material/deep_learning/dnn_in_keras_weight_init.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": " dnn_in_keras_weight_init.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Mte4_QlwwrlF"
      },
      "source": [
        "# 모듈 임포팅"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2YegBF6qExut"
      },
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "\n",
        "import time"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HNGWUXDmv2nF"
      },
      "source": [
        "from tensorflow.keras import optimizers\n",
        "from tensorflow.keras.layers import Dense"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jJBqaE5oyaIi"
      },
      "source": [
        "# 데이터 준비\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VT4QUNhPmIej"
      },
      "source": [
        "def get_sin_data(start=0, end=10, step=0.1):\n",
        "  x = np.arange(start,end,step)\n",
        "  np.random.shuffle(x)\n",
        "  y = np.sin(x)\n",
        "\n",
        "  split_index = int(x.shape[0]*0.6)\n",
        "\n",
        "  train_x, test_x = x[:split_index], x[split_index:]\n",
        "  train_y, test_y = y[:split_index], y[split_index:]\n",
        "\n",
        "  return (train_x, train_y), (test_x, test_y)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SBsPIRBRV9hO"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dIFRkvcp7Bpn"
      },
      "source": [
        "# 웨이트 초기값\n",
        "\n",
        "Dense() 생성시에 kernel_initializer, bias_initializer로 설정한다.\n",
        "\n",
        "- 'he_normal'\n",
        "- 'lecun_normal'"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9ANsuHxv5-Fd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 654
        },
        "outputId": "260a414a-8a8a-4acf-b416-74e6aaed6346"
      },
      "source": [
        "(train_x, train_y), (test_x, test_y) = get_sin_data(start=0, end=10, step=0.1)\n",
        "\n",
        "model = keras.Sequential()\n",
        "# model.add(Dense(10, activation='tanh', input_shape=(1,)))\n",
        "# model.add(Dense(10, activation='tanh'))\n",
        "model.add(Dense(10, activation='tanh', input_shape=(1,), kernel_initializer='he_normal'))\n",
        "model.add(Dense(10, activation='tanh', kernel_initializer='he_normal'))\n",
        "model.add(Dense(1))\n",
        "\n",
        "\n",
        "model.compile(optimizer=\"SGD\", loss=\"mse\", metrics=[\"mse\"])\n",
        "model.summary()\n",
        "\n",
        "\n",
        "start_time = time.time()\n",
        "model.fit(train_x, train_y, epochs=1000, verbose=0, batch_size=20)\n",
        "print(\"elapsed : {}\".format(time.time() - start_time))    \n",
        "\n",
        "\n",
        "loss, mse = model.evaluate(test_x, test_y)\n",
        "print(\"loss=\",loss)\n",
        "print(\"mse=\",mse)\n",
        "\n",
        "\n",
        "y_ = model.predict(test_x)\n",
        "\n",
        "plt.scatter(test_x,test_y)\n",
        "plt.scatter(test_x,y_,color='r')\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /tensorflow-1.15.2/python3.7/tensorflow_core/python/ops/resource_variable_ops.py:1630: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "If using Keras pass *_constraint arguments to layers.\n",
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense (Dense)                (None, 10)                20        \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 10)                110       \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 11        \n",
            "=================================================================\n",
            "Total params: 141\n",
            "Trainable params: 141\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "elapsed : 7.034062385559082\n",
            "40/40 [==============================] - 0s 486us/sample - loss: 0.1343 - mean_squared_error: 0.1343\n",
            "loss= 0.13432020843029022\n",
            "mse= 0.13432021\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}