{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "999d65a0",
      "metadata": {
        "id": "999d65a0"
      },
      "source": [
        "# **自定義資料集（PyTorch）**\n",
        "此份程式碼為 Custom_dataset 的 PyTorch 參考寫法。\n",
        "\n",
        "## 本章節內容大綱\n",
        "* ### [torch.utils.data.Dataset](#Dataset)\n",
        "* ### [torch.utils.data.DataLoader (Dataset operation)](#DataLoader)\n",
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "3988fa89",
      "metadata": {
        "id": "3988fa89"
      },
      "source": [
        "## 匯入套件"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "a1ae3118",
      "metadata": {
        "id": "a1ae3118"
      },
      "outputs": [],
      "source": [
        "import torch"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "2c697117",
      "metadata": {
        "id": "2c697117"
      },
      "source": [
        "## torch.utils.data.Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "879fea0f",
      "metadata": {
        "id": "879fea0f"
      },
      "outputs": [],
      "source": [
        "value = torch.randn((5, 2))\n",
        "target = torch.randint(high=5, size=(5,), dtype=torch.int64)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "f40f303c",
      "metadata": {
        "id": "f40f303c"
      },
      "source": [
        "* ### TensorDataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "7b78c316",
      "metadata": {
        "id": "7b78c316"
      },
      "outputs": [],
      "source": [
        "# 等同於 tf.data.Dataset.from_tensor_slices\n",
        "dataset1 = torch.utils.data.TensorDataset(value, target)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "8a1d1dae",
      "metadata": {
        "id": "8a1d1dae"
      },
      "outputs": [],
      "source": [
        "dataset1"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "61319313",
      "metadata": {
        "id": "61319313"
      },
      "outputs": [],
      "source": [
        "it = iter(dataset1)\n",
        "print(next(it))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "76705383",
      "metadata": {
        "id": "76705383"
      },
      "outputs": [],
      "source": [
        "for idx, elem in enumerate(dataset1):\n",
        "    print(f'{idx}. {elem}')"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "6fc54fc0",
      "metadata": {
        "id": "6fc54fc0"
      },
      "source": [
        "* ### Custom Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "248b8f8c",
      "metadata": {
        "id": "248b8f8c"
      },
      "outputs": [],
      "source": [
        "class myDataset(torch.utils.data.Dataset):  # build dataset object by custom class\n",
        "    def __init__(self, X, y):\n",
        "        # --------------------------------------------\n",
        "        # Initialize paths, transforms, and so on\n",
        "        # --------------------------------------------\n",
        "        self.X = X\n",
        "        self.y = y\n",
        "        \n",
        "    def __getitem__(self, index):\n",
        "        # --------------------------------------------\n",
        "        # 1. Read from file (using numpy.fromfile, PIL.Image.open)\n",
        "        # 2. Preprocess the data (torchvision.Transform).\n",
        "        # 3. Return the data (e.g. image and label)\n",
        "        # --------------------------------------------\n",
        "        return self.X[index], self.y[index]\n",
        "        \n",
        "    def __len__(self):\n",
        "        # --------------------------------------------\n",
        "        # Indicate the total size of the dataset\n",
        "        # --------------------------------------------\n",
        "        return len(self.X)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "22c8d53f",
      "metadata": {
        "id": "22c8d53f"
      },
      "outputs": [],
      "source": [
        "dataset2 = myDataset(value, target)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "98729db5",
      "metadata": {
        "id": "98729db5"
      },
      "outputs": [],
      "source": [
        "it = iter(dataset2)\n",
        "print(next(it))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "cbb5a93b",
      "metadata": {
        "id": "cbb5a93b"
      },
      "outputs": [],
      "source": [
        "for idx, elem in enumerate(dataset2):\n",
        "    print(f'{idx}. {elem}')"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "aeb62475",
      "metadata": {
        "id": "aeb62475"
      },
      "source": [
        "## torch.utils.data.DataLoader"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "de7d6776",
      "metadata": {
        "id": "de7d6776"
      },
      "outputs": [],
      "source": [
        "dataset1_loader = torch.utils.data.DataLoader(dataset1, batch_size=2, shuffle=True)\n",
        "next(iter(dataset1_loader))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "0523babc",
      "metadata": {
        "id": "0523babc"
      },
      "outputs": [],
      "source": [
        "dataset2_loader = torch.utils.data.DataLoader(dataset2, batch_size=2, shuffle=True)\n",
        "next(iter(dataset2_loader))"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "59a0ea47",
      "metadata": {
        "id": "59a0ea47"
      },
      "source": [
        "參考資料：https://pytorch.org/docs/stable/data.html"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "38a6e573",
      "metadata": {
        "id": "38a6e573"
      },
      "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
}