pytorch one-hot tips

首先介绍一下np.eye()

numpy.eye(N,M=None,k=0,dtype=<class 'float'>,order='C)

参数	类型	Value
N	int	Indicates the number of lines of output
M	int型	输出的列数,默认为N
k	int型	对角线的下标,默认为0表示的是主对角线,负数表示的是低对角,正数表示的是高对角.
dtype	$1	数据的类型,可选项,The data type of the returned data
order	{‘C’,‘F’}	可选项,That is, the form of the output array is in accordance withClanguage line precedence’C’,还是按照FortranColumn-majority of the form‘F’存储在内存中

例子：

import numpy as np

labels = np.array([[1], [2], [0], [1]])
print("labels的大小：", labels.shape, "\n")

# Because our category is from0-2,所以这里是3个类
a = np.eye(3)[1]
print("If the corresponding category number is 1,那么转成one-hot的形式", a, "\n")

a = np.eye(3)[2]
print("If the corresponding category number is 2,那么转成one-hot的形式", a, "\n")

a = np.eye(3)[1, 0]
print("1转成one-hotThe first number of the array is ：", a, "\n")

# The difference between the results here and above,注意!!!
a = np.eye(3)[[1, 2, 0, 1]]
print("If the corresponding category number is 1,2,0,1,那么转成one-hot的形式\n", a)

res = np.eye(3)[labels.reshape(-1)]
print("labels转成one-hot形式的结果：\n", res, "\n")
print("labels转化成one-hot后的大小：", res.shape)