一、遇到的磕磕绊绊:
1.一维数组 和 二维矩阵数据之间的区别:
二、用到的一些代码:
#备注,矩阵点乘 就是 A*B
import torch
from d2l import torch as d2l #这个库是李沐自己写的,我去
from torch import nn
def corr(X,k):
h,w = k.shape
Y = torch.zeros((X.shape[0]-h+1,X.shape[1]-w+1))
for i in range(Y.shape[0]):
for j in range(Y.shape[1]):
Y[i,j] = ((X[i:i+h,j:j+w])*k).sum() #这里证明了,先h 后w,谢谢
return Y
X =torch.tensor([[0.0,1.0,2.0],[1.0,2.0,3.0],[2.0,3.0,4.0]])
K =torch.tensor([[0.0,1.0],[2.0,3.0]])
corr(X,K)
#手动利用nn.module的模板实现 二维卷积层:
class my_conv(nn.Module):
def __init__(self,kernel_size):
super().init()
self.weight =nn.Parameter(torch.rand(kernel_size))
self.bias = nn.Parameter(torch.zeros(1))
def forward(self,x):
return corr(x,self.weight)+self.bias
#利用卷积层 实现对图像的 颜色边缘进行检测:
X=torch.ones(6,8)
X[:,2:6] = 0
X
import matplotlib.pyplot as plt
show_x = X.numpy()
plt.imshow(show_x)
plt.show()
Kernel = torch.tensor([[1.0,-1.0]])
Y = corr(X,Kernel)
show_out = Y.numpy()
plt.imshow(show_out)
plt.show()
Y
#通过使用nn.conv2d 自己学出 可以分辨 颜色边界的 kernel
conv2d = nn.Conv2d(in_channels=1,out_channels=1,kernel_size=(1,2),bias = False)
X = X.reshape(1,1,6,8)
Y = Y.reshape(1,1,6,7)
for i in range(10):
conv2d.zero_grad() #先清空前一次的grad
Y_hat = conv2d(X)
loss = (Y_hat-Y)**2 #手动计算MSE
loss.sum().backward()
#手动进行optim.step
conv2d.weight.data[:] -=3e-2*conv2d.weight.grad
#每个epoch进行必要的输出:
print('epoch : ',i ," loss = ",loss.sum().item())
print(conv2d.weight.data.reshape(1,2))