Training and validation losses are nan When using FillMissing

Part 1 (2019)
1

Hi, I have a problem with Fill Missing function.

At first I was using a separate python code to replace all of my missing values with “-1”, and when training the model after about 2 days of training my training and validation losses have decreased but is still in the millions, so I was trying to remove them by the use of fillMissing and adding a separate “True/Falce Column”. Now when I try to train the model I get training and validation losses as nan…

as another finding I see that the first 3 columns in the tabular data were replaced with nan instead of the number.

Please help

2

i’m not sure if that would help but I have noticed that if you have a column that contains nan values only, then FillMissing will replace it all with nan -
since
df[‘my_all_null_col’].median() = nan

you can avoid that by applying this beforehand:

df = df.dropna(axis='columns', how='all')

note: of course, this only applies for continuous columns, since fill missing does not apply to categorical columns (in a categorical columns ‘nan’ is yet another category).