{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "94874952",
      "metadata": {
        "id": "94874952"
      },
      "source": [
        "# **自定義類別物件**\n",
        "此份程式碼為 Custom_building 的 PyTorch 參考寫法。\n",
        "## 本章節內容大綱\n",
        "* ### [Custom Loss](#Loss)\n",
        "* ### [Custom Layer](#Layer)\n",
        "* ### [Custom Model](#Model)\n",
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "48c7f0ca",
      "metadata": {
        "id": "48c7f0ca"
      },
      "source": [
        "## 匯入套件"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b55f3395",
      "metadata": {
        "id": "b55f3395"
      },
      "outputs": [],
      "source": [
        "!pip install torchsummary"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "aed34e7d",
      "metadata": {
        "id": "aed34e7d"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "\n",
        "import torch\n",
        "from torchsummary import summary"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "5514391e",
      "metadata": {
        "id": "5514391e"
      },
      "outputs": [],
      "source": [
        "y_true = torch.randn((10, 4))\n",
        "y_pred = torch.randn((10, 4))"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "a4a38805",
      "metadata": {
        "id": "a4a38805"
      },
      "source": [
        "## Loss"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "2e2339c6",
      "metadata": {
        "id": "2e2339c6"
      },
      "outputs": [],
      "source": [
        "# build loss by torch.nn module\n",
        "mse_loss = torch.nn.MSELoss()\n",
        "mse_loss(y_pred, y_true)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "55ce5237",
      "metadata": {
        "id": "55ce5237"
      },
      "outputs": [],
      "source": [
        "class my_mse(torch.nn.Module):  # build loss object by custom loss\n",
        "    def __init__(self):\n",
        "        super(my_mse, self).__init__()\n",
        "\n",
        "    def forward(self, outputs, targets):\n",
        "        return torch.mean((outputs - targets)**2)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "04013ceb",
      "metadata": {
        "id": "04013ceb"
      },
      "outputs": [],
      "source": [
        "my_mse_loss = my_mse()\n",
        "my_mse_loss(y_true, y_pred)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "11cc631c",
      "metadata": {
        "id": "11cc631c"
      },
      "source": [
        "## Layer"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "d5468b23",
      "metadata": {
        "id": "d5468b23"
      },
      "outputs": [],
      "source": [
        "x = torch.randn((10, 5))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "8ae95b79",
      "metadata": {
        "id": "8ae95b79"
      },
      "outputs": [],
      "source": [
        "# build layer by torch.nn module\n",
        "linear_layer = torch.nn.Linear(5, 4)\n",
        "linear_layer(x)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b24106d8",
      "metadata": {
        "id": "b24106d8"
      },
      "outputs": [],
      "source": [
        "# use same initial parameters\n",
        "w = linear_layer.weight\n",
        "b = linear_layer.bias"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "66fbba4c",
      "metadata": {
        "id": "66fbba4c"
      },
      "outputs": [],
      "source": [
        "class my_linear(torch.nn.Module):  # build layer object by custom class\n",
        "    def __init__(self, w, b):\n",
        "        super().__init__()\n",
        "        self.w = w\n",
        "        self.b = b\n",
        "\n",
        "    def forward(self, x):\n",
        "        return torch.matmul(x, self.w.transpose(0, 1)) + self.b"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b54ac45c",
      "metadata": {
        "id": "b54ac45c"
      },
      "outputs": [],
      "source": [
        "my_linear_layer = my_linear(w, b)\n",
        "my_linear_layer(x)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "101c3754",
      "metadata": {
        "id": "101c3754"
      },
      "source": [
        "## Model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ee9ecf1d",
      "metadata": {
        "id": "ee9ecf1d"
      },
      "outputs": [],
      "source": [
        "num_classes = 4"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "6cef4001",
      "metadata": {
        "id": "6cef4001"
      },
      "outputs": [],
      "source": [
        "def build_model(input_shape, output_shape):  # build model by torch.nn module\n",
        "    model = torch.nn.Sequential(\n",
        "        torch.nn.Linear(input_shape, 16),\n",
        "        torch.nn.Linear(16, 32),\n",
        "        torch.nn.Linear(32, output_shape))\n",
        "    return model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ca1c3829",
      "metadata": {
        "id": "ca1c3829"
      },
      "outputs": [],
      "source": [
        "model = build_model(5, num_classes)\n",
        "summary(model, x[0].shape, device='cpu')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b0c18bad",
      "metadata": {
        "id": "b0c18bad"
      },
      "outputs": [],
      "source": [
        "class my_net(torch.nn.Module):  # build model object by custom class\n",
        "    def __init__(self, input_shape, output_shape):\n",
        "        super(my_net, self).__init__()\n",
        "        self.hidden_layer_1 = torch.nn.Linear(input_shape, 16)\n",
        "        self.hidden_layer_2 = torch.nn.Linear(16, 32)\n",
        "        self.output_layer = torch.nn.Linear(32, output_shape)\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = self.hidden_layer_1(x)\n",
        "        x = self.hidden_layer_2(x)\n",
        "        outputs = self.output_layer(x)\n",
        "        return outputs"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "c203e0e9",
      "metadata": {
        "id": "c203e0e9"
      },
      "outputs": [],
      "source": [
        "my_model = my_net(5, 4)\n",
        "summary(my_model, x[0].shape, device='cpu')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "2570399d",
      "metadata": {
        "id": "2570399d"
      },
      "outputs": [],
      "source": []
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3 (ipykernel)",
      "language": "python",
      "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": "3.7.12"
    },
    "colab": {
      "provenance": []
    },
    "accelerator": "GPU",
    "gpuClass": "standard"
  },
  "nbformat": 4,
  "nbformat_minor": 5
}