linear_func

‘‘‘


class torch.nn.Linear（in_features，out_features，bias = True ）[来源]

参数：

in_features - 每个输入样本的大小
out_features - 每个输出样本的大小
bias - 如果设置为False，则图层不会学习附加偏差。默认值：True
‘‘‘



import torch


x = torch.randn(3, 2)  # 输入的维度是（3,2）
m = torch.nn.Linear(2, 4)  # 2,4是指维度
output = m(x)
print("x",x)
print(‘m.weight.shape:\n ‘, m.weight.shape,m.weight)
print(‘m.bias.shape:\n‘, m.bias.shape,m.bias)
print(‘output.shape:\n‘, output.shape,output)
‘‘‘
x tensor([[-0.4972,  1.2745],
        [ 1.2993, -0.6580],
        [-0.2165,  0.8603]])
        
m.weight.shape:torch.Size([4, 2]) Parameter containing:
tensor([[-0.5528, -0.1309],
        [ 0.6907,  0.5723],
        [-0.2242,  0.1904],
        [ 0.1678, -0.6903]], requires_grad=True)
        
m.bias.shape:torch.Size([4]) Parameter containing:
tensor([-0.1663, -0.0111,  0.4852,  0.5688], requires_grad=True)

output.shape:torch.Size([3, 4]) 
tensor([[-0.0584,  0.3749,  0.8394, -0.3944],
        [-0.7984,  0.5097,  0.0685,  1.2410],
        [-0.1593,  0.3317,  0.6975, -0.0614]], grad_fn=<AddmmBackward>)


-0.4972*-0.5528+1.2745* -0.1309+-0.1663=-0.0584
‘‘‘

linear_func

原文：https://www.cnblogs.com/hapyygril/p/11586860.html