to_categorical(y_train)\n", | |
| "y_test = to_categorical(y_test)\n", | |
| "\n", | |
| "def create_conv_layers(input, nb_filters, kernel_size):\n", | |
| " x = Conv2D(nb_filters, kernel_size=kernel_size, padding=\"same\", data_format=\"channels_first\")(input)\n", | |
| " x = BatchNormalization(axis=1)(x)\n", | |
| " x = Activation(\"relu\")(x)\n", | |
| " return x\n", | |
| "\n", | |
| "# Model\n", | |
| "input = Input(shape=(1, 28, 28))\n", | |
| "x = create_conv_layers(input, 256, 5)\n", | |
| "x = create_conv_layers(x, 256, 5)\n", | |
| "x = MaxPooling2D((2,2), data_format=\"channels_first\")(x)\n", | |
| "x = create_conv_layers(x, 384, 3)\n", | |
| "x = create_conv_layers(x, 384, 3)\n", | |
| "x = create_conv_layers(x, 384, 3)\n", | |
| "x = Flatten()(x)\n", | |
| "x = Dense(10, activation=\"softmax\")(x)\n", | |
| "\n", | |
| "model = Model(input, x)\n", | |
| "model.compile(Adam(lr=0.0001), loss=\"categorical_crossentropy\", metrics=[\"acc\"])\n", | |
| "model.fit(X_train, y_train, batch_size=128, epochs=20, validation_data=(X_test, y_test))" | |
| ], | |
| "execution_count": 6, | |
| "outputs": [ | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n", | |
| "11493376/11490434 [==============================] - 1s 0us/step\n", | |
| "Train on 60000 samples, validate on 10000 samples\n", | |
| "Epoch 1/20\n" | |
| ], | |
| "name": "stdout" | |
| }, | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "/usr/local/lib/python3.6/dist-packages/mxnet/module/bucketing_module.py:408: UserWarning: Optimizer created manually outside Module but rescale_grad is not normalized to 1.0/batch_size/num_workers (1.0 vs. 0.0078125). Is this intended?\n", | |
| " force_init=force_init)\n" | |
| ], | |
| "name": "stderr" | |
| }, | |
| { | |
| "output_type": "stream", | |
| "text": [ | |
| "60000/60000 [==============================] - 138s 2ms/step - loss: 0.1319 - acc: 0.9628 - val_loss: 0.1098 - val_acc: 0.9716\n", | |
| "Epoch 2/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0522 - acc: 0.9850 - val_loss: 0.0747 - val_acc: 0.9805\n", | |
| "Epoch 3/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0400 - acc: 0.9881 - val_loss: 0.0798 - val_acc: 0.9803\n", | |
| "Epoch 4/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0350 - acc: 0.9897 - val_loss: 0.0415 - val_acc: 0.9898\n", | |
| "Epoch 5/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0319 - acc: 0.9911 - val_loss: 0.0848 - val_acc: 0.9785\n", | |
| "Epoch 6/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0299 - acc: 0.9920 - val_loss: 0.0463 - val_acc: 0.9891\n", | |
| "Epoch 7/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0297 - acc: 0.9924 - val_loss: 0.0826 - val_acc: 0.9815\n", | |
| "Epoch 8/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0213 - acc: 0.9938 - val_loss: 0.0363 - val_acc: 0.9920\n", | |
| "Epoch 9/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0240 - acc: 0.9938 - val_loss: 0.0380 - val_acc: 0.9923\n", | |
| "Epoch 10/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0213 - acc: 0.9942 - val_loss: 0.0831 - val_acc: 0.9841\n", | |
| "Epoch 11/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0247 - acc: 0.9940 - val_loss: 0.0403 - val_acc: 0.9923\n", | |
| "Epoch 12/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0169 - acc: 0.9957 - val_loss: 0.0631 - val_acc: 0.9889\n", | |
| "Epoch 13/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0182 - acc: 0.9955 - val_loss: 0.0635 - val_acc: 0.9885\n", | |
| "Epoch 14/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0193 - acc: 0.9951 - val_loss: 0.0624 - val_acc: 0.9889\n", | |
| "Epoch 15/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0180 - acc: 0.9957 - val_loss: 0.0371 - val_acc: 0.9920\n", | |
| "Epoch 16/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0152 - acc: 0.9961 - val_loss: 0.0626 - val_acc: 0.9900\n", | |
| "Epoch 17/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0182 - acc: 0.9960 - val_loss: 0.0634 - val_acc: 0.9901\n", | |
| "Epoch 18/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0155 - acc: 0.9963 - val_loss: 0.0577 - val_acc: 0.9900\n", | |
| "Epoch 19/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0154 - acc: 0.9964 - val_loss: 0.0803 - val_acc: 0.9871\n", | |
| "Epoch 20/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0112 - acc: 0.9970 - val_loss: 0.0977 - val_acc: 0.9840\n" | |
| ], | |
| "name": "stdout" | |
| }, | |
| { | |
| "output_type": "execute_result", | |
| "data": { | |
| "text/plain": [ | |
| "<keras.callbacks.History at 0x7f79edf6ce10>" | |
| ] | |
| }, | |
| "metadata": { | |
| "tags": [] | |
| }, | |
| "execution_count": 6 | |
| } | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "ltuI7xLpTu9Y", | |
| "colab_type": "text" | |
| }, | |
| "cell_type": "markdown", | |
| "source": [ | |
| "\n", | |
| "결과는 다음과 비슷하게 나올껍니다.\n", | |
| "\n", | |
| "```\n", | |
| "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n", | |
| "11493376/11490434 [==============================] - 1s 0us/step\n", | |
| "Train on 60000 samples, validate on 10000 samples\n", | |
| "Epoch 1/20\n", | |
| "/usr/local/lib/python3.6/dist-packages/mxnet/module/bucketing_module.py:408: UserWarning: Optimizer created manually outside Module but rescale_grad is not normalized to 1.0/batch_size/num_workers (1.0 vs. 0.0078125). Is this intended?\n", | |
| " force_init=force_init)\n", | |
| "60000/60000 [==============================] - 138s 2ms/step - loss: 0.1319 - acc: 0.9628 - val_loss: 0.1098 - val_acc: 0.9716\n", | |
| "Epoch 2/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0522 - acc: 0.9850 - val_loss: 0.0747 - val_acc: 0.9805\n", | |
| "Epoch 3/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0400 - acc: 0.9881 - val_loss: 0.0798 - val_acc: 0.9803\n", | |
| "Epoch 4/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0350 - acc: 0.9897 - val_loss: 0.0415 - val_acc: 0.9898\n", | |
| "Epoch 5/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0319 - acc: 0.9911 - val_loss: 0.0848 - val_acc: 0.9785\n", | |
| "Epoch 6/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0299 - acc: 0.9920 - val_loss: 0.0463 - val_acc: 0.9891\n", | |
| "Epoch 7/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0297 - acc: 0.9924 - val_loss: 0.0826 - val_acc: 0.9815\n", | |
| "Epoch 8/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0213 - acc: 0.9938 - val_loss: 0.0363 - val_acc: 0.9920\n", | |
| "Epoch 9/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0240 - acc: 0.9938 - val_loss: 0.0380 - val_acc: 0.9923\n", | |
| "Epoch 10/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0213 - acc: 0.9942 - val_loss: 0.0831 - val_acc: 0.9841\n", | |
| "Epoch 11/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0247 - acc: 0.9940 - val_loss: 0.0403 - val_acc: 0.9923\n", | |
| "Epoch 12/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0169 - acc: 0.9957 - val_loss: 0.0631 - val_acc: 0.9889\n", | |
| "Epoch 13/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0182 - acc: 0.9955 - val_loss: 0.0635 - val_acc: 0.9885\n", | |
| "Epoch 14/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0193 - acc: 0.9951 - val_loss: 0.0624 - val_acc: 0.9889\n", | |
| "Epoch 15/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0180 - acc: 0.9957 - val_loss: 0.0371 - val_acc: 0.9920\n", | |
| "Epoch 16/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0152 - acc: 0.9961 - val_loss: 0.0626 - val_acc: 0.9900\n", | |
| "Epoch 17/20\n", | |
| "60000/60000 [==============================] - 126s 2ms/step - loss: 0.0182 - acc: 0.9960 - val_loss: 0.0634 - val_acc: 0.9901\n", | |
| "Epoch 18/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0155 - acc: 0.9963 - val_loss: 0.0577 - val_acc: 0.9900\n", | |
| "Epoch 19/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0154 - acc: 0.9964 - val_loss: 0.0803 - val_acc: 0.9871\n", | |
| "Epoch 20/20\n", | |
| "60000/60000 [==============================] - 125s 2ms/step - loss: 0.0112 - acc: 0.9970 - val_loss: 0.0977 - val_acc: 0.9840\n", | |
| "<keras.callbacks.History at 0x7f79edf6ce10>\n", | |
| "```\n", | |
| "\n", | |
| "덧붙여서 이것을 친숙한 TensorFlow 백엔드에서 실행시키면 이렇게 나옵니다.\n", | |
| "\n", | |
| "```\n", | |
| "Epoch 1/20\n", | |
| "60000/60000 [==============================] - 153s 3ms/step - loss: 0.1154 - acc: 0.9648 - val_loss: 0.0774 - val_acc: 0.9795\n", | |
| "```\n", | |
| "\n" | |
| ] | |
| }, | |
| { | |
| "metadata": { | |
| "id": "uxEVmwW0lSY2", | |
| "colab_type": "code", | |
| "colab": {} | |
| }, | |
| "cell_type": "code", | |
| "source": [ | |
| "" | |
| ], | |
| "execution_count": 0, | |
| "outputs": [] | |
| } | |
| ] | |
| } |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment