Issue with Timestamp Comparison in dt_transform

[mfquintero]

I'm getting an issue running robynpy since the current implementation converts the dates to a formatted string (YYYY-MM-DD) using dt.strftime() in the _prepare_data method of the FeatureEngineering class, which results in dt_transform["ds"] being of type string. This causes issues when comparing dt_transform["ds"] with the window_start in the _create_rolling_window_data method of the same class.

[rora00]

Where do you see the implementation of dt.strftime in the FeatureEngineering class? It looks like the implementation in the current version of RobynPy 0.1.2 uses pd.to_datetime() to convert the date column from str to datetime so that "ds" can be compared with "window_start" and "window_end". The tutorial notebook uses a dataset in the format you specified of (YYYY-MM-DD) as a str which gets converted successfully to a datetime object which then gets used in the feature engineering step.

Could you provide an example of the date column in your dataset which may be causing a potential faulty datetime conversion to an object instead of datetime? Potentially the limitation of pd.to_datetime itself is that it does not handle two scenarios: (1) Where dates fall outside of the permitted range or (2) Have a mix between timezone naive and aware datetimes.