torch.ge()与nn.masked

ge/0573.jpg" alt="torch.ge()与nn.masked"/>

torch.ge()与nn.masked

1.torch.ge(input, other, *, out=None) → Tensor

Computes input≥other element-wise. 返回的是bool tensor，shape与input相同

Parameters

• input (Tensor) – the tensor to compare

• other (Tensor or float) – the tensor or value to compare

Keyword Arguments

out (Tensor, optional) – the output tensor.

Returns

A boolean tensor that is True where input is greater than or equal to other and False elsewhere

2.nn.masked_select()

配合torch.ge使用，用于筛选出满足ge条件的input中的数据

Returns a new 1-D tensor which indexes the input tensor according to the boolean mask mask which is a BoolTensor.

The shapes of the mask tensor and the input tensor don’t need to match, but they must be broadcastable.

Parameters

• input (Tensor) – the input tensor.

• mask (BoolTensor) – the tensor containing the binary mask to index with

Keyword Arguments

out (Tensor, optional) – the output tensor.

import torch
​
x = torch.randn([3, 4])
print(x)
# 将x中的每一个元素与0.5进行比较
# 当元素大于等于0.5返回True,否则返回False
mask = x.ge(0.5)
print(mask)
print(torch.masked_select(x, mask))​
'''
tensor([[ 1.2001,  1.2968, -0.6657, -0.6907],[-2.0099,  0.6249, -0.5382,  1.4458],[ 0.0684,  0.4118,  0.1011, -0.5684]])
tensor([[ True,  True, False, False],[False,  True, False,  True],[False, False, False, False]])
tensor([1.2001, 1.2968, 0.6249, 1.4458])
'''