5_4_ML.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/minyoung347/python101/blob/main/5_4_ML.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tPMwQAH7Tk29"
      },
      "source": [
        "### 과소적합과 과대적합"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VfRNGJB8Tk3B"
      },
      "source": [
        "**MNIST에 백색 잡음 픽셀과 0 픽셀을 추가하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:01:24.254476Z",
          "iopub.status.busy": "2022-05-21T01:01:24.253958Z",
          "iopub.status.idle": "2022-05-21T01:01:27.373051Z",
          "shell.execute_reply": "2022-05-21T01:01:27.372074Z"
        },
        "id": "KBz18RQVTk3C",
        "outputId": "c7610568-12e3-49f6-b5e3-320509650b70",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n",
            "11490434/11490434 [==============================] - 2s 0us/step\n"
          ]
        }
      ],
      "source": [
        "from tensorflow.keras.datasets import mnist\n",
        "import numpy as np\n",
        "\n",
        "(train_images, train_labels), _ = mnist.load_data()\n",
        "train_images = train_images.reshape((60000, 28 * 28))\n",
        "train_images = train_images.astype(\"float32\") / 255\n",
        "\n",
        "train_images_with_noise_channels = np.concatenate(\n",
        "    [train_images, np.random.random((len(train_images), 784))], axis=1)\n",
        "\n",
        "train_images_with_zeros_channels = np.concatenate(\n",
        "    [train_images, np.zeros((len(train_images), 784))], axis=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:01:27.377455Z",
          "iopub.status.busy": "2022-05-21T01:01:27.376945Z",
          "iopub.status.idle": "2022-05-21T01:01:27.388938Z",
          "shell.execute_reply": "2022-05-21T01:01:27.387917Z"
        },
        "id": "8cUOncJ1Tk3F",
        "outputId": "743cb223-e93a-44d9-88e1-94637c468e33",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(60000, 1568)"
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ],
      "source": [
        "train_images_with_noise_channels.shape"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Eh9dzVI3Tk3I"
      },
      "source": [
        "**백색 잡음과 0을 추가한 MNIST 데이터에서 모델 훈련하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:01:27.392783Z",
          "iopub.status.busy": "2022-05-21T01:01:27.392444Z",
          "iopub.status.idle": "2022-05-21T01:02:25.502977Z",
          "shell.execute_reply": "2022-05-21T01:02:25.502078Z"
        },
        "id": "H8BUR0xETk3J",
        "outputId": "849f9740-1997-4719-8eb4-fc6a9a38940b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "375/375 [==============================] - 3s 4ms/step - loss: 0.6236 - accuracy: 0.8100 - val_loss: 0.2897 - val_accuracy: 0.9160\n",
            "Epoch 2/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.2606 - accuracy: 0.9189 - val_loss: 0.2095 - val_accuracy: 0.9373\n",
            "Epoch 3/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.1731 - accuracy: 0.9467 - val_loss: 0.1685 - val_accuracy: 0.9488\n",
            "Epoch 4/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.1212 - accuracy: 0.9623 - val_loss: 0.1535 - val_accuracy: 0.9535\n",
            "Epoch 5/10\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 0.0897 - accuracy: 0.9721 - val_loss: 0.1311 - val_accuracy: 0.9597\n",
            "Epoch 6/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.0660 - accuracy: 0.9790 - val_loss: 0.1197 - val_accuracy: 0.9649\n",
            "Epoch 7/10\n",
            "375/375 [==============================] - 3s 7ms/step - loss: 0.0512 - accuracy: 0.9833 - val_loss: 0.1173 - val_accuracy: 0.9654\n",
            "Epoch 8/10\n",
            "375/375 [==============================] - 3s 8ms/step - loss: 0.0357 - accuracy: 0.9885 - val_loss: 0.1183 - val_accuracy: 0.9672\n",
            "Epoch 9/10\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 0.0269 - accuracy: 0.9914 - val_loss: 0.2000 - val_accuracy: 0.9468\n",
            "Epoch 10/10\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 0.0189 - accuracy: 0.9943 - val_loss: 0.1216 - val_accuracy: 0.9699\n",
            "Epoch 1/10\n",
            "375/375 [==============================] - 4s 8ms/step - loss: 0.2990 - accuracy: 0.9131 - val_loss: 0.1825 - val_accuracy: 0.9437\n",
            "Epoch 2/10\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 0.1225 - accuracy: 0.9641 - val_loss: 0.1057 - val_accuracy: 0.9684\n",
            "Epoch 3/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.0795 - accuracy: 0.9767 - val_loss: 0.0965 - val_accuracy: 0.9714\n",
            "Epoch 4/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.0574 - accuracy: 0.9831 - val_loss: 0.0811 - val_accuracy: 0.9753\n",
            "Epoch 5/10\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.0427 - accuracy: 0.9876 - val_loss: 0.0814 - val_accuracy: 0.9754\n",
            "Epoch 6/10\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.0329 - accuracy: 0.9900 - val_loss: 0.0727 - val_accuracy: 0.9793\n",
            "Epoch 7/10\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.0245 - accuracy: 0.9929 - val_loss: 0.0711 - val_accuracy: 0.9774\n",
            "Epoch 8/10\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.0192 - accuracy: 0.9946 - val_loss: 0.0779 - val_accuracy: 0.9790\n",
            "Epoch 9/10\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 0.0140 - accuracy: 0.9963 - val_loss: 0.0754 - val_accuracy: 0.9793\n",
            "Epoch 10/10\n",
            "375/375 [==============================] - 3s 7ms/step - loss: 0.0103 - accuracy: 0.9973 - val_loss: 0.0821 - val_accuracy: 0.9790\n"
          ]
        }
      ],
      "source": [
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "\n",
        "def get_model():\n",
        "    model = keras.Sequential([\n",
        "        layers.Dense(512, activation=\"relu\"),\n",
        "        layers.Dense(10, activation=\"softmax\")\n",
        "    ])\n",
        "    model.compile(optimizer=\"rmsprop\",\n",
        "                  loss=\"sparse_categorical_crossentropy\",\n",
        "                  metrics=[\"accuracy\"])\n",
        "    return model\n",
        "\n",
        "model = get_model()\n",
        "history_noise = model.fit(\n",
        "    train_images_with_noise_channels, train_labels,\n",
        "    epochs=10,\n",
        "    batch_size=128,\n",
        "    validation_split=0.2)\n",
        "\n",
        "model = get_model()\n",
        "history_zeros = model.fit(\n",
        "    train_images_with_zeros_channels, train_labels,\n",
        "    epochs=10,\n",
        "    batch_size=128,\n",
        "    validation_split=0.2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vx3EJaPaTk3K"
      },
      "source": [
        "**검증 정확도 비교 그래프 그리기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:02:25.507068Z",
          "iopub.status.busy": "2022-05-21T01:02:25.506610Z",
          "iopub.status.idle": "2022-05-21T01:02:26.352048Z",
          "shell.execute_reply": "2022-05-21T01:02:26.351089Z"
        },
        "id": "FZnn2Uw4Tk3L",
        "outputId": "eec2635e-b8f0-46e9-c078-8659147a578f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 273
        }
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "val_acc_noise = history_noise.history[\"val_accuracy\"]\n",
        "val_acc_zeros = history_zeros.history[\"val_accuracy\"]\n",
        "epochs = range(1, 11)\n",
        "plt.plot(epochs, val_acc_noise, \"b-\",\n",
        "         label=\"Validation accuracy with noise channels\")\n",
        "plt.plot(epochs, val_acc_zeros, \"b--\",\n",
        "         label=\"Validation accuracy with zeros channels\")\n",
        "plt.title(\"Effect of noise channels on validation accuracy\")\n",
        "plt.xlabel(\"Epochs\")\n",
        "plt.ylabel(\"Accuracy\")\n",
        "plt.legend()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mMxFZdiQTk3O"
      },
      "source": [
        "**랜덤하게 섞은 레이블로 MNIST 모델 훈련하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:02:26.358294Z",
          "iopub.status.busy": "2022-05-21T01:02:26.357787Z",
          "iopub.status.idle": "2022-05-21T01:05:33.171731Z",
          "shell.execute_reply": "2022-05-21T01:05:33.170427Z"
        },
        "id": "8ocMHIMMTk3O",
        "outputId": "91677c26-4409-4415-d1aa-acbc6eb4f630",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 2.3144 - accuracy: 0.1029 - val_loss: 2.3062 - val_accuracy: 0.1022\n",
            "Epoch 2/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.2992 - accuracy: 0.1155 - val_loss: 2.3092 - val_accuracy: 0.1053\n",
            "Epoch 3/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.2913 - accuracy: 0.1257 - val_loss: 2.3173 - val_accuracy: 0.1019\n",
            "Epoch 4/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.2796 - accuracy: 0.1363 - val_loss: 2.3233 - val_accuracy: 0.1044\n",
            "Epoch 5/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.2651 - accuracy: 0.1488 - val_loss: 2.3367 - val_accuracy: 0.1006\n",
            "Epoch 6/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.2455 - accuracy: 0.1637 - val_loss: 2.3585 - val_accuracy: 0.0967\n",
            "Epoch 7/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 2.2225 - accuracy: 0.1797 - val_loss: 2.3658 - val_accuracy: 0.1041\n",
            "Epoch 8/100\n",
            "375/375 [==============================] - 3s 7ms/step - loss: 2.1973 - accuracy: 0.1949 - val_loss: 2.3866 - val_accuracy: 0.1031\n",
            "Epoch 9/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.1693 - accuracy: 0.2101 - val_loss: 2.4050 - val_accuracy: 0.1028\n",
            "Epoch 10/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.1363 - accuracy: 0.2293 - val_loss: 2.4254 - val_accuracy: 0.0988\n",
            "Epoch 11/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.1020 - accuracy: 0.2442 - val_loss: 2.4686 - val_accuracy: 0.0987\n",
            "Epoch 12/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.0669 - accuracy: 0.2590 - val_loss: 2.4766 - val_accuracy: 0.0996\n",
            "Epoch 13/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 2.0296 - accuracy: 0.2777 - val_loss: 2.5092 - val_accuracy: 0.0988\n",
            "Epoch 14/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.9920 - accuracy: 0.2931 - val_loss: 2.5357 - val_accuracy: 0.0960\n",
            "Epoch 15/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.9529 - accuracy: 0.3128 - val_loss: 2.5618 - val_accuracy: 0.0974\n",
            "Epoch 16/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.9131 - accuracy: 0.3289 - val_loss: 2.6033 - val_accuracy: 0.1010\n",
            "Epoch 17/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 1.8744 - accuracy: 0.3430 - val_loss: 2.6344 - val_accuracy: 0.0972\n",
            "Epoch 18/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.8353 - accuracy: 0.3614 - val_loss: 2.6855 - val_accuracy: 0.0996\n",
            "Epoch 19/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 1.7965 - accuracy: 0.3739 - val_loss: 2.7087 - val_accuracy: 0.0990\n",
            "Epoch 20/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.7595 - accuracy: 0.3913 - val_loss: 2.7442 - val_accuracy: 0.1004\n",
            "Epoch 21/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.7224 - accuracy: 0.4033 - val_loss: 2.8015 - val_accuracy: 0.1007\n",
            "Epoch 22/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.6840 - accuracy: 0.4201 - val_loss: 2.8321 - val_accuracy: 0.1047\n",
            "Epoch 23/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.6465 - accuracy: 0.4335 - val_loss: 2.8914 - val_accuracy: 0.0999\n",
            "Epoch 24/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.6106 - accuracy: 0.4448 - val_loss: 2.9487 - val_accuracy: 0.0972\n",
            "Epoch 25/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.5747 - accuracy: 0.4627 - val_loss: 2.9773 - val_accuracy: 0.0993\n",
            "Epoch 26/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.5400 - accuracy: 0.4730 - val_loss: 3.0387 - val_accuracy: 0.0942\n",
            "Epoch 27/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.5077 - accuracy: 0.4856 - val_loss: 3.0738 - val_accuracy: 0.1024\n",
            "Epoch 28/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 1.4729 - accuracy: 0.5007 - val_loss: 3.1564 - val_accuracy: 0.0991\n",
            "Epoch 29/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 1.4404 - accuracy: 0.5114 - val_loss: 3.2144 - val_accuracy: 0.0997\n",
            "Epoch 30/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.4101 - accuracy: 0.5231 - val_loss: 3.2292 - val_accuracy: 0.0997\n",
            "Epoch 31/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.3784 - accuracy: 0.5341 - val_loss: 3.2822 - val_accuracy: 0.1002\n",
            "Epoch 32/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.3492 - accuracy: 0.5462 - val_loss: 3.3291 - val_accuracy: 0.1006\n",
            "Epoch 33/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.3190 - accuracy: 0.5560 - val_loss: 3.3966 - val_accuracy: 0.1018\n",
            "Epoch 34/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.2921 - accuracy: 0.5668 - val_loss: 3.4617 - val_accuracy: 0.0994\n",
            "Epoch 35/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.2627 - accuracy: 0.5771 - val_loss: 3.4972 - val_accuracy: 0.1018\n",
            "Epoch 36/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 1.2359 - accuracy: 0.5857 - val_loss: 3.5758 - val_accuracy: 0.0980\n",
            "Epoch 37/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 1.2098 - accuracy: 0.5940 - val_loss: 3.6352 - val_accuracy: 0.0980\n",
            "Epoch 38/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.1855 - accuracy: 0.6044 - val_loss: 3.6869 - val_accuracy: 0.0967\n",
            "Epoch 39/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.1593 - accuracy: 0.6132 - val_loss: 3.7388 - val_accuracy: 0.0983\n",
            "Epoch 40/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.1345 - accuracy: 0.6199 - val_loss: 3.7988 - val_accuracy: 0.0972\n",
            "Epoch 41/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.1114 - accuracy: 0.6293 - val_loss: 3.8738 - val_accuracy: 0.0992\n",
            "Epoch 42/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.0886 - accuracy: 0.6374 - val_loss: 3.9414 - val_accuracy: 0.1002\n",
            "Epoch 43/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 1.0653 - accuracy: 0.6454 - val_loss: 3.9901 - val_accuracy: 0.1002\n",
            "Epoch 44/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.0425 - accuracy: 0.6536 - val_loss: 4.0531 - val_accuracy: 0.1029\n",
            "Epoch 45/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 1.0214 - accuracy: 0.6620 - val_loss: 4.0856 - val_accuracy: 0.0996\n",
            "Epoch 46/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 1.0014 - accuracy: 0.6679 - val_loss: 4.1766 - val_accuracy: 0.1004\n",
            "Epoch 47/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 0.9799 - accuracy: 0.6793 - val_loss: 4.2291 - val_accuracy: 0.1005\n",
            "Epoch 48/100\n",
            "375/375 [==============================] - 2s 4ms/step - loss: 0.9602 - accuracy: 0.6835 - val_loss: 4.3434 - val_accuracy: 0.0967\n",
            "Epoch 49/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.9395 - accuracy: 0.6914 - val_loss: 4.3403 - val_accuracy: 0.0986\n",
            "Epoch 50/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.9195 - accuracy: 0.6977 - val_loss: 4.4553 - val_accuracy: 0.1012\n",
            "Epoch 51/100\n",
            "375/375 [==============================] - 2s 6ms/step - loss: 0.9019 - accuracy: 0.7039 - val_loss: 4.5187 - val_accuracy: 0.1015\n",
            "Epoch 52/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.8807 - accuracy: 0.7127 - val_loss: 4.5895 - val_accuracy: 0.1005\n",
            "Epoch 53/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.8677 - accuracy: 0.7170 - val_loss: 4.6734 - val_accuracy: 0.0983\n",
            "Epoch 54/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.8490 - accuracy: 0.7221 - val_loss: 4.7303 - val_accuracy: 0.0992\n",
            "Epoch 55/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.8310 - accuracy: 0.7296 - val_loss: 4.7249 - val_accuracy: 0.0987\n",
            "Epoch 56/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.8134 - accuracy: 0.7356 - val_loss: 4.8545 - val_accuracy: 0.0994\n",
            "Epoch 57/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.7970 - accuracy: 0.7387 - val_loss: 4.9254 - val_accuracy: 0.0994\n",
            "Epoch 58/100\n",
            "375/375 [==============================] - 2s 5ms/step - loss: 0.7827 - accuracy: 0.7456 - val_loss: 4.9923 - val_accuracy: 0.1038\n",
            "Epoch 59/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.7673 - accuracy: 0.7513 - val_loss: 5.0831 - val_accuracy: 0.0991\n",
            "Epoch 60/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.7537 - accuracy: 0.7560 - val_loss: 5.1157 - val_accuracy: 0.0999\n",
            "Epoch 61/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.7370 - accuracy: 0.7596 - val_loss: 5.1658 - val_accuracy: 0.0996\n",
            "Epoch 62/100\n",
            "375/375 [==============================] - 1s 3ms/step - loss: 0.7222 - accuracy: 0.7664 - val_loss: 5.3002 - val_accuracy: 0.0995\n",
            "Epoch 63/100\n",
            "375/375 [==============================] - 1s 4ms/step - loss: 0.7091 - accuracy: 0.7706 - val_loss: 5.3626 - val_accuracy: 0.0984\n",
            "Epoch 64/100\n",
            "134/375 [=========>....................] - ETA: 0s - loss: 0.6486 - accuracy: 0.7976"
          ]
        }
      ],
      "source": [
        "(train_images, train_labels), _ = mnist.load_data()\n",
        "train_images = train_images.reshape((60000, 28 * 28))\n",
        "train_images = train_images.astype(\"float32\") / 255\n",
        "\n",
        "random_train_labels = train_labels[:]\n",
        "np.random.shuffle(random_train_labels)\n",
        "\n",
        "model = keras.Sequential([\n",
        "    layers.Dense(512, activation=\"relu\"),\n",
        "    layers.Dense(10, activation=\"softmax\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"sparse_categorical_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "model.fit(train_images, random_train_labels,\n",
        "          epochs=100,\n",
        "          batch_size=128,\n",
        "          validation_split=0.2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PunfvZ7uTk3Q"
      },
      "source": [
        "## 머신 러닝 모델 평가"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pRY-dxWTTk3S"
      },
      "source": [
        "**잘못된 높은 학습률로 MNIST 모델 훈련하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:05:33.177460Z",
          "iopub.status.busy": "2022-05-21T01:05:33.177117Z",
          "iopub.status.idle": "2022-05-21T01:05:52.585061Z",
          "shell.execute_reply": "2022-05-21T01:05:52.584241Z"
        },
        "id": "ZqvcFMPRTk3T"
      },
      "outputs": [],
      "source": [
        "(train_images, train_labels), _ = mnist.load_data()\n",
        "train_images = train_images.reshape((60000, 28 * 28))\n",
        "train_images = train_images.astype(\"float32\") / 255\n",
        "\n",
        "model = keras.Sequential([\n",
        "    layers.Dense(512, activation=\"relu\"),\n",
        "    layers.Dense(10, activation=\"softmax\")\n",
        "])\n",
        "model.compile(optimizer=keras.optimizers.RMSprop(1.),\n",
        "              loss=\"sparse_categorical_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "model.fit(train_images, train_labels,\n",
        "          epochs=10,\n",
        "          batch_size=128,\n",
        "          validation_split=0.2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "77iXsoAYTk3U"
      },
      "source": [
        "**같은 모델을 적절한 학습률로 훈련하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:05:52.588549Z",
          "iopub.status.busy": "2022-05-21T01:05:52.588260Z",
          "iopub.status.idle": "2022-05-21T01:06:11.526008Z",
          "shell.execute_reply": "2022-05-21T01:06:11.525112Z"
        },
        "id": "TQAqlYm0Tk3U"
      },
      "outputs": [],
      "source": [
        "model = keras.Sequential([\n",
        "    layers.Dense(512, activation=\"relu\"),\n",
        "    layers.Dense(10, activation=\"softmax\")\n",
        "])\n",
        "model.compile(optimizer=keras.optimizers.RMSprop(1e-2),\n",
        "              loss=\"sparse_categorical_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "model.fit(train_images, train_labels,\n",
        "          epochs=10,\n",
        "          batch_size=128,\n",
        "          validation_split=0.2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Br1F2jcwTk3V"
      },
      "source": [
        "### 모델 용량 늘리기"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ubvdcv_-Tk3V"
      },
      "source": [
        "**MNIST 데이터를 사용한 간단한 로지스틱 회귀 모델**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:06:11.530131Z",
          "iopub.status.busy": "2022-05-21T01:06:11.529560Z",
          "iopub.status.idle": "2022-05-21T01:06:25.955536Z",
          "shell.execute_reply": "2022-05-21T01:06:25.954654Z"
        },
        "id": "xl5ejzSRTk3W"
      },
      "outputs": [],
      "source": [
        "model = keras.Sequential([layers.Dense(10, activation=\"softmax\")])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"sparse_categorical_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_small_model = model.fit(\n",
        "    train_images, train_labels,\n",
        "    epochs=20,\n",
        "    batch_size=128,\n",
        "    validation_split=0.2)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:06:25.959685Z",
          "iopub.status.busy": "2022-05-21T01:06:25.959399Z",
          "iopub.status.idle": "2022-05-21T01:06:26.277516Z",
          "shell.execute_reply": "2022-05-21T01:06:26.276506Z"
        },
        "id": "-yiFhvT4Tk3W"
      },
      "outputs": [],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "val_loss = history_small_model.history[\"val_loss\"]\n",
        "epochs = range(1, 21)\n",
        "plt.plot(epochs, val_loss, \"b--\",\n",
        "         label=\"Validation loss\")\n",
        "plt.title(\"Effect of insufficient model capacity on validation loss\")\n",
        "plt.xlabel(\"Epochs\")\n",
        "plt.ylabel(\"Loss\")\n",
        "plt.legend()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:06:26.281161Z",
          "iopub.status.busy": "2022-05-21T01:06:26.280863Z",
          "iopub.status.idle": "2022-05-21T01:06:48.235589Z",
          "shell.execute_reply": "2022-05-21T01:06:48.234644Z"
        },
        "id": "TYz3NAseTk3X"
      },
      "outputs": [],
      "source": [
        "model = keras.Sequential([\n",
        "    layers.Dense(96, activation=\"relu\"),\n",
        "    layers.Dense(96, activation=\"relu\"),\n",
        "    layers.Dense(10, activation=\"softmax\"),\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"sparse_categorical_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_large_model = model.fit(\n",
        "    train_images, train_labels,\n",
        "    epochs=20,\n",
        "    batch_size=128,\n",
        "    validation_split=0.2)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "g1lVn3NcTk3Z"
      },
      "source": [
        "#### 네트워크 크기 축소"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j3r8_V04Tk3Z"
      },
      "source": [
        "**원본 모델**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:06:48.240327Z",
          "iopub.status.busy": "2022-05-21T01:06:48.239635Z",
          "iopub.status.idle": "2022-05-21T01:07:01.599335Z",
          "shell.execute_reply": "2022-05-21T01:07:01.598428Z"
        },
        "id": "yZ__XxRqTk3Z"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.datasets import imdb\n",
        "(train_data, train_labels), _ = imdb.load_data(num_words=10000)\n",
        "\n",
        "def vectorize_sequences(sequences, dimension=10000):\n",
        "    results = np.zeros((len(sequences), dimension))\n",
        "    for i, sequence in enumerate(sequences):\n",
        "        results[i, sequence] = 1.\n",
        "    return results\n",
        "train_data = vectorize_sequences(train_data)\n",
        "\n",
        "model = keras.Sequential([\n",
        "    layers.Dense(16, activation=\"relu\"),\n",
        "    layers.Dense(16, activation=\"relu\"),\n",
        "    layers.Dense(1, activation=\"sigmoid\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"binary_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_original = model.fit(train_data, train_labels,\n",
        "                             epochs=20, batch_size=512, validation_split=0.4)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ur2zKeT6Tk3a"
      },
      "source": [
        "**작은 용량의 모델**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:07:01.603485Z",
          "iopub.status.busy": "2022-05-21T01:07:01.603168Z",
          "iopub.status.idle": "2022-05-21T01:07:09.910749Z",
          "shell.execute_reply": "2022-05-21T01:07:09.909893Z"
        },
        "id": "J07S8kRUTk3a"
      },
      "outputs": [],
      "source": [
        "model = keras.Sequential([\n",
        "    layers.Dense(4, activation=\"relu\"),\n",
        "    layers.Dense(4, activation=\"relu\"),\n",
        "    layers.Dense(1, activation=\"sigmoid\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"binary_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_smaller_model = model.fit(\n",
        "    train_data, train_labels,\n",
        "    epochs=20, batch_size=512, validation_split=0.4)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k6040fZeTk3a"
      },
      "source": [
        "**큰 용량의 모델**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:07:09.914619Z",
          "iopub.status.busy": "2022-05-21T01:07:09.914070Z",
          "iopub.status.idle": "2022-05-21T01:07:55.440909Z",
          "shell.execute_reply": "2022-05-21T01:07:55.439855Z"
        },
        "id": "gRyaH_CkTk3b",
        "outputId": "04104850-5449-492a-f5c5-189478e0896a",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "30/30 [==============================] - 3s 57ms/step - loss: 0.5666 - accuracy: 0.7082 - val_loss: 0.3856 - val_accuracy: 0.8372\n",
            "Epoch 2/20\n",
            "30/30 [==============================] - 1s 27ms/step - loss: 0.3223 - accuracy: 0.8647 - val_loss: 0.2869 - val_accuracy: 0.8815\n",
            "Epoch 3/20\n",
            "30/30 [==============================] - 1s 27ms/step - loss: 0.2348 - accuracy: 0.9115 - val_loss: 0.4342 - val_accuracy: 0.8208\n",
            "Epoch 4/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.1858 - accuracy: 0.9289 - val_loss: 0.3126 - val_accuracy: 0.8779\n",
            "Epoch 5/20\n",
            "30/30 [==============================] - 1s 27ms/step - loss: 0.1490 - accuracy: 0.9475 - val_loss: 0.3091 - val_accuracy: 0.8860\n",
            "Epoch 6/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.1114 - accuracy: 0.9591 - val_loss: 0.2866 - val_accuracy: 0.8880\n",
            "Epoch 7/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0811 - accuracy: 0.9743 - val_loss: 0.3305 - val_accuracy: 0.8847\n",
            "Epoch 8/20\n",
            "30/30 [==============================] - 1s 27ms/step - loss: 0.0777 - accuracy: 0.9768 - val_loss: 0.3185 - val_accuracy: 0.8857\n",
            "Epoch 9/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0158 - accuracy: 0.9983 - val_loss: 0.4368 - val_accuracy: 0.8861\n",
            "Epoch 10/20\n",
            "30/30 [==============================] - 1s 43ms/step - loss: 0.0800 - accuracy: 0.9802 - val_loss: 0.3761 - val_accuracy: 0.8844\n",
            "Epoch 11/20\n",
            "30/30 [==============================] - 1s 30ms/step - loss: 0.0066 - accuracy: 0.9997 - val_loss: 0.4818 - val_accuracy: 0.8858\n",
            "Epoch 12/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.0265 - accuracy: 0.9955 - val_loss: 1.0362 - val_accuracy: 0.7885\n",
            "Epoch 13/20\n",
            "30/30 [==============================] - 1s 27ms/step - loss: 0.0144 - accuracy: 0.9961 - val_loss: 0.4863 - val_accuracy: 0.8843\n",
            "Epoch 14/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 0.5754 - val_accuracy: 0.8807\n",
            "Epoch 15/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0018 - accuracy: 0.9999 - val_loss: 2.0367 - val_accuracy: 0.7221\n",
            "Epoch 16/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0778 - accuracy: 0.9832 - val_loss: 0.4950 - val_accuracy: 0.8839\n",
            "Epoch 17/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 0.0014 - accuracy: 0.9999 - val_loss: 0.5766 - val_accuracy: 0.8826\n",
            "Epoch 18/20\n",
            "30/30 [==============================] - 1s 29ms/step - loss: 6.5325e-04 - accuracy: 1.0000 - val_loss: 0.6355 - val_accuracy: 0.8831\n",
            "Epoch 19/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 4.1426e-04 - accuracy: 1.0000 - val_loss: 0.6804 - val_accuracy: 0.8832\n",
            "Epoch 20/20\n",
            "30/30 [==============================] - 1s 28ms/step - loss: 3.0846e-04 - accuracy: 1.0000 - val_loss: 0.7004 - val_accuracy: 0.8819\n"
          ]
        }
      ],
      "source": [
        "model = keras.Sequential([\n",
        "    layers.Dense(512, activation=\"relu\"),\n",
        "    layers.Dense(512, activation=\"relu\"),\n",
        "    layers.Dense(1, activation=\"sigmoid\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"binary_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_larger_model = model.fit(\n",
        "    train_data, train_labels,\n",
        "    epochs=20, batch_size=512, validation_split=0.4)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HkMwsNWiTk3b"
      },
      "source": [
        "#### 가중치 규제 추가하기"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "slmc7NwkTk3b"
      },
      "source": [
        "**모델에 L2 가중치 추가하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:07:55.445676Z",
          "iopub.status.busy": "2022-05-21T01:07:55.444826Z",
          "iopub.status.idle": "2022-05-21T01:08:04.225182Z",
          "shell.execute_reply": "2022-05-21T01:08:04.224432Z"
        },
        "id": "BbKqHmXtTk3b",
        "outputId": "bc955a55-cffe-4398-96d2-5ba7a8585fbe",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "30/30 [==============================] - 3s 53ms/step - loss: 0.6246 - accuracy: 0.7769 - val_loss: 0.5054 - val_accuracy: 0.8563\n",
            "Epoch 2/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.4252 - accuracy: 0.8911 - val_loss: 0.4004 - val_accuracy: 0.8826\n",
            "Epoch 3/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.3399 - accuracy: 0.9137 - val_loss: 0.3832 - val_accuracy: 0.8762\n",
            "Epoch 4/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.3021 - accuracy: 0.9217 - val_loss: 0.3547 - val_accuracy: 0.8885\n",
            "Epoch 5/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.2785 - accuracy: 0.9289 - val_loss: 0.3531 - val_accuracy: 0.8867\n",
            "Epoch 6/20\n",
            "30/30 [==============================] - 1s 24ms/step - loss: 0.2564 - accuracy: 0.9413 - val_loss: 0.3512 - val_accuracy: 0.8861\n",
            "Epoch 7/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.2477 - accuracy: 0.9426 - val_loss: 0.3678 - val_accuracy: 0.8818\n",
            "Epoch 8/20\n",
            "30/30 [==============================] - 1s 23ms/step - loss: 0.2418 - accuracy: 0.9453 - val_loss: 0.3708 - val_accuracy: 0.8818\n",
            "Epoch 9/20\n",
            "30/30 [==============================] - 1s 23ms/step - loss: 0.2278 - accuracy: 0.9528 - val_loss: 0.3656 - val_accuracy: 0.8844\n",
            "Epoch 10/20\n",
            "30/30 [==============================] - 1s 24ms/step - loss: 0.2266 - accuracy: 0.9521 - val_loss: 0.3704 - val_accuracy: 0.8825\n",
            "Epoch 11/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.2213 - accuracy: 0.9504 - val_loss: 0.3794 - val_accuracy: 0.8786\n",
            "Epoch 12/20\n",
            "30/30 [==============================] - 1s 23ms/step - loss: 0.2155 - accuracy: 0.9556 - val_loss: 0.3920 - val_accuracy: 0.8761\n",
            "Epoch 13/20\n",
            "30/30 [==============================] - 1s 26ms/step - loss: 0.2082 - accuracy: 0.9592 - val_loss: 0.3858 - val_accuracy: 0.8774\n",
            "Epoch 14/20\n",
            "30/30 [==============================] - 1s 30ms/step - loss: 0.2046 - accuracy: 0.9621 - val_loss: 0.3908 - val_accuracy: 0.8776\n",
            "Epoch 15/20\n",
            "30/30 [==============================] - 1s 30ms/step - loss: 0.2050 - accuracy: 0.9599 - val_loss: 0.3908 - val_accuracy: 0.8789\n",
            "Epoch 16/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.1951 - accuracy: 0.9666 - val_loss: 0.5730 - val_accuracy: 0.8187\n",
            "Epoch 17/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.2026 - accuracy: 0.9610 - val_loss: 0.4214 - val_accuracy: 0.8726\n",
            "Epoch 18/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1877 - accuracy: 0.9681 - val_loss: 0.4199 - val_accuracy: 0.8719\n",
            "Epoch 19/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1981 - accuracy: 0.9601 - val_loss: 0.4073 - val_accuracy: 0.8779\n",
            "Epoch 20/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1848 - accuracy: 0.9692 - val_loss: 0.4092 - val_accuracy: 0.8771\n"
          ]
        }
      ],
      "source": [
        "from tensorflow.keras import regularizers\n",
        "model = keras.Sequential([\n",
        "    layers.Dense(16,\n",
        "                 kernel_regularizer=regularizers.l2(0.002),\n",
        "                 activation=\"relu\"),\n",
        "    layers.Dense(16,\n",
        "                 kernel_regularizer=regularizers.l2(0.002),\n",
        "                 activation=\"relu\"),\n",
        "    layers.Dense(1, activation=\"sigmoid\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"binary_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_l2_reg = model.fit(\n",
        "    train_data, train_labels,\n",
        "    epochs=20, batch_size=512, validation_split=0.4)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "766AbsD9Tk3c"
      },
      "source": [
        "**케라스에서 사용할 수 있는 가중치 규제**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:08:04.229728Z",
          "iopub.status.busy": "2022-05-21T01:08:04.228866Z",
          "iopub.status.idle": "2022-05-21T01:08:04.237099Z",
          "shell.execute_reply": "2022-05-21T01:08:04.235990Z"
        },
        "id": "tT6wJynITk3c",
        "outputId": "605ab40c-8614-40ed-ba60-b5ccb0328795",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.regularizers.L1L2 at 0x7f971c3175b0>"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ],
      "source": [
        "from tensorflow.keras import regularizers\n",
        "regularizers.l1(0.001)\n",
        "regularizers.l1_l2(l1=0.001, l2=0.001)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ultTwtPKTk3c"
      },
      "source": [
        "#### 드롭아웃 추가"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MWW1i75kTk3c"
      },
      "source": [
        "**IMDB 모델에 드롭아웃 추가하기**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "execution": {
          "iopub.execute_input": "2022-05-21T01:08:04.241161Z",
          "iopub.status.busy": "2022-05-21T01:08:04.240762Z",
          "iopub.status.idle": "2022-05-21T01:08:13.113623Z",
          "shell.execute_reply": "2022-05-21T01:08:13.112554Z"
        },
        "id": "K6C938cRTk3d",
        "outputId": "9d1c3e9e-201d-4f5c-e080-4b0e2b3368c8",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "30/30 [==============================] - 3s 53ms/step - loss: 0.6489 - accuracy: 0.6125 - val_loss: 0.5486 - val_accuracy: 0.8080\n",
            "Epoch 2/20\n",
            "30/30 [==============================] - 1s 24ms/step - loss: 0.5327 - accuracy: 0.7443 - val_loss: 0.4282 - val_accuracy: 0.8618\n",
            "Epoch 3/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.4510 - accuracy: 0.8065 - val_loss: 0.3572 - val_accuracy: 0.8783\n",
            "Epoch 4/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.3877 - accuracy: 0.8421 - val_loss: 0.3425 - val_accuracy: 0.8644\n",
            "Epoch 5/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.3459 - accuracy: 0.8671 - val_loss: 0.2915 - val_accuracy: 0.8861\n",
            "Epoch 6/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.2980 - accuracy: 0.8939 - val_loss: 0.2798 - val_accuracy: 0.8876\n",
            "Epoch 7/20\n",
            "30/30 [==============================] - 1s 26ms/step - loss: 0.2696 - accuracy: 0.9057 - val_loss: 0.2742 - val_accuracy: 0.8924\n",
            "Epoch 8/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.2365 - accuracy: 0.9215 - val_loss: 0.2784 - val_accuracy: 0.8912\n",
            "Epoch 9/20\n",
            "30/30 [==============================] - 1s 21ms/step - loss: 0.2165 - accuracy: 0.9294 - val_loss: 0.3048 - val_accuracy: 0.8854\n",
            "Epoch 10/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1974 - accuracy: 0.9363 - val_loss: 0.2951 - val_accuracy: 0.8911\n",
            "Epoch 11/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1777 - accuracy: 0.9467 - val_loss: 0.3120 - val_accuracy: 0.8894\n",
            "Epoch 12/20\n",
            "30/30 [==============================] - 1s 24ms/step - loss: 0.1631 - accuracy: 0.9504 - val_loss: 0.3226 - val_accuracy: 0.8871\n",
            "Epoch 13/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1428 - accuracy: 0.9557 - val_loss: 0.3598 - val_accuracy: 0.8851\n",
            "Epoch 14/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1310 - accuracy: 0.9624 - val_loss: 0.3692 - val_accuracy: 0.8863\n",
            "Epoch 15/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1210 - accuracy: 0.9638 - val_loss: 0.4084 - val_accuracy: 0.8845\n",
            "Epoch 16/20\n",
            "30/30 [==============================] - 1s 22ms/step - loss: 0.1106 - accuracy: 0.9652 - val_loss: 0.4011 - val_accuracy: 0.8842\n",
            "Epoch 17/20\n",
            "30/30 [==============================] - 1s 24ms/step - loss: 0.1049 - accuracy: 0.9667 - val_loss: 0.4394 - val_accuracy: 0.8814\n",
            "Epoch 18/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.0936 - accuracy: 0.9708 - val_loss: 0.4647 - val_accuracy: 0.8834\n",
            "Epoch 19/20\n",
            "30/30 [==============================] - 1s 25ms/step - loss: 0.0924 - accuracy: 0.9725 - val_loss: 0.4865 - val_accuracy: 0.8836\n",
            "Epoch 20/20\n",
            "30/30 [==============================] - 1s 23ms/step - loss: 0.0882 - accuracy: 0.9734 - val_loss: 0.5038 - val_accuracy: 0.8826\n"
          ]
        }
      ],
      "source": [
        "model = keras.Sequential([\n",
        "    layers.Dense(16, activation=\"relu\"),\n",
        "    layers.Dropout(0.5),\n",
        "    layers.Dense(16, activation=\"relu\"),\n",
        "    layers.Dropout(0.5),\n",
        "    layers.Dense(1, activation=\"sigmoid\")\n",
        "])\n",
        "model.compile(optimizer=\"rmsprop\",\n",
        "              loss=\"binary_crossentropy\",\n",
        "              metrics=[\"accuracy\"])\n",
        "history_dropout = model.fit(\n",
        "    train_data, train_labels,\n",
        "    epochs=20, batch_size=512, validation_split=0.4)"
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "zICezbMiBXQ8"
      },
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "colab": {
      "name": "chapter05_fundamentals-of-ml.i",
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "default:Python",
      "language": "python",
      "name": "conda-env-default-py"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.10"
    },
    "accelerator": "GPU",
    "gpuClass": "standard"
  },
  "nbformat": 4,
  "nbformat_minor": 0
}