BUG: Fix to_json when converting Period column by colonesej · Pull Request #32665 · pandas-dev/pandas (original) (raw)

I made this roundtrip test completely based on the one for a time series with a DatetimeIndex. Some points:

• I could not find any option to make read_json treat the data and generate a PeriodIndex when orient equals to "column" or "index". Conversion does not carry frequency information. I'm not sure what exactly should be the behaviour so just converted manually to Period and inferred freq is OK.
• When orient = "split", the test fails as read_json cannot convert back the dict created by to_json. It works with a DatetimeIndex, which conversion results in float formatted dates. With Period it converts to separated keys for different aspects of Period object. I could not find where these were determined nor what should be preferred.

To illustrate, using series.to_json(orient='split') when index is:

DatetimeIndex:

'{"name":null,"index":[946857600000,946944000000,947030400000,947116800000,947203200000,947462400000,947548800000,947635200000,947721600000,947808000000,948067200000,948153600000,948240000000,948326400000,948412800000,948672000000,948758400000,948844800000,948931200000,949017600000,949276800000,949363200000,949449600000,949536000000,949622400000,949881600000,949968000000,950054400000,950140800000,950227200000],"data":[-0.7273489893,0.43763153,0.0064352361,-1.0478282669,0.6605651751,0.2847295244,1.6006677017,-1.2731074841,0.3004426454,0.274394972,0.0402248331,1.5284606656,0.5741635466,-0.3126890855,1.4765392606,0.3663030854,0.3132193099,-0.0730319057,-0.3956868944,1.4626210011,0.2129331654,-0.6671879793,-1.6477379639,0.7626651704,-1.7852083837,0.0707591559,1.3089143648,-1.0282706187,-0.7853299219,-0.2119391625]}'

PeriodIndex:

'{"name":null,"index":[{"day":3,"dayofyear":3,"daysinmonth":31,"freqstr":"B","is_leap_year":true,"month":1,"ordinal":7827,"qyear":2000,"start_time":946857600000,"week":1,"weekofyear":1},