nikosbosse / Metaculus-Twitter-bot

17 stars 4 forks source link

fix: remove future data points & exclude question 10557 #29

Closed edomt closed 1 year ago

edomt commented 1 year ago

This PR fixes https://github.com/nikosbosse/Metaculus-Twitter-bot/issues/28.

Bad formatting

This will fix the bad formatting in the last few tweets of question 10557:

image

This is due to question 10557 having – for some reason – data points in the future:

time                            lower  middle  upper
2022-11-04 22:44:54.744327168   0.10   0.20    0.25
2022-11-06 21:23:29.295725312   0.10   0.20    0.29
2022-11-08 23:28:00.000000000   0.10   0.20    0.29
2023-12-30 17:00:00.000000000   0.10   0.20    0.29
2022-12-26 09:27:23.602747648   0.26   0.35    0.50
2023-12-30 17:00:00.000000000   0.10   0.20    0.29

Possible issue with question 10557

On top of this, there seems to be an issue with question 10557 specifically. As shown above, in the time series the recent swing in the forecast seems to always be attributed to the current day (2022-12-26, currently); when we know from the bot that this swing appeared first on 19 Dec. Because of this, everyday the bot sees a new swing and tweets an alert.

Combined with the first issue of data points in the future, my best guess is that there is an issue with question 105557 on Metaculus's end.

For now, I've added an excluded_questions section in the filters, and I've excluded Q10557 from the bot's list.

nikosbosse commented 1 year ago

Thanks so much, Edouard! I let the Metaculus Team know about the issue. Upon looking into it more closely I assume that this bug is related to the fact that 10557 is a subquestion of a question group. I will check with the Metaculus team whether

dschwarz26 commented 1 year ago

10557

We'll look into this, thanks for finding the data points in the future, that's pretty suspicious!

Regarding Twitter cards for subquestions: coming soon! We're redesigning our Twitter cards to support some new use cases too. I'd say rough ETA would be mid-Feb.