pytorch implementation of masked CRF

[ZenBel]

Hello, and thank you for your pytorch implementation of masked CRF. I have a question concerning this part of the code:

for i in range(self.mask_tran_matrix.shape[0]):
    for j in range(self.mask_tran_matrix.shape[1]):
        if self.mask_tran_matrix[i][j] == -10.0:
            self.transitions.data[i][j] = torch.tensor([-1.0])

Why do you assign torch.tensor([-1.0]) instead of torch.tensor([-10.0])?

[zhw666888]

Thank you for your question,this part is wrong,the new version has been updated,thanks for your attention!