torch.index_fill的理解

通过按index中给定的顺序 选择索引，用val值填充 自己（自张量）的元素。
dim（int）–索引所依据的维度
index（LongTensor）–要填充的自张量的索引
val（浮点数）–要填充的值

import torch

a = torch.randn(4, 3)
print(a)
# tensor([[-1.7189,  0.9798, -0.0428],
#         [ 0.7184, -0.2824, -1.0289],
#         [ 1.2858,  0.8423, -1.0473],
#         [-0.0269, -0.9876, -2.3126]])

index = torch.tensor([0, 2])
b=a.index_fill(1, index, 9)#要填充1维
print(b)
# tensor([[ 9.0000,  0.9798,  9.0000],
#         [ 9.0000, -0.2824,  9.0000],
#         [ 9.0000,  0.8423,  9.0000],
#         [ 9.0000, -0.9876,  9.0000]])

c=a.index_fill(0, index, 9)#要填充0维
print(c)
# tensor([[ 9.0000,  9.0000,  9.0000],
#         [ 0.7184, -0.2824, -1.0289],
#         [ 9.0000,  9.0000,  9.0000],
#         [-0.0269, -0.9876, -2.3126]])