nixtla
nixtla copied to clipboard
[FIX] Same output type as input in timegpt forecast function
When calling the API, the output of timegpt for the time column always uses the format "yyyy-mm-dd hh:mm:ss", even when the initial timestamps don't go to the hour-minute-second level.
When using the SDK, the timegpt.forecast
function returns dates in the same format as the input. Hence, if the time column doesn't include the hour, minutes and seconds, then the output doesn't include them either.
Here's an example using the AirPassengers dataset
timestamp API | timestamp SDK | TimeGPT |
---|---|---|
1961-01-01 00:00:00 | 1961-01-01 | 437.83 |
1961-02-01 00:00:00 | 1961-02-01 | 426.06 |
1961-03-01 00:00:00 | 1961-03-01 | 463.11 |
This issue is relevant because it would be better to be consistent in the API and the SDK. I also think that including the hour, minutes, and seconds could be burdensome for users that don't need this level of precision.