[one-hot]one-hot实现的几种方式

one-hot实现的几种方式

方式一：最简单的方式，可以一行代码搞定

import numpy as np
import torch

'one-hot使用方法：最简单'
num_classes = 10
arr = [1, 3, 4, 5]
one_hots = np.eye(num_classes)[arr]
print(one_hots)

arr1 = torch.LongTensor(arr)
print(arr1)
one_hots = torch.eye(num_classes)[arr]
print(one_hots)

arr2 = np.array(arr)
print(arr2)
one_hots = np.eye(num_classes)[arr]
print(one_hots)

方式二

'one-hot使用方法二：'
one_hot = np.zeros([len(arr2), num_classes])
for i, j in enumerate(arr):
    one_hot[i][int(j)] = 1
print(one_hot)

方式三

'one-hot使用方式三：'
# one_hot = torch.zeros((len(arr1), num_classes)).scatter_(dim=1, index=arr1[:, None], value=1)
num_classes = 10
y = torch.tensor([1, 3, 4, 5])
one_hot = torch.zeros(y.shape[0], num_classes).scatter_(1, y.reshape(-1, 1), 1)
print(one_hot)

方式四：白嫖的方式【最爽】

from torch.nn.functional import one_hot

'pytorch创造one_hot'
num_classes = 10
y = torch.tensor([1, 3, 4, 5])
one_hot = torch.zeros(y.shape[0], num_classes).scatter_(1, y.reshape(-1, 1), 1)
one_hot = one_hot.scatter_(y) 

print(one_hot)