Error: Cannot normalize 'days_from_start' column as it isn't numerical. Are you sure it doesn't belong in the categorical set of columns?

Part 1 (2019)
1

Getting this error when attempting to create a TabularList …

Cannot normalize 'days_from_start' column as it isn't numerical. Are you sure it doesn't belong in the categorical set of columns?

Here is the code:

procs = [FillMissing, Categorify, Normalize]

il = TabularList.from_df(sample_df, cat_cols, cont_cols, procs=procs)
ils = il.random_split_by_pct(valid_pct=0.1, seed=42)
ll = ils.label_from_df(dep_var)  # <= exception occurs here

There may be nulls in this column but I would have expected the FillMissing proc to take care of them before Normalize was even called. The dtype = int64 for that column (and the same problem happens for other columns that are dtype = float.

Any ideas?

2

SOLVED

Not sure why this is required, but you have to also ensure your DataFrame passed into TabularList.from_df() consists of only those columns defined by cat_cols, cont_cols, and dep_var).