PyTorchでMNIST手書き認識を実現
25917 ワード
じっけんかんきょう
MNISTデータセットの紹介
MNISTは6万枚の28 x 28の訓練サンプル、1万枚のテストサンプルを含み、CVの「Hello Word」と言える.本論文で使用したCNNネットワークはMNISTデータの識別率を99%に向上させた.次は実戦を始めます.
パッケージのインポート
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
torch.__version__
スーパーパラメータの定義
BATCH_SIZE=512
EPOCHS=20
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
データセット
PyTorchに付属しているdatasetを直接使用し、DataLoaderを使用してトレーニングデータとテストデータをそれぞれ読み取ります.データセットをダウンロードした場合はFalseを選択できます
train_loader = torch.utils.data.DataLoader(
datasets.MNIST('data', train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])),
batch_size=BATCH_SIZE, shuffle=True)
test_loader = torch.utils.data.DataLoader(
datasets.MNIST('data', train=False, transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])),
batch_size=BATCH_SIZE, shuffle=True)
ネットワークの定義
このネットワークは、2つのボリューム層と2つの線形層を含み、最後に10次元、すなわち0〜9 10個の数字を出力する.
class ConvNet(nn.Module):
def __init__(self):
super().__init__()
self.conv1=nn.Conv2d(1,10,5) # input:(1,28,28) output:(10,24,24)
self.conv2=nn.Conv2d(10,20,3) # input:(10,12,12) output:(20,10,10)
self.fc1 = nn.Linear(20*10*10,500)
self.fc2 = nn.Linear(500,10)
def forward(self,x):
in_size = x.size(0)
out = self.conv1(x)
out = F.relu(out)
out = F.max_pool2d(out, 2, 2)
out = self.conv2(out)
out = F.relu(out)
out = out.view(in_size,-1)
out = self.fc1(out)
out = F.relu(out)
out = self.fc2(out)
out = F.log_softmax(out,dim=1)
return out
インスタンス化されたネットワーク
model = ConvNet().to(DEVICE) # gpu
optimizer = optim.Adam(model.parameters()) # Adam
トレーニング関数の定義
def train(model, device, train_loader, optimizer, epoch):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if(batch_idx+1)%30 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
テスト関数の定義
def test(model, device, test_loader):
model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += F.nll_loss(output, target, reduction='sum').item() #
pred = output.max(1, keepdim=True)[1] #
correct += pred.eq(target.view_as(pred)).sum().item()
test_loss /= len(test_loader.dataset)
print('
Test set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)
'.format(
test_loss, correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
訓練を始める
for epoch in range(1, EPOCHS + 1):
train(model, DEVICE, train_loader, optimizer, epoch)
test(model, DEVICE, test_loader)
じっけんけっか
Train Epoch: 1 [14848/60000 (25%)] Loss: 0.375058
Train Epoch: 1 [30208/60000 (50%)] Loss: 0.255248
Train Epoch: 1 [45568/60000 (75%)] Loss: 0.128060
Test set: Average loss: 0.0992, Accuracy: 9690/10000 (97%)
Train Epoch: 2 [14848/60000 (25%)] Loss: 0.093066
Train Epoch: 2 [30208/60000 (50%)] Loss: 0.087888
Train Epoch: 2 [45568/60000 (75%)] Loss: 0.068078
Test set: Average loss: 0.0599, Accuracy: 9816/10000 (98%)
Train Epoch: 3 [14848/60000 (25%)] Loss: 0.043926
Train Epoch: 3 [30208/60000 (50%)] Loss: 0.037321
Train Epoch: 3 [45568/60000 (75%)] Loss: 0.068404
Test set: Average loss: 0.0416, Accuracy: 9859/10000 (99%)
Train Epoch: 4 [14848/60000 (25%)] Loss: 0.031654
Train Epoch: 4 [30208/60000 (50%)] Loss: 0.041341
Train Epoch: 4 [45568/60000 (75%)] Loss: 0.036493
Test set: Average loss: 0.0361, Accuracy: 9873/10000 (99%)
Train Epoch: 5 [14848/60000 (25%)] Loss: 0.027688
Train Epoch: 5 [30208/60000 (50%)] Loss: 0.019488
Train Epoch: 5 [45568/60000 (75%)] Loss: 0.018023
Test set: Average loss: 0.0344, Accuracy: 9875/10000 (99%)
Train Epoch: 6 [14848/60000 (25%)] Loss: 0.024212
Train Epoch: 6 [30208/60000 (50%)] Loss: 0.018689
Train Epoch: 6 [45568/60000 (75%)] Loss: 0.040412
Test set: Average loss: 0.0350, Accuracy: 9879/10000 (99%)
Train Epoch: 7 [14848/60000 (25%)] Loss: 0.030426
Train Epoch: 7 [30208/60000 (50%)] Loss: 0.026939
Train Epoch: 7 [45568/60000 (75%)] Loss: 0.010722
Test set: Average loss: 0.0287, Accuracy: 9892/10000 (99%)
Train Epoch: 8 [14848/60000 (25%)] Loss: 0.021109
Train Epoch: 8 [30208/60000 (50%)] Loss: 0.034845
Train Epoch: 8 [45568/60000 (75%)] Loss: 0.011223
Test set: Average loss: 0.0299, Accuracy: 9904/10000 (99%)
Train Epoch: 9 [14848/60000 (25%)] Loss: 0.011391
Train Epoch: 9 [30208/60000 (50%)] Loss: 0.008091
Train Epoch: 9 [45568/60000 (75%)] Loss: 0.039870
Test set: Average loss: 0.0341, Accuracy: 9890/10000 (99%)
Train Epoch: 10 [14848/60000 (25%)] Loss: 0.026813
Train Epoch: 10 [30208/60000 (50%)] Loss: 0.011159
Train Epoch: 10 [45568/60000 (75%)] Loss: 0.024884
Test set: Average loss: 0.0286, Accuracy: 9901/10000 (99%)
Train Epoch: 11 [14848/60000 (25%)] Loss: 0.006420
Train Epoch: 11 [30208/60000 (50%)] Loss: 0.003641
Train Epoch: 11 [45568/60000 (75%)] Loss: 0.003402
Test set: Average loss: 0.0377, Accuracy: 9894/10000 (99%)
Train Epoch: 12 [14848/60000 (25%)] Loss: 0.006866
Train Epoch: 12 [30208/60000 (50%)] Loss: 0.012617
Train Epoch: 12 [45568/60000 (75%)] Loss: 0.008548
Test set: Average loss: 0.0311, Accuracy: 9908/10000 (99%)
Train Epoch: 13 [14848/60000 (25%)] Loss: 0.010539
Train Epoch: 13 [30208/60000 (50%)] Loss: 0.002952
Train Epoch: 13 [45568/60000 (75%)] Loss: 0.002313
Test set: Average loss: 0.0293, Accuracy: 9905/10000 (99%)
Train Epoch: 14 [14848/60000 (25%)] Loss: 0.002100
Train Epoch: 14 [30208/60000 (50%)] Loss: 0.000779
Train Epoch: 14 [45568/60000 (75%)] Loss: 0.005952
Test set: Average loss: 0.0335, Accuracy: 9897/10000 (99%)
Train Epoch: 15 [14848/60000 (25%)] Loss: 0.006053
Train Epoch: 15 [30208/60000 (50%)] Loss: 0.002559
Train Epoch: 15 [45568/60000 (75%)] Loss: 0.002555
Test set: Average loss: 0.0357, Accuracy: 9894/10000 (99%)
Train Epoch: 16 [14848/60000 (25%)] Loss: 0.000895
Train Epoch: 16 [30208/60000 (50%)] Loss: 0.004923
Train Epoch: 16 [45568/60000 (75%)] Loss: 0.002339
Test set: Average loss: 0.0400, Accuracy: 9893/10000 (99%)
Train Epoch: 17 [14848/60000 (25%)] Loss: 0.004136
Train Epoch: 17 [30208/60000 (50%)] Loss: 0.000927
Train Epoch: 17 [45568/60000 (75%)] Loss: 0.002084
Test set: Average loss: 0.0353, Accuracy: 9895/10000 (99%)
Train Epoch: 18 [14848/60000 (25%)] Loss: 0.004508
Train Epoch: 18 [30208/60000 (50%)] Loss: 0.001272
Train Epoch: 18 [45568/60000 (75%)] Loss: 0.000543
Test set: Average loss: 0.0380, Accuracy: 9894/10000 (99%)
Train Epoch: 19 [14848/60000 (25%)] Loss: 0.001699
Train Epoch: 19 [30208/60000 (50%)] Loss: 0.000661
Train Epoch: 19 [45568/60000 (75%)] Loss: 0.000275
Test set: Average loss: 0.0339, Accuracy: 9905/10000 (99%)
Train Epoch: 20 [14848/60000 (25%)] Loss: 0.000441
Train Epoch: 20 [30208/60000 (50%)] Loss: 0.000695
Train Epoch: 20 [45568/60000 (75%)] Loss: 0.000467
Test set: Average loss: 0.0396, Accuracy: 9894/10000 (99%)
まとめ
実際のプロジェクトのワークフロー:データセットを見つけて、データを前処理して、私たちのモデルを定義して、スーパーパラメータを調整して、テスト訓練して、更に訓練の結果を通じてスーパーパラメータを調整してあるいはモデルを調整します.