pytorch_gradient_descent.ipynb

pytorch_gradient_descent.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "pytorch_gradient_descent.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/suzusuzu/17471b63cb112988dc953b201ddd9fc5/pytorch_gradient_descent.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RdGgtliP4I8Y",
        "colab_type": "code",
        "outputId": "31b05d3e-8bc8-47b1-ee66-f38376bb25ae",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "! pip install -U -q torch\n",
        "\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.optim as optim\n",
        "\n",
        "# 評価関数\n",
        "def _sphere(x):\n",
        "    return torch.pow(x, 2).sum()\n",
        "\n",
        "class Model(nn.Module):\n",
        "    def __init__(self, dim=100, func=None):\n",
        "        super(Model, self).__init__()\n",
        "        if func is None:\n",
        "            func = _sphere\n",
        "        self.func = func\n",
        "        # 設計変数\n",
        "        self.x = nn.Parameter(torch.rand(dim))\n",
        "\n",
        "    def forward(self):\n",
        "        return self.func(self.x)\n",
        "    \n",
        "    def vars(self):\n",
        "        return self.x.detach().numpy()\n",
        "\n",
        "    def objective(self):\n",
        "        with torch.no_grad():\n",
        "            return self.func(self.x).numpy()\n",
        "\n",
        "# 次元は10000000次元\n",
        "model = Model(dim=10000000)\n",
        "optimizer = optim.SGD(model.parameters(), lr=0.1)\n",
        "\n",
        "N = 100\n",
        "\n",
        "print('初期の評価値', model.objective())\n",
        "\n",
        "print('optimization...')\n",
        "for i in range(N):\n",
        "    output = model()\n",
        "    optimizer.zero_grad()\n",
        "    loss = output\n",
        "    loss.backward()\n",
        "    optimizer.step()\n",
        "    print('\\rloss:', loss.item(), end='')\n",
        "print()\n",
        "\n",
        "print('評価値', model.objective())"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "初期の評価値 3333505.5\n",
            "optimization...\n",
            "loss: 2.1613312810482566e-13\n",
            "評価値 1.383244e-13\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}