Bug in RevIN for POTS data

[WenjieDu]

1. System Info

PyPOTS v0.6

2. Information

• The official example scripts
• My own created scripts

3. Reproduction

As shown below, for POTS data case, x = x / self.stdev is not applied.

PyPOTS/pypots/nn/modules/revin/layers.py

Lines 65 to 74 in ac3318f

	else:
	# pypots implementation for POTS data
	missing_sum = (
	torch.sum(missing_mask == 1, dim=dim2reduce, keepdim=True) + 1e-9
	)
	self.mean = torch.sum(x, dim=dim2reduce, keepdim=True) / missing_sum
	x = x - self.mean
	x_enc = x.masked_fill(missing_mask == 0, 0)
	variance = torch.sum(x_enc * x_enc, dim=dim2reduce, keepdim=True) + 1e-9
	self.stdev = torch.sqrt(variance / missing_sum)

4. Expected behavior

Normalization process has not been finished for POTS data case.