Altair: "TypeError: data type not understood" with dtype: period[M]

[tehfink]

Thanks for a great project! With these versions:

• altair: 4.0.1
• pandas: 0.25.3
• numpy: 1.18.1
• Python: 3.8.1

I get the following error when trying to pass a dataframe with a period to altair: TypeError: data type not understood

The error disappears when the 'Sale month' column is not passed to altair.

sale['Sale month'] = pd.to_datetime(sale['Sale month'], format='%b-%y').dt.to_period('M')

sale['Sale month']
0        2011-07
1        2011-08
2        2011-09
3        2011-10
4        2011-11
          ...   
15845    2013-11
15846    2013-12
15847    2014-01
15848    2014-02
15849    2014-03
Name: Sale month, Length: 1173, dtype: period[M]

alt.Chart(sale[['Value', 'Sale month']].head()).mark_bar().encode(
)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~/Documents/…/virtualenv/lib/python3.8/site-packages/altair/vegalite/v4/api.py in to_dict(self, *args, **kwargs)
    353         copy = self.copy(deep=False)
    354         original_data = getattr(copy, 'data', Undefined)
--> 355         copy.data = _prepare_data(original_data, context)
    356 
    357         if original_data is not Undefined:

~/Documents/…/virtualenv/lib/python3.8/site-packages/altair/vegalite/v4/api.py in _prepare_data(data, context)
     82     # convert dataframes  or objects with __geo_interface__ to dict
     83     if isinstance(data, pd.DataFrame) or hasattr(data, '__geo_interface__'):
---> 84         data = pipe(data, data_transformers.get())
     85 
     86     # convert string input to a URLData

~/Documents/…/virtualenv/lib/python3.8/site-packages/toolz/functoolz.py in pipe(data, *funcs)
    632     """
    633     for func in funcs:
--> 634         data = func(data)
    635     return data
    636 

~/Documents/…/virtualenv/lib/python3.8/site-packages/toolz/functoolz.py in __call__(self, *args, **kwargs)
    301     def __call__(self, *args, **kwargs):
    302         try:
--> 303             return self._partial(*args, **kwargs)
    304         except TypeError as exc:
    305             if self._should_curry(args, kwargs, exc):

~/Documents/…/virtualenv/lib/python3.8/site-packages/altair/vegalite/data.py in default_data_transformer(data, max_rows)
     11 @curry
     12 def default_data_transformer(data, max_rows=5000):
---> 13     return pipe(data, limit_rows(max_rows=max_rows), to_values)
     14 
     15 

~/Documents/…/virtualenv/lib/python3.8/site-packages/toolz/functoolz.py in pipe(data, *funcs)
    632     """
    633     for func in funcs:
--> 634         data = func(data)
    635     return data
    636 

~/Documents/…/virtualenv/lib/python3.8/site-packages/toolz/functoolz.py in __call__(self, *args, **kwargs)
    301     def __call__(self, *args, **kwargs):
    302         try:
--> 303             return self._partial(*args, **kwargs)
    304         except TypeError as exc:
    305             if self._should_curry(args, kwargs, exc):

~/Documents/…/virtualenv/lib/python3.8/site-packages/altair/utils/data.py in to_values(data)
    138         return {'values': data}
    139     elif isinstance(data, pd.DataFrame):
--> 140         data = sanitize_dataframe(data)
    141         return {'values': data.to_dict(orient='records')}
    142     elif isinstance(data, dict):

~/Documents/…/virtualenv/lib/python3.8/site-packages/altair/utils/core.py in sanitize_dataframe(df)
    221             # otherwise it will give an error on np.issubdtype(dtype, np.integer)
    222             continue
--> 223         elif np.issubdtype(dtype, np.integer):
    224             # convert integers to objects; np.int is not JSON serializable
    225             df[col_name] = df[col_name].astype(object)

~/Documents/…/virtualenv/lib/python3.8/site-packages/numpy/core/numerictypes.py in issubdtype(arg1, arg2)
    391     """
    392     if not issubclass_(arg1, generic):
--> 393         arg1 = dtype(arg1).type
    394     if not issubclass_(arg2, generic):
    395         arg2_orig = arg2

TypeError: data type not understood

[jakevdp]

The sanitization code needs to be updated to recognize period types. That said, Vega-Lite has no data type representing time periods, so I think the best possible outcome would probably be a more informative error.

[jdewanwala]

Hi Jake, any idea of when this issue will be resolved? Still getting the TypeError: data type not understood error when PeriodIndex is present in the dataframe. Thank you!

[jakevdp]

It will be resolved when someone fixes it. Are you interested?

[jakevdp]

The relevant code that would need to be modified is here: https://github.com/altair-viz/altair/blob/03f37ca8b2701bf8d9e61646cacf8eee2edcad87/altair/utils/core.py#L243-L350

That said, it's unclear what is the best way to represent time periods in Vega-Lite, which doesn't have the concept of time periods. I suppose it could be represented as timestamps with respect to some standard start time, such as 2012-01-01 00:00:00

[jakevdp]

Quick question on this: how would you expect your period columns to be represented within the Vega-Lite chart? Should it be a timestamp? A string? Silently ignored?

[jdewanwala]

Hi Jake, "I suppose it could be represented as timestamps with respect to some standard start time, such as 2012-01-01 00:00:00" - that would be my thought.

Could the sanitization code be used to convert periods into timestamps before they get passed to Vega-Lite?

And say if PeriodIndex has the frequency of "M" (Monthly), perhaps there is an implicit conversion to 'month(period_column)' in the Altair code? Not sure what's the best way to do this second part. Alternately, the easy way out may be for the user to explicitly specify 'month(period_column)', knowing that the period_column has been cast to time_stamp through sanitization.

Let me know your thoughts. Thank you.

[jakevdp]

We don't have any mechanism at the moment for data sanitization to affect the encoding specification, and I think we should probably avoid making those kinds of assumptions for the user. Silent changes of defaultsthat are convenient in some situations often lead to a lot of confusion in others.