SelectAdaptivePool2d does not fail explicitly

[hugoWR]

SelectAdaptivePool2d does not fail explicitly when the pooling type is not supported. This can lead to awkward situation where a typo leads to an unexpected type of pooling.

from timm.layers import SelectAdaptivePool2d

inp = torch.rand(8, 1, 64, 40)

select_max = SelectAdaptivePool2d(pool_type='Max')
select_avg = SelectAdaptivePool2d(pool_type='avg')
select_wrong = SelectAdaptivePool2d(pool_type='gemdjfhlvbdsuhbf')


(select_avg(inp) == select_max(inp)).all() # True
(select_wrong(inp) == select_max(inp)).all() # True

Was this design intended ?

[rwightman]

@hugoWR I suppose it could be worth updating that old logic. Two things could be added, an assert pool_type == 'avg', '...in the else and a pool_type = pool_type.lower() before the if block is probably worthwhile.

[hugoWR]

I see it was introduced recently, that's a nice little change to have, thanks!