Standardization and normalization is switched

[reneburghardt]

Standardization is subtracting the sample mean and dividing by the sample std. Normalization in our case is variance normalization, so just dividing by sample std.
In ChunkDataset it seems that it is switched. Here the mean statistic, which is later used in the transform, is set to zeros for standardize (and not normalize):

experanto/experanto/datasets.py

Line 368 in 47f5f84

if mode == 'standardize':

[pollytur]

@reneburghardt how about we rename standardize to normalize_variance_only to avoid this conclusion? I do not want to rename normalize to standartize as they are often used interchangeably and this might lead to more confusion and (current) normalize calls torchvision.transforms.Normalize under the hood (which is somewhat intuitive)

Any opinions?

[reneburghardt]

Yes I agree @pollytur