NLLLoss function wrt to getting a error of "bool value of tensor with more than one value is ambiguous"

[Bhargava]

In PyTorch nn.NLLLoss() function, when we apply it directly to calculate loss value wrt input and Target value, we get a error mentioning “bool value of Tensor with more than one value is ambiguous”.
When we mention something like criterion=nn.NLLLoss() and then apply criterion (input,Target) we get the loss tensor output. Any idea why this happens?

[KFrank]

Hi Bhargava!

Without seeing you code, I’m only guessing. However …

@Bhargava

In PyTorch nn.NLLLoss() function, when we apply it directly to calculate loss value wrt input and Target value, we get a error mentioning “bool value of Tensor with more than one value is ambiguous”.

NLLLoss is a class, instances of which are callable function objects.

(nll_loss is a function.)

My guess is that you are calling the constructor of NLLLoss with your
input and Target tensors. The second argument, size_average to
NLLLoss’s constructor is a bool, hence your error.

nn.NLLLoss (input, Target)   # bad -- calling constructor incorrectly
nn.NLLLoss() (input, Target)   # good -- calling a default-constructed function object correctly
nn.functional.nll_loss (input, Target)   # good -- calling ordinary function

Good luck!

K. Frank