【PyTorch】手書き文字判別においてニューラルネットワークはどうあるべきか？精度９９％を目指して活性化関数などを変更してみた

import torch
import torch.nn as nn
import torch.nn.functional as f
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
import matplotlib.pyplot as plt

# train用データ
train_loader = DataLoader(
    datasets.MNIST('./',
                    train=True,
                    download=True,
                    transform=transforms.Compose([
                        transforms.ToTensor()
                    ])),
    batch_size=batch_size,
    shuffle=True
    )

# eval用データ
test_loader = DataLoader(
    datasets.MNIST('./',
                    train=False,
                    transform=transforms.Compose([
                        transforms.ToTensor()
                    ])),
    batch_size=batch_size,
    shuffle=True)

data_loader_dict = {'train': train_loader, 'test': test_loader}

class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.gelu = nn.GELU()
        self.pool = nn.MaxPool2d(2, stride=2)

        self.conv1 = nn.Conv2d(1,28,3)
        self.conv2 = nn.Conv2d(28,32,3)

        self.fc1 = nn.Linear(32*5*5, 100)
        self.fc2 = nn.Linear(100, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = self.gelu(x)
        x = self.pool(x)
        x = self.conv2(x)
        x = self.gelu(x)
        x = self.pool(x)
        x = x.reshape(x.size()[0], -1)
        x = self.fc1(x)
        x = self.gelu(x)
        x = self.fc2(x)
        return f.log_softmax(x, dim=1)

epochs = 20
batch_size = 200

train_loss_list = []
test_loss_list = []
test_acc_list = []

net = Net()

optimizer = torch.optim.AdamW(params=net.parameters(), lr=0.005) # 最適化アルゴリズムを選択

for epoch in range(epochs):
    """ Training """
    loss = None
    net.train()
    for i, (data, target) in enumerate(data_loader_dict['train']):
        optimizer.zero_grad()
        output = net(data)
        loss = f.nll_loss(output, target)
        loss.backward()
        optimizer.step()

        print("\rTraining log: {0} epoch ({1} / {2}). Loss: {3}%".format(epoch+1, (i+1)*batch_size, len(train_loader)*batch_size, loss.item()), end="")

    train_loss_list.append(loss)

    """ eval """
    net.eval()
    test_loss = 0
    correct = 0

    with torch.no_grad():
        for data, target in data_loader_dict['test']:
            output = net(data)
            test_loss += f.nll_loss(output, target, reduction='sum').item()
            pred = output.argmax(dim=1, keepdim=True)
            correct += pred.eq(target.view_as(pred)).sum().item()

    test_loss /= 10000

    print("")
    print('Test loss (avg): {0}, Accuracy: {1}'.format(test_loss, correct / 10000), sep="")

    test_loss_list.append(test_loss)
    test_acc_list.append(correct / 10000)

# 結果の出力と描画
print("test_acc", test_acc_list)
plt.figure()
plt.plot(range(1, epochs+1), train_loss_list, label='train_loss')
plt.plot(range(1, epochs+1), test_loss_list, label='test_loss')
plt.xlabel('epoch')
plt.legend()

plt.figure()
plt.plot(range(1, epochs+1), test_acc_list)
plt.title('test accuracy')
plt.xlabel('epoch')