muellerzr / fastai-experiments-and-tips Goto Github PK

I must be missing something basic here.

then you get the parameters for the transformations like the mean and sd for normalization
and split the data into two parts - train_df and test_df

train_df and test_df are not normalized. But the mean and sd are saved for normalization

in each iteration of the loop you compute
data_fold (different splitting, do the normalization based on the original mean and sd
data_test use all the data in test_df and the original normalization mean and sd

I would have thought the data_test is the same every iteration - so why is recomputed?

do you have code working for latest fastai version to calculate
number of layers and number neurons for each layer ?
is there difference for tabular data when first layers is embedding layer