PR to handle columns of a dataframe that are not of type float #218
Conversation
…is thrown when its columns are not float, this commit solves this issue
|
May be if I change the way I retrieve column names depending on whether it is |
|
Sorry for the delay! I'm almost graduating so I have been neglecting the repo. |
| values = self.convert_and_round(data_row) | ||
|
|
||
| # get column names if numpy.ndarray or pandas df: | ||
| try: |
There was a problem hiding this comment.
Could we possibly check if the type is a pandas df rather than doing that via try / catch? It's a little awkward to have the most common use case be in an exception.
| else: | ||
| col_names = [] | ||
| if len(col_names) != 0: | ||
| if np.sum([1 if str(k).isdigit() else 0 for k in col_names]) != len(col_names): |
There was a problem hiding this comment.
I don't understand this line. Why is it meaningful if column names are all digits, and why does it matter if all columns have digit names? Also, what happens if this is False? values seem to be undefined then
When a row of a pandas dataframe is passed to an explainer tabular instance:
explainer.explain_instance(df[cols],...), error is thrown because they are not of type float.This PR handles it without problem by checking if columns are of type digits and solves the issue.