Posted in Python onMarch 22, 2022
目录
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三水点靠木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三水点靠木!
python使用torch随机初始化参数
- Author -
就是一顿骚操作- Original Sources -
声明:登载此文出于传递更多信息之目的,并不意味着赞同其观点或证实其描述。
Reply on: @reply_date@
@reply_contents@