Logits Overflow Problem

[mohitsharma29]

Hey, thanks for releasing your code. I installed the pip version of your code and am trying to determine the fairness-accuracy tradeoff on my dataset.

1. Here is the error I am getting. Could you help us figure out what this issue is?

RuntimeError Traceback (most recent call last)
Cell In[17], line 162
159 for lam in np.arange(10, 1000, 10):
160 fermi_instance = FERMI(train_embeddings, test_embeddings, train_soft.reshape(-1, 1), test_soft.reshape(-1, 1),
161 S_train=np.eye(2)[train_attrs], S_test=np.eye(2)[test_attrs], batch_size=64, epochs=2000, lam=lam)
--> 162 all_weights[lam] = FERMI_Logistic_Regression(fermi_instance)

Cell In[17], line 142
140 loss_min = criterion(logits, YTorch)
141 else:
--> 142 loss_min = criterion(logits, YTorch) + fermi.fairness_regularizer(XTorch, STorch)
143 # loss_min = criterion(logits, YTorch)
145 minimizer.zero_grad()

File ~/.conda/envs/llms/lib/python3.11/site-packages/torch/nn/modules/module.py:1739, in Module._wrapped_call_impl(self, *args, **kwargs)
1737 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1738 else:
-> 1739 return self._call_impl(*args, **kwargs)

File ~/.conda/envs/llms/lib/python3.11/site-packages/torch/nn/modules/module.py:1750, in Module._call_impl(self, *args, **kwargs)
1745 # If we don't have any hooks, we want to skip the rest of the logic in
1746 # this function, and just call forward.
1747 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1748 or _global_backward_pre_hooks or _global_backward_hooks
1749 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1750 return forward_call(*args, **kwargs)
1752 result = None
1753 called_always_called_hooks = set()

File ~/.conda/envs/llms/lib/python3.11/site-packages/torch/nn/modules/loss.py:699, in BCELoss.forward(self, input, target)
698 def forward(self, input: Tensor, target: Tensor) -> Tensor:
--> 699 return F.binary_cross_entropy(
700 input, target, weight=self.weight, reduction=self.reduction
701 )

File ~/.conda/envs/llms/lib/python3.11/site-packages/torch/nn/functional.py:3569, in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
3566 new_size = _infer_size(target.size(), weight.size())
3567 weight = weight.expand(new_size)
-> 3569 return torch._C._nn.binary_cross_entropy(input, target, weight, reduction_enum)

RuntimeError: all elements of input should be between 0 and 1

2. The code library seems to only support DP. Where can I find extensions for EOP/EODDS?