BinaryMatthewsCorrCoef outputs 0 if target and prediction contains only one case either positive or negative case

[lokesh1199]

Hi,

When the given target and pred contains only one case either positive or negative (case - 1), The output of BinaryMatthewsCorrCoef is always 0. But we can clearly say that predictions are same as the targets. And for the case - 2, If the preds and targets are completely different, the MCC is also zero. Is this expected behaviour with MCC or some other issue.

mcc = BinaryMatthewsCorrCoef()

# preds and targets are equal (case - 1)
print(mcc(torch.ones(10), torch.ones(10)))
print(mcc(torch.zeros(10), torch.zeros(10)))

# preds and targets are completely different (case - 2)
print(mcc(torch.ones(10), torch.zeros(10)))

tensor(0)
tensor(0)
tensor(0)

In this blog, They are handling these cases slightly different. Below are the MCC values using their implementation. These values represent model performance more accurately than the torchmetrics version.

case - 1 -> MCC = 1.0
case - 2 -> MCC = -1.0

Kindly tell if this should be handled differently?