String categorical features causes error in "Error Bias Predicted vs Actual" visualization

[frtendero]

I have a dataset with some string categorical features. When generating regression performance report, I'm encountering the following error with the "Predicted vs Actual" scatter plot:

ValueError:
    Invalid value of type 'builtins.str' received for the 'cmax' property of scatter.marker       
        Received value: 'mycategoricalvalue'

    The 'cmax' property is a number and may be specified as:
      - An int or float

Root of the error is in the creation of the scatter plot segment_fig for categorical features, at site-packages\evidently\dashboard\widgets\reg_underperform_segments_table_widget.py", line 211, in calculate, which tries to create a color bar, which could make sense if categorical variable is ordinal, but not for this case.

A potential fix creating the figure for string feature types as follows:

segment_fig = px.scatter(
    merged_data,
    x=target_name,
    y=prediction_name,
    color=feature_name,
    facet_col='dataset',
)

merged_data is already created (from error bias histogram plot) and contains dataset column to make distinction of 'reference' / 'current' dataset.

This modification would result in:

[0lgaF]

Hi @frtendero !
Thanks for reporting!
The bug is fixed. 👩‍🔧 The update will be published in the next release.
#291