Jupyter kernel abort when plotting a column with pandas type "category"

[dhuntley1023]

I ran into the issue documented in Issue #230 and was thinking that if I marked the column as dtype=category in pandas that the dataprep plot would use that as a hint that the columns was nominal and not continuous. However when I do this, it is reliably causing a Jupyter kernel abort with no useful messages printed to the shell where jupyter notebook was started.

Repro:

• Create a new environment as ```conda create -n test dataprep notebook=6.1.5`` (Note the notebook version is required to get around the issue I mentioned in #396 )
• Start the notebook and run the code below
• After several seconds, a message displays 'The kernel appears to have died. It will restart automatically.'

import pandas as pd
import dataprep.eda as dp

df=pd.DataFrame({'Continuous': [1,2,3,4,5], 'Categorical': [0,1,1,0,0]})
df['Categorical'] = df['Categorical'].astype('category')
df.info()
dp.plot(df, 'Categorical')

[dovahcrow]

@dhuntley1023 Thanks for the reporting! This sounds like an issue that happened in the C code, either pandas or dask. @brandonlockhart I think you are more familiar with the categorical support than I. Do you have any thoughts?

[brandonlockhart]

I cannot reproduce the error, can you @dovahcrow? However, I use Jupyter installed with pip, not conda. @dhuntley1023 could you please try .astype('object') rather than .astype('category') to convert a column to nominal values and see if that works?

[dhuntley1023]

When I use .astype('object'), it works normally (i.e. no crashes). Would it be helpful to get a dump of the package versions in my environment?

[brandonlockhart]

Thanks for following up @dhuntley1023!

Would it be helpful to get a dump of the package versions in my environment?

Yes, that would be helpful, thanks.

I'm unsure why category is causing a problem. Under-the-hood of dataprep we can convert category to object like we do for string type #377, but I think some more investigation would be useful to see if we can figure out the problem.

[dovahcrow]

let's see with a newer release if this problem is fixed