1. 简介
(1) 结构
(2) 激活函数
2.模型设计
(1) Model
import torch.nn as nn
from collections import OrderedDict
layers = OrderedDict() # 创建顺序的dict结构
for i, n_hidden in enumerate(n_hiddens):
layers['fc{}'.format(i+1)] = nn.Linear(current_dims, n_hidden)
layers['relu{}'.format(i+1)] = nn.ReLU()
layers['drop{}'.format(i+1)] = nn.Dropout(0.2)
current_dims = n_hidden
layers['out'] = nn.Linear(current_dims, n_class)
model = nn.Sequential(layers) # 顺序的执行layers
print(model)
model = torch.nn.DataParallel(model, device_ids= range(ngpu)) # 数据并行
(2) Optimizer
# 采用随机梯度下降算法
# lr表示学习率
# weight_decay表示权重衰减，防止模型过拟合
# momentum加速模型的迭代，参见3
optimizer = optim.SGD(model.parameters(), lr=0.01, weight_decay=0.0001, momentum=0.9)
for epoch in range(epochs):
model.train() # 训练模式
if epoch in [80,200]:
optimizer.param_groups[0]['lr'] *= 0.1 # 随着模型跌打的次数增加，学习率降低
for batch_idx, (data, target) in enumerate(train_data):
data, target = Variable(data), Variable(target)
optimizer.zero_grad() # 清除上一轮的梯度，否则会进行累加
output = model(data) # 动态图结构
loss = F.cross_entropy(output, target) # 交叉熵
loss.backward()
optimizer.step()
(3) Evaluation
# 模型训练时，按照epoch的次数进行效果评估
if epoch % 10 == 0:
model.eval() # 评估模式
test_loss = 0
correct = 0
for data, target in test_date:
data, target = Variable(data, volatile=True), Variable(target)
output = model(data)
test_loss += F.cross_entropy(output, target).data[0]
pred = output.data.max(1)[1]
correct += pred.cpu().eq(indx_target).sum()
test_loss = test_loss / len(test_loader)
acc = 100. * correct / len(test_loader.dataset)
3. 参考资料
1. https://github.com/aaron-xichen/pytorch-playground
2. PyTorch API
3. On the importance of initialization and momentum in deep learning
4.Fully Connected Neural Network Algorithms
5.Fully Connected Neural Network与Activation Function