PyTorch實戰mnist圖像分類
- 項目結構
- 項目代碼
項目結構
項目結構如圖,代碼都放在mnistclassify.py裡面,data資料是代碼執行過程中自己下載下傳的。
項目代碼
- 導入包,建構訓練集測試集
from random import shuffle
from turtle import forward
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchvision import datasets,transforms
import matplotlib.pyplot as plt
import numpy as np
# 定義超參數
input_size = 28
num_classes = 10
num_epoches = 3
batch_size = 64
# 訓練集
train_dateset = datasets.MNIST(root='./data',train=True,transform=transforms.ToTensor(),download=True)
# 測試集
test_dateset = datasets.MNIST(root='./data',train=True,transform=transforms.ToTensor())
# 建構batch資料
train_loader = torch.utils.data.DataLoader(dataset=train_dateset,batch_size=batch_size,shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dateset,batch_size=batch_size,shuffle=True)
- 建構神經網絡
# 建構網絡
class CNN(nn.Module):
def __init__(self) -> None:
super(CNN, self).__init__()
self.conv1 = nn.Sequential(
nn.Conv2d(
in_channels=1, # 灰階圖
out_channels=16, # 輸出特征圖個數
kernel_size=5, # 卷積核大小
stride=1, # 步長
padding=2, # 邊緣填充,如果stride=1,希望卷積後的圖像和原來的圖像一樣大則設定padding=(kernal_size-1)/2
), # 輸出特征圖為(16,28,28)
nn.ReLU(),
nn.MaxPool2d(kernel_size=2) # 2*2最大池化,結果為(16,14,14)
)
self.conv2 = nn.Sequential( # 輸入(16,14,14)
nn.Conv2d(16, 32, 5, 1, 2), # 輸出(32,14,14)
nn.ReLU(),
nn.MaxPool2d(2), # 輸出(32,7,7)
)
self.out = nn.Linear(32 * 7 *7, 10) # 全連接配接得到結果
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size(0), -1) # 将結果轉換為向量,友善下一步全連接配接(32*7*7)
output = self.out(x)
return output
- 執行個體化網絡開始訓練
# 預測準确率
def accuracy(predictins, labels):
pred = torch.max(predictins.data, 1)[1]
rights = pred.eq(labels.data.view_as(pred)).sum()
return rights, len(labels)
# 執行個體化神經網絡
net = CNN()
# 損失函數
criterion = nn.CrossEntropyLoss()
# 優化器
optimizer = optim.Adam(net.parameters(), lr=0.001)
# 開始訓練循環
for epoch in range(num_epoches):
# 儲存目前epoch結果
train_rights = []
for batch_idx, (data, target) in enumerate(train_loader):
net.train()
output = net(data)
loss = criterion(output, target)
optimizer.zero_grad()
loss.backward()
optimizer.step()
right = accuracy(output, target)
train_rights.append(right)
if batch_idx % 100 == 0:
net.eval()
val_rights = []
for (data, target) in test_loader:
output = net(data)
right = accuracy(output, target)
val_rights.append(right)
# 準确率計算
train_r = (sum([tup[0] for tup in train_rights]), sum([tup[1] for tup in train_rights]))
val_r = (sum([tup[0] for tup in val_rights]), sum([tup[1] for tup in val_rights]))
print('目前epoch:{} [{}/{}({:.0f}%)]\t損失: {:.6f}\t訓練集準确率: {:.2f}%\t測試集準确率: {:.2f}%'.format(
epoch, batch_idx * batch_size, len(train_loader.dataset),
100. * batch_idx / len(train_loader),
loss.data,
100. * train_r[0].numpy() / train_r[1],
100. * val_r[0].numpy() / val_r[1],
))
- 訓練結果
目前epoch:0 [0/60000(0%)] 損失: 2.290263 訓練集準确率: 6.25% 測試集準确率: 11.39%
目前epoch:0 [6400/60000(11%)] 損失: 0.222888 訓練集準确率: 76.14% 測試集準确率: 90.28%
目前epoch:0 [12800/60000(21%)] 損失: 0.275965 訓練集準确率: 84.60% 測試集準确率: 94.70%
目前epoch:0 [19200/60000(32%)] 損失: 0.071834 訓練集準确率: 88.24% 測試集準确率: 95.60%
目前epoch:0 [25600/60000(43%)] 損失: 0.029019 訓練集準确率: 90.25% 測試集準确率: 96.68%
目前epoch:0 [32000/60000(53%)] 損失: 0.159890 訓練集準确率: 91.48% 測試集準确率: 97.08%
目前epoch:0 [38400/60000(64%)] 損失: 0.080257 訓練集準确率: 92.39% 測試集準确率: 97.00%
目前epoch:0 [44800/60000(75%)] 損失: 0.100067 訓練集準确率: 93.11% 測試集準确率: 97.57%
目前epoch:0 [51200/60000(85%)] 損失: 0.105826 訓練集準确率: 93.66% 測試集準确率: 97.84%
目前epoch:0 [57600/60000(96%)] 損失: 0.042444 訓練集準确率: 94.11% 測試集準确率: 98.05%
目前epoch:1 [0/60000(0%)] 損失: 0.169493 訓練集準确率: 93.75% 測試集準确率: 98.01%
目前epoch:1 [6400/60000(11%)] 損失: 0.033878 訓練集準确率: 98.04% 測試集準确率: 97.87%
目前epoch:1 [12800/60000(21%)] 損失: 0.108467 訓練集準确率: 98.05% 測試集準确率: 98.01%
目前epoch:1 [19200/60000(32%)] 損失: 0.007603 訓練集準确率: 97.97% 測試集準确率: 98.35%
目前epoch:1 [25600/60000(43%)] 損失: 0.202825 訓練集準确率: 98.04% 測試集準确率: 98.49%
目前epoch:1 [32000/60000(53%)] 損失: 0.113783 訓練集準确率: 98.11% 測試集準确率: 98.47%
目前epoch:1 [38400/60000(64%)] 損失: 0.027782 訓練集準确率: 98.11% 測試集準确率: 98.46%
目前epoch:1 [44800/60000(75%)] 損失: 0.034398 訓練集準确率: 98.12% 測試集準确率: 98.51%
目前epoch:1 [51200/60000(85%)] 損失: 0.013913 訓練集準确率: 98.18% 測試集準确率: 98.51%
目前epoch:1 [57600/60000(96%)] 損失: 0.021681 訓練集準确率: 98.19% 測試集準确率: 98.91%
目前epoch:2 [0/60000(0%)] 損失: 0.052889 訓練集準确率: 96.88% 測試集準确率: 98.72%
目前epoch:2 [6400/60000(11%)] 損失: 0.070504 訓練集準确率: 98.95% 測試集準确率: 98.86%
目前epoch:2 [12800/60000(21%)] 損失: 0.104337 訓練集準确率: 98.67% 測試集準确率: 98.85%
目前epoch:2 [19200/60000(32%)] 損失: 0.028965 訓練集準确率: 98.72% 測試集準确率: 98.70%
目前epoch:2 [25600/60000(43%)] 損失: 0.048499 訓練集準确率: 98.70% 測試集準确率: 98.82%
目前epoch:2 [32000/60000(53%)] 損失: 0.021659 訓練集準确率: 98.70% 測試集準确率: 98.80%
目前epoch:2 [38400/60000(64%)] 損失: 0.002921 訓練集準确率: 98.72% 測試集準确率: 98.95%
目前epoch:2 [44800/60000(75%)] 損失: 0.015612 訓練集準确率: 98.70% 測試集準确率: 98.92%
目前epoch:2 [51200/60000(85%)] 損失: 0.043291 訓練集準确率: 98.71% 測試集準确率: 99.08%
目前epoch:2 [57600/60000(96%)] 損失: 0.033159 訓練集準确率: 98.72% 測試集準确率: 99.01%
如有代碼不懂或者報錯請評論區留言,部落客幫忙解答、調試。