1、从均匀分布中生成值
w = torch.zeros(3, 5)
w
Out[75]:
tensor([[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]])
torch.nn.init.uniform_(w, a=10, b=15)
Out[76]:
tensor([[11.8949, 11.0836, 10.6348, 13.4524, 12.8051],
[14.5289, 11.3441, 10.0570, 11.0310, 11.3643],
[10.2919, 11.2083, 13.5757, 13.3987, 11.0059]])2、分布N(mean, std)中生成值
从给定均值和标准差的正态分布N(mean, std)中生成值，填充输入的张量或变量
w = torch.zeros(3, 5)
torch.nn.init.normal_(w, mean=0, std=0.1)
Out[78]:
tensor([[-0.1810, -0.0781, 0.0562, 0.0239, -0.0599],
[ 0.0340, 0.1520, 0.0534, 0.1895, 0.0135],
[ 0.0149, -0.1131, -0.0643, 0.0160, -0.2282]])3、使用值val填充输入Tensor
w = torch.empty(2, 5)
torch.nn.init.constant_(w, val=0.6)
Out[80]:
tensor([[0.6000, 0.6000, 0.6000, 0.6000, 0.6000],
[0.6000, 0.6000, 0.6000, 0.6000, 0.6000]])3.1、使用0，或者1 填充数据
torch.nn.init.zeros_(w)
Out[83]:
tensor([[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]])
torch.nn.init.ones_(w)
Out[85]:
tensor([[1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1.]])4、用单位矩阵填充二维输入张量
w = torch.empty(3, 5)
torch.nn.init.eye_(w)
Out[82]:
tensor([[1., 0., 0., 0., 0.],
[0., 1., 0., 0., 0.],
[0., 0., 1., 0., 0.]])
torch.nn.init.zeros_(w)5、其他常用的初始化方法
torch.nn.init.xavier_normal_(w)
Out[87]:
tensor([[ 0.2121, -0.3607, 0.6342, 0.1501, 0.0018],
[-0.0737, 0.6971, -0.2628, 0.1004, -0.0322],
[ 0.0093, 0.7139, 0.0263, 0.7135, 0.6979]])
torch.nn.init.xavier_uniform_(w)
Out[88]:
tensor([[-0.1675, -0.1284, -0.4856, 0.5762, -0.6135],
[ 0.0711, -0.8592, -0.0317, 0.6801, 0.4777],
[ 0.2965, -0.5528, -0.5425, 0.5166, 0.5759]])
torch.nn.init.kaiming_normal_(w)
Out[89]:
tensor([[ 0.0015, 0.0681, 0.5349, -0.0972, -0.8459],
[ 0.6095, -0.0047, 0.2383, 1.1911, -1.2320],
[-0.7059, -0.0080, 0.4166, 0.6686, -0.9375]])
torch.nn.init.kaiming_uniform_(w)
Out[90]:
tensor([[-0.2876, 0.3591, 0.7630, 0.5041, -0.6685],
[-0.6666, 0.5787, 0.9411, -0.0918, 1.0930],
[-0.5985, -0.9909, 0.4831, -0.6703, 0.0351]])到此这篇关于python使用torch随机初始化参数的文章就介绍到这了,更多相关torch随机初始化参数内容请搜索编程网以前的文章或继续浏览下面的相关文章希望大家以后多多支持编程网！
