pytorch常用小结

import numpy as np

features = np.array([[1,2,3],[4,5,6],[7,8,9]])
components = [True,True,True]
features = features[:, components]
print(features)
fea = np.array([[1,2,3],[4,5,6],[7,8,9]])
components = [True,False,True]
fea = fea[:, components]
print(fea)

结果：

[[1 2 3]
[4 5 6]
[7 8 9]]

[[1 3]
[4 6]
[7 9]]

2.生成均值为0，方差为1的离散正态分布

import torch

batch_size = 5
embedding_dim = 2
mean = torch.zeros(batch_size, embedding_dim)
print(mean)
std = mean + 1
print(std)
fakez = torch.normal(mean=mean, std=std) # 500*128，均值为0，方差为1的离散正态分布
print(fakez)

orch.randn(*sizes, out=None) → Tensor

返回一个张量，包含了从标准正态分布（均值为0，方差为1，即高斯白噪声）中抽取的一组随机数。

torch.rand(*sizes, out=None) → Tensor

返回一个张量，包含了从区间[0, 1)的均匀分布中抽取的一组随机数。

3.torch中的repeat函数：之后的维度，对应维度相乘

 alpha = torch.rand( 3, 1, 1)

 2 print(alpha)

 3 alpha = alpha.repeat(1, 2, 3)

 4 print(alpha) 

结果：

tensor([[[0.0029]],

[[0.0730]],

[[0.7339]]])

tensor([[[0.0029, 0.0029, 0.0029],
[0.0029, 0.0029, 0.0029]],

[[0.0730, 0.0730, 0.0730],
[0.0730, 0.0730, 0.0730]],

[[0.7339, 0.7339, 0.7339],
[0.7339, 0.7339, 0.7339]]])

将1改成2：

alpha = torch.rand( 3, 1, 1)
print(alpha)
alpha = alpha.repeat(2, 2, 3)
print(alpha)

结果：6个两行三列

tensor([[[0.8134]],

[[0.6626]],

[[0.5064]]])

tensor([[[0.8134, 0.8134, 0.8134],
[0.8134, 0.8134, 0.8134]],

[[0.6626, 0.6626, 0.6626],
[0.6626, 0.6626, 0.6626]],

[[0.5064, 0.5064, 0.5064],
[0.5064, 0.5064, 0.5064]],

[[0.8134, 0.8134, 0.8134],
[0.8134, 0.8134, 0.8134]],

[[0.6626, 0.6626, 0.6626],
[0.6626, 0.6626, 0.6626]],

[[0.5064, 0.5064, 0.5064],
[0.5064, 0.5064, 0.5064]]])