import torchc++
import torch.nn as nn網絡
import torch.nn.functional as F #這個是函數形式定義,相對來講更低層一點。ide
"""函數
Net(oop
(conv1): Conv2d(1, 6, kernel_size=(3, 3), stride=(1, 1))優化
(conv2): Conv2d(6, 16, kernel_size=(3, 3), stride=(1, 1))ci
(fc1): Linear(in_features=576, out_features=120, bias=True)get
(fc2): Linear(in_features=120, out_features=84, bias=True)input
(fc3): Linear(in_features=84, out_features=10, bias=True)it
)
"""
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
# 1 input image channel, 6 output channels, 3x3 square convolution
# kernel
self.conv1 = nn.Conv2d(1, 6, 3)#注意這裏1的大小是1*1*32*32
self.conv2 = nn.Conv2d(6, 16, 3)
# an affine operation: y = Wx + b
self.fc1 = nn.Linear(16 * 6 * 6, 120) # 6*6 from image dimension
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
#卷經層計算公式:(w-k+2p)/s+1 k表明卷積尺寸,p是填充,s是步幅;
#池化層計算公式:(w-k)/s+1
def forward(self, x):
# Max pooling over a (2, 2) window
# conv1 layer calc:32-3+2*0/1+1 = 30
# pool1 layer calc:30/2 = 15
x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
#print(y.size())
# If the size is a square you can only specify a single number
#conv2 layer calc:15-3+2*0/1+1=13
#pool2 layer calc:13/2=6 這裏的問題是剩下的省略了?
x = F.max_pool2d(F.relu(self.conv2(x)), 2)#x 16*6*6
x = x.view(-1, self.num_flat_features(x))#展開
x = F.relu(self.fc1(x))#三個全鏈接層
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
#原有尺寸爲1*16*6*6;準確尺寸是16*6*6,第一個維度不要;
def num_flat_features(self, x):
size = x.size()[1:] # all dimensions except the batch dimension 取每個維度尺寸
num_features = 1
for s in size:#將每個維度尺寸乘;最終便是總的尺寸。
num_features *= s
return num_features
net = Net()
print(net)
params = list(net.parameters())#存儲其全部參數,依然是張量,打印後格式爲:
"""
torch.Size([6, 1, 3, 3]) torch.Size([6]) torch.Size([16, 6, 3, 3]) torch.Size([16]) torch.Size([120, 576]) torch.Size([120])
torch.Size([84, 120]) torch.Size([84])
torch.Size([10, 84]) torch.Size([10])
"""
for i in range(len(params)):
print(params[i].size(),end=' ')
input = torch.randn(1, 1, 32, 32)#隨機輸入
out = net(input)
print(out)
net.zero_grad( )#將全部梯度歸零
out.backward(torch.randn(1,10))#反向傳播,注意這裏的傳入方式是以前所說的雅各布矩陣
#偏差
output = net(input)
target = torch.randn(10)
target = target.view(1,-1)
loss = nn.MSELoss()(output,target)#相似c++函數傳入兩個參數,直觀上不大好看;
print(loss)#會進行相似輸出:tensor(0.4308, grad_fn=<MseLossBackward>)
#若是須要查看其調用屬性,可進行以下相似操做
print(loss.grad_fn) # MSELoss
print(loss.grad_fn.next_functions[0][0]) # Linear
print(loss.grad_fn.next_functions[0][0].next_functions[0][0]) # ReLU
#也能夠採起以下方式查看
tmpPrint = loss.grad_fn
for i in range(2):
print(tmpPrint.next_functions[0][0])
tmpPrint = tmpPrint.next_functions[0][0]
#反向傳播
net.zero_grad()#主要目的是防止微分時出現異常;
print('conv1.bias.grad before backward',net.conv1.bias.grad)
loss.backward()
print('conv1.bias.grad after backward',net.conv1.bias.grad)
#更新權重
learning_rate=0.01
for f in net.parameters():
f.data.sub_(f.grad.data*learning_rate)
#根據NN的特色,咱們可能會考慮使用不一樣的更新規則;因此整個流程簡化爲:
import torch.optim as optim
# create your optimizer
optimizer = optim.SGD(net.parameters(), lr=0.01)#優化器
# in your training loop:
optimizer.zero_grad() # zero the gradient buffers
output = net(input)
loss = criterion(output, target)
loss.backward()
optimizer.step() # Does the update
"""
因此整個流程變成:
1.定義網絡結構
2.輸入數據正向計算
3.計算損失
4.進行反向傳播
5.更新參數(方法可選)
"""