dask_xgboost_error: 'Series' object has no attribute 'compute‘ about featurewiz HOT 4 CLOSED

Hi @zw10 👍
I assume you are using the latest version. Can you confirm?

There is a dask data related problem when I see an issue like this. Can you please help me?
2. From the error, it seems that the dataset y_test (test data) is being assumed to be dask by featurewiz while it is not. Can you please post your calling statement for featurewiz? That and a few lines of data (both train and test) would help.

you can attach the data snippet as a zip file and drop it in the comment section. It accepts zip files only as attachments.

Thank you,
AutoViML

from featurewiz.

Hi - thanks for the quick response!

Yes using the latest version as far I'm aware - 0.80. Agreed - for some reason I'm calling the function with a pandas dataframe, but FeatureWiz seems to be assuming a dask dataframe.

                output = FW.featurewiz(df, col, corr_limit=0.70, verbose=2, sep=',', header=0
                                       , test_data='', feature_engg='interactions', category_encoders=''
                                       , dask_xgboost_flag=True, nrows=1000)

I also attach the first few rows of the input dataframe.

error_sample.zip

from featurewiz.

Hi @zw10 👍

I think i have found the problem. I have fixed it and you can upgrade via:
pip install featurewiz --upgrade

Another very simple trick when you hit a problem with dask again is to just set dask_xgboost_flag = 'False` in your calling statement and try running it. That might avoid the dask error. If that works, all the best.
AutoViML

from featurewiz.

Thank you - confirming that the fix is working as intended, after the upgrade to v0.81

And dask_xgboost_flag='False' works for v0.80

from featurewiz.