benchmark.ipynb

benchmark.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "benchmark.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/mpdroid/1fe417aa5168a8d88318681f7ee13078/benchmark.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HXJNifgNQ2yD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Original at https://github.com/fastai/course-v3/blob/master/nbs/dl1/lesson7-resnet-mnist.ipynb\n",
        "\n",
        "RUN_LOC = \"COLAB\"\n",
        "# LOCAL, COLAB or KAGGLE\n",
        "# Additional installs needed for Colab and Kaggle\n",
        "\n",
        "WANDB_ID = \"__use_your_own__\"\n",
        "\n",
        "from fastai.vision import *\n",
        "import os\n",
        "if RUN_LOC == \"COLAB\" or RUN_LOC == \"KAGGLE\" :\n",
        "    os.system('pip install --upgrade wandb')\n",
        "\n",
        "import wandb\n",
        "from wandb.fastai import WandbCallback\n",
        "import subprocess\n",
        "os.system('wandb login ' + WANDB_ID)\n",
        "wandb.init(project=\"benchmarks\")\n",
        "\n",
        "##############################\n",
        "# Print system configuration\n",
        "##############################\n",
        "import fastai.utils.collect_env\n",
        "fastai.utils.collect_env.show_install(1)\n",
        "\n",
        "import subprocess\n",
        "\n",
        "def run(command):\n",
        "    process = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE)\n",
        "    out, err = process.communicate()\n",
        "    print(out.decode('utf-8').strip())\n",
        "\n",
        "print('# CPU')\n",
        "run('cat /proc/cpuinfo | egrep -m 1 \"^model name\"')\n",
        "run('cat /proc/cpuinfo | egrep -m 1 \"^cpu MHz\"')\n",
        "run('cat /proc/cpuinfo | egrep -m 1 \"^cpu cores\"')\n",
        "\n",
        "print('# RAM')\n",
        "run('cat /proc/meminfo | egrep \"^MemTotal\"')\n",
        "\n",
        "\n",
        "##############\n",
        "# MODEL\n",
        "##############\n",
        "def conv2(ni,nf): return conv_layer(ni,nf,stride=2)\n",
        "\n",
        "def conv_and_res(ni,nf): return nn.Sequential(conv2(ni, nf), res_block(nf))\n",
        "\n",
        "model = nn.Sequential(\n",
        "    conv_and_res(1, 8),\n",
        "    conv_and_res(8, 16),\n",
        "    conv_and_res(16, 32),\n",
        "    conv_and_res(32, 16),\n",
        "    conv2(16, 10),\n",
        "    Flatten()\n",
        ")\n",
        "\n",
        "##############\n",
        "# TRAINING\n",
        "##############\n",
        "path = untar_data(URLs.MNIST)\n",
        "image_list = ImageList.from_folder(path, convert_mode='L')\n",
        "defaults.cmap='binary'\n",
        "sample_data = image_list.split_by_folder(train='training', valid='testing')\n",
        "label_list = sample_data.label_from_folder()\n",
        "x,y = label_list.train[0]\n",
        "xforms = ([*rand_pad(padding=3, size=28, mode='zeros')], [])\n",
        "label_list = label_list.transform(xforms)\n",
        "batch_size = 128\n",
        "data = label_list.databunch(bs=batch_size).normalize()\n",
        "\n",
        "learn = Learner(data, model, loss_func=nn.CrossEntropyLoss(), metrics=accuracy,  callback_fns=WandbCallback)\n",
        "print(learn.summary())\n",
        "\n",
        "# learn.lr_find(end_lr=100)\n",
        "# learn.recorder.plot()\n",
        "\n",
        "learn.fit_one_cycle(12, max_lr=0.05)"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}