[Bug] ValueError caused by column with nan values

[tswsxk]

When using shapash along with the following codes:

xpl = SmartExplainer(
    model=model,
)

xpl.compile(
    x=test_df,
    ...
)

it will call the init_app in SmartApp class where the following codes are used to calculate the std of a certain column (line 192 - 199):

        for col in list(self.dataframe.columns):
            typ = self.dataframe[col].dtype
            if typ == float:
                std = self.dataframe[col].std()
                if std != 0:
                    digit = max(round(log10(1 / std) + 1) + 2, 0)
                    self.round_dataframe[col] = self.dataframe[col].map(f"{{:.{digit}f}}".format).astype(float)

However, when a column with nan values, std will be nan and execute the following line:

                    digit = max(round(log10(1 / std) + 1) + 2, 0)

and result in ValueError:

File "xxx/.local/lib/python3.10/site-packages/shapash/webapp/smart_app.py", line 197, in init_data
    digit = max(round(log10(1 / std) + 1) + 2, 0)
ValueError: cannot convert float NaN to integer

Python version : python3.10

Shapash version : shapash-2.5.0

Operating System : CentOS Linux release 8.2.2.2004

[guillaume-vignal]

Thank you to report this issue, if I understand correctly you should have only nan values in your columns, is it right ? Actually the code should not bug if you have a column with just some nan values in it. Can you tell us why you have such a column used in your model ? We don't see the use case. In any case we will look at your PR and see the best way to tackle this issue.

[tswsxk]

Hi, there. Usually, the nan columns could be removed during data preprocessing. However, in my case, the data is several time series and one column contains only nan values before a specific date. Thus, when I conduct experiments, where I need to split the data according to dates for training. However, the all-nan values in this column before a specific date cause this error.

[guillaume-vignal]

It has been fixed with the version 2.6.0 of shapash (https://github.com/MAIF/shapash/pull/553)