New windy thresholds?

[ajstewart]

The thread here that popped up when the weather was broken a few days ago: https://old.reddit.com/r/TheSilphRoad/comments/aks0k7/weather_gone_germany/

Suggests a windy threshold of:

Windy Override means that sufficient wind (wind speed over 24.1 km/h or wind gust speed over 31.5) and regardless of what the forecast is, it'll be windy in game. If it's a precipitation type (rain, snow) it'll never be windy, but you'll see the wind effect.

That post lead me down a rabbit hole of attempting to predict our weather and a google search lead me to your bot! Nice work!

[5310]

Thank you! And also thanks for the new resource.

Since DST started our instance of this bot has been pretty inaccurate, which makes us suspect that the forecast hours might have changed, despite us thinking the forecast hours were local to the timezone. I've been trying out some new models, and this'll help.

It's just that we get wind very very rarely around here, so I'll be a while before I can be sure of any new wind models.

[ajstewart]

No problem, that excellent resource put me on this path (there are some other comments in there about rain as well).

We get windy a lot here, I could test out the wind thresholding suggested in that thread. I'd probably need a little help coding it though, still working my way around js.

I'm currently in every hour mode to try and nail down our weather pull times.

[5310]

I actually just pushed th0rnleaf's wind model a while ago <8d317e2#diff-a66005028d1762c28eeaeaedfb75fbd5R71> ! Apparently, the old model was predicting (with the wrong pull hours, and no corroboration in-game) that one of our cells would get superficial winds. Hopefully it picks up later in the week.

…

On Thu, Jan 31, 2019 at 11:21 AM Adam Stewart ***@***.***> wrote: No problem, that excellent resource put me on this path (there are some other comments in there about rain as well). We get windy a lot here, I could test out the wind thresholding suggested in that thread. I'd probably need a little help coding it though, still working my way around js. I'm currently in every hour mode to try and nail down our weather pull times. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAG5EjOJQt6MeF0wpkWW7pCMLn5PvtXCks5vIoRzgaJpZM4abVXu> .

[ajstewart]

Ah cool, ok my js is good enough that I matched your commit with my own change.

We actually have windy weather now! I will keep monitoring it.

[5310]

Lucky you. And thanks again.

…

On Thu, Jan 31, 2019 at 12:27 PM Adam Stewart ***@***.***> wrote: Ah cool, ok my js is good enough that I matched your commit with my own change. We actually have windy weather now! I will keep monitoring it. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAG5Et7kpb7sSSKbrtHrzqWoMT_BobDBks5vIpPFgaJpZM4abVXu> .

[ajstewart]

Looking at your code and the model proposed by th0rnleaf, am I right in thinking that currently wind and rain override is possibly being applied to too many accuweather labels?

E.g. I just had a forecast of mostly cloudy w/ thunderstorms but the bot has interpreted as rain. Though th0rnleaf model suggests this should be approached as cloudy (of course ignoring that the model might not be accurate, seems to be the best there is though).

[5310]

Are you sure your current forecast comes from the most recent pull hour? Or at least hasn't changed since the last pull hour?

When our instance of the bot was still running back in summer other than superficial rain and wind we actually never had it fail for at least the first two or three months the members were submitting observations. Which is to say, simple maps from AW forecast label to PoGO label was always correct, (and we get a lot of rain in summer too.) Up until the time we were able to decide on our pull hour we had a lot of suspicious false positive hours where most of them matched but then one hour out of a week would maybe diverge like it does in your case.

Back in summer our pull hours were (fo' sho') midnight, 8am, noon, and 4pm. Say at 1pm the current fresh forecast for 2pm was cloudy, but it was actually rainy back on the noon forecast, we'd still get rain at 2.

[ajstewart]

Sorry I didn’t explain properly - it almost certainly isn’t the pull hour, I’m purely talking about how it interprets what it sees if it *was* the pull hour. So mostlycloudy with thunderstorm apparently should be cloudy according to that new model with no rain or wind override. That’s what I mean, I’m talking about the PoGo model regardless if it’s the pull hour or not.

…

On 31 Jan 2019, at 18:49, Sayantan Chaudhuri ***@***.***> wrote: Are you sure your current forecast comes from the most recent pull hour? Or at least hasn't changed since the last pull hour? When our instance of the bot was still running back in summer other than superficial rain and wind we actually never had it fail for at least the first two or three months the members were submitting observations. Which is to say, simple maps from AW forecast label to PoGO label was always correct, (and we get a lot of rain in summer too.) Up until the time we were able to decide on our pull hour we had a lot of suspicious false positive hours where most of them matched but then one hour out of a week would maybe diverge like it does in your case. Back in summer our pull hours were (fo' sho') midnight, 8am, noon, and 4pm. Say at 1pm the current fresh forecast for 2pm was cloudy, but it was actually rainy back on the noon forecast, we'd still get rain at 2. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[5310]

Ah, then it must indeed have been an override. The model the bot follows considers rain snow fog and wind to all act as overrides if extreme enough. In this case, the chance of rain might've been above 60%.

[5310]

Perhaps I should consider testing a more simplified model. The code is modular enough.

[ajstewart]

I would try following the model in his spreadsheet (https://docs.google.com/spreadsheets/d/1v51qbI1egh6eBTk-NTaRy3Qlx2Y2v9kDYqmvHlmntJE/edit#gid=0) so the code would need some sort of filter on if the AccuWeather label is non-override one (for example = mostlycloudywtstorms) then do not apply wind or rain override.

I have a candidate pull hour that is still going strong and the only hour it got it wrong was the one above, but it would have been correct following th0rnleaf model, in that case cloudy, which it was in-game.

[5310]

I'd actually already started implementing th0rnleaf's model from scratch before decided to save time by making the frankenmodel... I'll split them apart later today and ping this thread.

[5310]

(Glitch might have a glitch with commits containing apostrophe's...)

I've pushed th0rnleaf's model here: https://github.com/5310/discord-bot-castform/blob/6df2aadbb13d6e01eefbceef6391e6efb435eed0/pogo-th0rnleaf.js

[ajstewart]

Cool thanks for doing it so fast! I'll pull this now ready for the next hour. Still a promising run going either 3pm or 4pm local time forecast when you apply the th0rnleaf model, which is around the time you mentioned interestingly enough. Just to note I'm also pulling the forecasts at 15 past the hour.

I'll report back with how I get on.

[5310]

I for one hope it's not 4pm. Our 4pms don't match any of the first three hours and sometimes even later hours too! :D

…

On Thu, Jan 31, 2019 at 4:59 PM Adam Stewart ***@***.***> wrote: Cool thanks for doing it so fast! I'll pull this now ready for the next hour. Still a promising run going either 3pm or 4pm local time forecast when you apply the th0rnleaf model, which is around the time you mentioned interestingly enough. Just to note I'm also pulling the forecasts at 15 past the hour. I'll report back with how I get on. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAG5EqLFrCPEpqqQgFoPi_5VUAu99PIKks5vItN-gaJpZM4abVXu> .