Calculate binary entropy loss using a function in pytorch

Question

I have a problem about calculating binary cross entropy. The way I know that works out in pytorch is:

import torch
import torch.nn as nn
import torch.nn.functional as F
def lossfunc():
    return F.binary_cross_entropy

criterion = lossFunc()
input = torch.randn((3, 2), requires_grad=True)
target = torch.rand((3, 2), requires_grad=False)
loss = criterion(torch.sigmoid(input),target)

But how to complete the lossfunc() in such way, because I don't know how to pass the arguments to the function:

#the function that add sigmoid to input and calculate the binary cross entropy loss
def lossfunc():
   return

criterion = lossFunc()
input = torch.randn((3, 2), requires_grad=True)
target = torch.rand((3, 2), requires_grad=False)
loss = criterion(input,target)

Harshit Kumar · Accepted Answer · 2019-11-14 05:09:52Z

1

I think you're confusing the nn api with the functional F api. In functional api, loss function F.binary_cross_entropy can be used as a function directly.

In nn api, you need to create an object of the loss class such as criterion = nn.BCELoss()

Thus, you can simply do:

def lossFunc(input, target):
   return F.binary_cross_entropy(torch.sigmoid(input),target)

input = torch.randn((3, 2), requires_grad=True)
target = torch.rand((3, 2), requires_grad=False)
loss = lossFunc(input,target)

Also, PyTorch provides nn.nn.BCEWithLogitsLoss() and F.binary_cross_entropy_with_logits() that combines both sigmoid and binary cross-entropy.

edited Nov 14, 2019 at 5:09

answered Nov 14, 2019 at 5:03

Harshit Kumar

13.1k10 gold badges62 silver badges79 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Calculate binary entropy loss using a function in pytorch

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related