Uncertainty intervals of forecast are wrong when it goes negative.

[Dapid]

Uncertainty intervals of forecast never goes negative.

How to reproduce the bug

1 Take any data that includes negative values.
2 Run a forecast, including uncertainty interval

Expected results

I'd expect the uncertainty to be around the forecast.

Actual results

The uncertainty interval is clipped from below at 0, and behaves oddly.

Screenshots

Environment

(please complete the following information):

• browser type and version: Firefox 105.0.2
• superset version: 2.0.0
• python version: 3.8.14
• prophet version: 1.1.1
• any feature flags active: run as per the tutorial

Checklist

Make sure to follow these steps before submitting your issue - thank you!

• I have checked the superset logs for python stacktraces and included it here as text if there are any.
• I have reproduced the issue with at least the latest released version of superset.
• I have checked the issue tracker for the same issue and I haven't found one similar.

Additional context

Here is some data one can use. I used Date as the X axis and min(temperature) for Y (days, weeks, or months, work fine).
Note that using max(temperature) instead yields nice-looking results, since almost all of it is positive.

[ShaliniIruvuru]

Hi it will be very help for us to resolve this issue if you provide detailed information,

[rusackas]

I'm not sure if this is still an issue in current versions of Superset (3.x). If it is, we can re-open this, or feel free to open a new issue with updated context and a reproducible case using example data. We're no longer supporting Superset 2.x or prior, and it's been a while since this thread saw any activity, so I'm closing this as stale.

[Dapid]

The problem is still here in 3.1. You need any data with negative values.

For example, creating a line chart based on this query, and adding a forecast works as expected:

SELECT DATETIME(year, 'start of day') AS year,
       MIN(SP_POP_AG14_MA_IN) AS "MIN(SP_POP_AG14_MA_IN)"
FROM main.wb_health_population
GROUP BY DATETIME(year, 'start of day')
ORDER BY "MIN(SP_POP_AG14_MA_IN)" DESC
LIMIT 10000
OFFSET 0;

Changing the sign of the value yields instead:

And in fact, plotting MIN(SP_POP_AG14_MA_IN) - 170 works fine:

But plotting MIN(SP_POP_AG14_MA_IN) - 172 is odd again, even though there aren't any negative values.

[rusackas]

Ahh, thanks for the context! Re-opening!

[rusackas]

This issue has gone stale, but I'm hesitant to close it. Not sure if @villebro took a look at it and/or has any clues to go on.

[Dapid]

The problem remains on v4.0.2.

I can't find where the plot is actually being done, so I haven' been able to track down the problem further. But I have checked the raw data, and prophet is modelling it correctly, is just the display.

The issue doesn't have much activity, but the bug persists, so I think closing it is wrong.

[rogerlga]

I'm struggling with this issue as well. Any attention would be greatly appreciated.