2.4.1 神经网络分类任务 MNIST.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# torch.nn.functional中有很多功能，后续会常用。那么什么是用nn.Module，什么时候用nn.functional呢\n",
    "# 一般情况下，如果 模型有可学习的参数，最好用nn.Module，其他情况用nn.functional更简单一些\n",
    "# MSB:可以认为是大端存储\n",
    "# python中的...省略号,表示索引操作时省略冒号 a[:,:,1] = a[...,1]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "'''\n",
    "TRAINING SET LABEL FILE (train-labels-idx1-ubyte):\n",
    "[offset] [type]          [value]          [description]\n",
    "0000     32 bit integer  0x00000801(2049) magic number (MSB first)\n",
    "0004     32 bit integer  60000            number of items\n",
    "0008     unsigned byte   ??               label\n",
    "0009     unsigned byte   ??               label\n",
    "........\n",
    "xxxx     unsigned byte   ??               label\n",
    "The labels values are 0 to 9.\n",
    "\n",
    "TRAINING SET IMAGE FILE (train-images-idx3-ubyte):\n",
    "[offset] [type]          [value]          [description]\n",
    "0000     32 bit integer  0x00000803(2051) magic number\n",
    "0004     32 bit integer  60000            number of images\n",
    "0008     32 bit integer  28               number of rows\n",
    "0012     32 bit integer  28               number of columns\n",
    "0016     unsigned byte   ??               pixel\n",
    "0017     unsigned byte   ??               pixel\n",
    "........\n",
    "xxxx     unsigned byte   ??               pixel\n",
    "Pixels are organized row-wise. Pixel values are 0 to 255. 0 means background (white), 255 means foreground (black).\n",
    "\n",
    "TEST SET LABEL FILE (t10k-labels-idx1-ubyte):\n",
    "[offset] [type]          [value]          [description]\n",
    "0000     32 bit integer  0x00000801(2049) magic number (MSB first)\n",
    "0004     32 bit integer  10000            number of items\n",
    "0008     unsigned byte   ??               label\n",
    "0009     unsigned byte   ??               label\n",
    "........\n",
    "xxxx     unsigned byte   ??               label\n",
    "The labels values are 0 to 9.\n",
    "\n",
    "TEST SET IMAGE FILE (t10k-images-idx3-ubyte):\n",
    "[offset] [type]          [value]          [description]\n",
    "0000     32 bit integer  0x00000803(2051) magic number\n",
    "0004     32 bit integer  10000            number of images\n",
    "0008     32 bit integer  28               number of rows\n",
    "0012     32 bit integer  28               number of columns\n",
    "0016     unsigned byte   ??               pixel\n",
    "0017     unsigned byte   ??               pixel\n",
    "........\n",
    "xxxx     unsigned byte   ??               pixel\n",
    "Pixels are organized row-wise. Pixel values are 0 to 255. 0 means background (white), 255 means foreground (black).\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(28, 28)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 12 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import gzip\n",
    "import struct\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# data_name = 'd:/GitRepos/data/{}/train-images-idx3-ubyte.gz'.format('MNIST')\n",
    "\n",
    "mnist = {\n",
    "    'trainset'  :'d:/GitRepos/data/MNIST/train-images-idx3-ubyte.gz',\n",
    "    'trainlabel':'d:/GitRepos/data/MNIST/train-labels-idx1-ubyte.gz',\n",
    "    'testset'   :'d:/GitRepos/data/MNIST/t10k-images-idx3-ubyte.gz',\n",
    "    'testlabel' :'d:/GitRepos/data/MNIST/t10k-labels-idx1-ubyte.gz'\n",
    "}\n",
    "fashion_mnist = {\n",
    "    'trainset'  :'d:/GitRepos/data/fashionMNIST/train-images-idx3-ubyte.gz',\n",
    "    'trainlabel':'d:/GitRepos/data/fashionMNIST/train-labels-idx1-ubyte.gz',\n",
    "    'testset'   :'d:/GitRepos/data/fashionMNIST/t10k-images-idx3-ubyte.gz',\n",
    "    'testlabel' :'d:/GitRepos/data/fashionMNIST/t10k-labels-idx1-ubyte.gz'\n",
    "}\n",
    "\n",
    "image_header = {\n",
    "    'magic' :0,\n",
    "    'cnt'   :0,\n",
    "    'rows'  :0,\n",
    "    'cols'  :0\n",
    "}\n",
    "label_header = {\n",
    "    'magic'  :0,\n",
    "    'itemNum':0\n",
    "}\n",
    "\n",
    "def get_header_idx(path : str,header_fmt : str):\n",
    "    offset = 0\n",
    "    # fmt = '>{}i'.format(head_size)\n",
    "    with gzip.open(path) as f:\n",
    "        bytescontent = f.read()\n",
    "        h_list = [x for x in struct.unpack_from(header_fmt,bytescontent,offset=offset)]\n",
    "        keys   = [k for k in image_header.keys()]\n",
    "        header = dict(zip(keys,h_list))\n",
    "    return header\n",
    "\n",
    "def image_loader_idx3(path : str,header_fmt : str,header : dict,yield_batch_idx : int,yield_batch_size : int):\n",
    "    '''\n",
    "    offset : starting position in byte\n",
    "    format : asigning number of bytes to read\n",
    "    '''\n",
    "    # \n",
    "    assert yield_batch_idx  > 0\n",
    "    assert yield_batch_size > 0\n",
    "    # header offset,use struct.calcsize() to get the correct offset caused by aligning\n",
    "    offset_head = struct.calcsize(header_fmt)\n",
    "    size_image = header['rows'] * header['cols']\n",
    "    # batch offset\n",
    "    fmt_batch = '{}B'.format((yield_batch_idx-1) * yield_batch_size * size_image)\n",
    "    offset_batch = struct.calcsize(fmt_batch)\n",
    "    # single image offset\n",
    "    fmt_image = '>{}B'.format(size_image)\n",
    "    with gzip.open(path) as f:\n",
    "        bindata = f.read()\n",
    "        images_list = (\n",
    "            np.array(\n",
    "            struct.unpack_from(\n",
    "                fmt_image,\n",
    "                bindata,\n",
    "                offset= struct.calcsize(fmt_image) * i + offset_head + offset_batch\n",
    "                )\n",
    "            ).reshape(header['rows'],header['cols'])\n",
    "            for i in range(yield_batch_size)\n",
    "        )\n",
    "    return images_list\n",
    "\n",
    "def label_loader_idx1(path : str,header_fmt : str,yield_batch_idx : int,yield_batch_size : int):\n",
    "    #\n",
    "    assert yield_batch_idx  > 0\n",
    "    assert yield_batch_size > 0\n",
    "    # head offset\n",
    "    offset_head = struct.calcsize(header_fmt)\n",
    "    # batch offset\n",
    "    fmt_batch = '{}B'.format((yield_batch_idx - 1) * yield_batch_size * 1)\n",
    "    offset_batch = struct.calcsize(fmt_batch)\n",
    "    fmt = '>B'\n",
    "    with gzip.open(path) as f:\n",
    "        bindata = f.read()\n",
    "        labels_list =(\n",
    "            struct.unpack_from(\n",
    "            fmt,\n",
    "            bindata,\n",
    "            offset=offset_head + struct.calcsize(fmt) * i + offset_batch\n",
    "            )\n",
    "            for i in range(yield_batch_size)\n",
    "        )\n",
    "    return labels_list\n",
    "\n",
    "def compress_as_gzip(inpath : str,outpath : str):\n",
    "    '''\n",
    "    open a non-GZip file, compress and save with extension .gz\n",
    "    '''\n",
    "    with open(inpath,'rb') as f:\n",
    "        bindata = f.read()\n",
    "        gobj = gzip.GzipFile(\n",
    "            filename=inpath,\n",
    "            mode='wb',\n",
    "            compresslevel=9,\n",
    "            fileobj=open(outpath,mode='wb')\n",
    "        )\n",
    "        gobj.write(bindata)\n",
    "\n",
    "# asigning data path\n",
    "m_data = mnist\n",
    "image_header = get_header_idx(path=m_data['trainset'],header_fmt='>4i')\n",
    "label_header = get_header_idx(path=m_data['trainlabel'],header_fmt='>2i' )\n",
    "\n",
    "# asigning batch position and size\n",
    "yield_batch_idx = 50\n",
    "yield_batch_size = 100\n",
    "\n",
    "# form a new header\n",
    "header_fmt_image = '>4i'\n",
    "header_fmt_label = '>2i'\n",
    "\n",
    "# load data\n",
    "train_image_loader = image_loader_idx3(path=m_data['trainset'],header_fmt=header_fmt_image,header=image_header,yield_batch_idx=yield_batch_idx,yield_batch_size=100)\n",
    "train_label_loader = label_loader_idx1(path=m_data['trainlabel'],header_fmt=header_fmt_label,yield_batch_idx=yield_batch_idx,yield_batch_size=100)\n",
    "\n",
    "# https://matplotlib.org/stable/gallery/style_sheets/style_sheets_reference.html\n",
    "# print(list(image_loader)[0].shape, plt.style.available)\n",
    "\n",
    "# show data\n",
    "plt.style.use('ggplot')\n",
    "fig, axs = plt.subplots(nrows=3,ncols=4)\n",
    "\n",
    "for ax in axs:\n",
    "    for a in ax:\n",
    "        a.set_title('{}'.format(next(train_label_loader)))\n",
    "        a.imshow(next(train_image_loader))\n",
    "\n",
    "next(train_image_loader).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ConvNet with Fully connected layer\n",
    "\n",
    "# import packages\n",
    "\n",
    "%matplotlib inline\n",
    "import torch\n",
    "import torch.utils.data.dataloader as dataloader\n",
    "import torch.nn as nn\n",
    "import torch.functional as F\n",
    "import torch.optim as optim\n",
    "from torchvision import datasets,transforms\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "# [compress_as_gzip(fashion_mnist[k],fashion_mnist[k]+'.gz') for k in fashion_mnist.keys()]\n",
    "\n",
    "# Data preparing\n",
    "train_sets = datasets.MNIST(root=\"../../GitRepos/data/\",\n",
    "                            train=True,\n",
    "                            transform=transforms.ToTensor(),\n",
    "                            download=True)\n",
    "\n",
    "test_sets = datasets.MNIST(root=\"../../GitRepos/data/\",\n",
    "                            train=False,\n",
    "                            transform=transforms.ToTensor())\n",
    "\n",
    "trainloader = dataloader.DataLoader(train_sets,\n",
    "                                    batch_size = 32,\n",
    "                                    shuffle = True,\n",
    "                                )\n",
    "testloader = dataloader.DataLoader(test_sets,\n",
    "                                batch_size = 32,\n",
    "                                shuffle = True,\n",
    "                                )\n",
    "\n",
    "# rely on size of input since there is fully connected layer\n",
    "# (28 + padding x 2)/stride - kernel + 1\n",
    "# 28 x 28, conv1 -> 28 + 2 x 2 - 5 + 1 = 28,maxpool -> 14\n",
    "# 14 x 14, conv2 -> 14 + 2 x 2 - 5 + 1 = 14,maxpool -> 7\n",
    "class CNN(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(CNN,self).__init__()\n",
    "        # 1 x 28 x28\n",
    "        self.conv1 = nn.Sequential(\n",
    "            nn.Conv2d(\n",
    "                in_channels     = 1,\n",
    "                out_channels    = 16,\n",
    "                kernel_size     = 5,\n",
    "                stride          = 1,\n",
    "                padding         = 2\n",
    "                ),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(kernel_size = 2)\n",
    "        )\n",
    "        # 16 x 14 x 14\n",
    "        self.conv2 = nn.Sequential(\n",
    "            nn.Conv2d(16,32,5,1,2),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(kernel_size = 2)\n",
    "        )\n",
    "        # 32 x 7 x 7\n",
    "        self.linear = nn.Linear(32*7*7,10)\n",
    "        # (batch_size,32*7*7)\n",
    "\n",
    "    def forward(self,x : torch.Tensor):\n",
    "        x = self.conv1(x)\n",
    "        # print(x.size())\n",
    "        x = self.conv2(x)\n",
    "        # print(x.size(0))\n",
    "        x = x.view(x.size(0),-1) # (batch_size,32*7*7)\n",
    "        out = self.linear(x)\n",
    "        return out\n",
    "\n",
    "\n",
    "# accuracy\n",
    "def accuracy(predictions : torch.Tensor, labels : torch.Tensor):\n",
    "    pred = torch.max(predictions,1)[1]\n",
    "    rights = pred.eq(labels.data.view_as(pred)).sum()\n",
    "    return rights,len(labels)\n",
    "\n",
    "# start training\n",
    "net = CNN()\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.Adam(net.parameters(),lr = 0.001)\n",
    "\n",
    "if False:\n",
    "    for epoch in range(1):\n",
    "        train_rights = []\n",
    "\n",
    "        for batch_idx,(data,target) in enumerate(trainloader):\n",
    "            net.train()\n",
    "            output = net(data)\n",
    "            loss = criterion(output,target)\n",
    "            optimizer.zero_grad()\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "            right = accuracy(output,target)\n",
    "\n",
    "            if batch_idx % 100 == 0:\n",
    "                net.eval()\n",
    "                right2 = 0\n",
    "                for b,(data,target) in enumerate(testloader):\n",
    "                    pred2 = net(data)\n",
    "                    right2 += accuracy(pred2,target)[0]\n",
    "\n",
    "                print('batch_idx {0}, right1 {1}, right2 {2}'.format(batch_idx,right[0],right2))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 1, 28, 28])\n"
     ]
    }
   ],
   "source": [
    "# net.eval()\n",
    "# 扩充维度unsqueeze() 缩减维度squeeze()去掉维度值=1的那个维度\n",
    "img = torch.Tensor(next(train_image_loader))\n",
    "# 扩充2次维度变成(batch_size, channels, w, h)\n",
    "img = torch.unsqueeze(img,dim=0)\n",
    "img = torch.unsqueeze(img,dim=0)\n",
    "print(img.size())\n",
    "# pred = net(img)\n",
    "\n",
    "# export model to .onnx\n",
    "# torch.onnx.export(\n",
    "#       net,\n",
    "#        img,\n",
    "#        '../../GitRepos/model/cnn.onnx',\n",
    "#        export_params = True,\n",
    "#        verbose = True,\n",
    "#        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 28, 28]) 5\n",
      "tensor([True]) tensor([[9.]]) tensor([9.])\n",
      "tensor([9]) [array([[-2113.1404 , -2054.1638 , -1824.9364 ,  -745.5421 ,  -350.51797,\n",
      "         -831.8036 , -2741.6367 ,  -319.57275,  -268.5156 ,  1474.91   ]],\n",
      "      dtype=float32)]\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import onnx\n",
    "import onnxruntime as ort\n",
    "\n",
    "input_names = []\n",
    "\n",
    "# model = onnx.load('../../GitRepos/model/cnn.onnx')\n",
    "ort_session = ort.InferenceSession('../../GitRepos/model/cnn.onnx')\n",
    "# input_feed是字典时,key值必须正确，可以在导出onnx时指定\n",
    "pred = ort_session.run(None,{'input.1':np.array(img)})\n",
    "output = torch.Tensor(np.array(pred[0]))\n",
    "mm = torch.max(output,1)[1]\n",
    "\n",
    "# datasets are lists of [[channel,w,h],label] 实现相应的接口，可以自己制作datasets\n",
    "print(train_sets[0][0].size(),train_sets[0][1])\n",
    "\n",
    "# view as : 相当于是忽略掉两者多余的维度，维度值=1的那个维度[]会被去掉\n",
    "print(torch.eq(mm,torch.Tensor([[9]]).view_as(mm)),torch.Tensor([[9]]),torch.Tensor([[9]]).view_as(mm))\n",
    "\n",
    "# display\n",
    "a = plt.axes()\n",
    "a.imshow(torch.squeeze(torch.squeeze(img)))\n",
    "print(mm,pred)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "pytorch",
   "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": "3.8.12 (default, Oct 12 2021, 03:01:40) [MSC v.1916 64 bit (AMD64)]"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "0c12c839adeec326d421be117b103b29d3b0068db2c28a64ae1df63f05821871"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}