Nous sommes actuellement le 26 Juin 2022, 06:43


[AI] [0047505] Bug Hillstat

Modérateur: Cyanide Team

  • Auteur
  • Message
Hors-ligne

suigfdhuibsyret

  • Messages: 11
  • Inscrit le: 05 Août 2013, 23:25

Re: Bug Hillstat

Message20 Juin 2017, 18:32

3emeType a écrit:
suigfdhuibsyret a écrit:just to prove our concerns a bit more, i just played liège-bastogne-liège on extreme. I altered dayer quintana's stats to 85 climb, 50 hill and decent stats around that. Won it first try. I'm uploading the stage as we speak. expect an edit when it's done uploading.

edit: https://www.youtube.com/watch?v=IopzAVbK2xY&


I believe that a combination of factors can explain your victory:
- strong back wind.
- strong cheated stats (flat/endurance/sprint/...)

It seems all your opponents are tired at the end. So when they need punch (their HIL stat), they can't use it, so the race is not judged on punch, so you can have no punch, it doesn't matter.

Could you do the same result without wind and cheated stats?

fragglerock a écrit:I won yesterday with froome in extreme, kwiatkowski 14º. If you need video, no problem.


Yes please. And even better: a replay.
I perform much better with Kwiatkowski so I really need you to understand how it can be the opposite. Thanks! ;)

Other points:
- I didn't say you can't win Liege with a pure climber, I said it would be much harder: with a very strong team, under more rare circumstances (back wind, ...), ...
- Liege has many kilometers uphill, like a mountain stage. If there is one "hilly" ace that can be won by a climber, that's this one. Can you win hilly races with less kilometers uphill with a pure climber?


does this count?
http://imgur.com/DoM6hkV
or this?
http://imgur.com/bVs02IZ

I'm uploading the whole stages again but it's gonna take a while.
All of these were first try too.

Look, it's fine to make a mistake but please don't act like it works like it should.

edit:
fleche wallon: https://www.youtube.com/watch?v=BdGSbK_Z45g
plouay: https://www.youtube.com/watch?v=JzyykCahfMw
Dernière édition par suigfdhuibsyret le 20 Juin 2017, 21:52, édité 1 fois au total.
Hors-ligne

Paul23

  • Messages: 10
  • Inscrit le: 24 Déc 2015, 22:32

Re: [AI] Bug Hillstat

Message20 Juin 2017, 18:41

To be fair, Lombardia is both. Hilly and mountainous. It's also a great goal for pure climbers.

But yeah, something is really wrong here.
Why not implement it that way, that you can set a value of HI/MO ratio for every stage and then for every climb in that stage. With that way, you could set the value f.e. at 0, which means that only the MO stat gets used, at 0.5, which is 50/50 and at 1 which is where only the HI stat gets used.(obviously everything in between as well)
With that you could stop guys with a good HI stat to be too strong in Mountain-stages and you could stop pure climbers to win at hilly races.


Also great to hear, that the sprints will get fixed.
Hors-ligne

suigfdhuibsyret

  • Messages: 11
  • Inscrit le: 05 Août 2013, 23:25

Re: [AI] Bug Hillstat

Message20 Juin 2017, 19:17

Paul23 a écrit:To be fair, Lombardia is both. Hilly and mountainous. It's also a great goal for pure climbers.

But yeah, something is really wrong here.
Why not implement it that way, that you can set a value of HI/MO ratio for every stage and then for every climb in that stage. With that way, you could set the value f.e. at 0, which means that only the MO stat gets used, at 0.5, which is 50/50 and at 1 which is where only the HI stat gets used.(obviously everything in between as well)
With that you could stop guys with a good HI stat to be too strong in Mountain-stages and you could stop pure climbers to win at hilly races.


Also great to hear, that the sprints will get fixed.


that is exactly how it worked up until pcm 2015
Hors-ligne

balaban

  • Messages: 20
  • Inscrit le: 21 Juin 2013, 08:53

Re: [AI] Bug Hillstat

Message20 Juin 2017, 19:36

suigfdhuibsyret a écrit:
Paul23 a écrit:To be fair, Lombardia is both. Hilly and mountainous. It's also a great goal for pure climbers.

But yeah, something is really wrong here.
Why not implement it that way, that you can set a value of HI/MO ratio for every stage and then for every climb in that stage. With that way, you could set the value f.e. at 0, which means that only the MO stat gets used, at 0.5, which is 50/50 and at 1 which is where only the HI stat gets used.(obviously everything in between as well)
With that you could stop guys with a good HI stat to be too strong in Mountain-stages and you could stop pure climbers to win at hilly races.


Also great to hear, that the sprints will get fixed.


that is exactly how it worked up until pcm 2015


paul23 has a good idea but he explained it very poorly.

He doesn't want the pcm 2015 system of setting a ratio for the stage.
He wants to be able to set different ratios for different climbs in the same stage.

That way you can set different ratios for the small climbs and for the long climbs.
Hors-ligne

Paul23

  • Messages: 10
  • Inscrit le: 24 Déc 2015, 22:32

Re: [AI] Bug Hillstat

Message20 Juin 2017, 19:58

balaban a écrit:
suigfdhuibsyret a écrit:
Paul23 a écrit:To be fair, Lombardia is both. Hilly and mountainous. It's also a great goal for pure climbers.

But yeah, something is really wrong here.
Why not implement it that way, that you can set a value of HI/MO ratio for every stage and then for every climb in that stage. With that way, you could set the value f.e. at 0, which means that only the MO stat gets used, at 0.5, which is 50/50 and at 1 which is where only the HI stat gets used.(obviously everything in between as well)
With that you could stop guys with a good HI stat to be too strong in Mountain-stages and you could stop pure climbers to win at hilly races.


Also great to hear, that the sprints will get fixed.


that is exactly how it worked up until pcm 2015


paul23 has a good idea but he explained it very poorly.

He doesn't want the pcm 2015 system of setting a ratio for the stage.
He wants to be able to set different ratios for different climbs in the same stage.

That way you can set different ratios for the small climbs and for the long climbs.


You got it! Thanks for clarifiying.
Hors-ligne

fragglerock

  • Messages: 28
  • Inscrit le: 18 Juin 2017, 20:30

Re: [AI] Bug Hillstat

Message21 Juin 2017, 08:43

After play a lot of stages i think the problem is smaller than I thought. Effectively climbers arrive at last 10 km always with more energy (more green, yellow and red bar), but in the last 10km even with more energy, the hill riders have better performance.

However, i´m still thinking this aspect needs a small adjustment. Is not very logical that climb riders arrive at the final of the stage with more energy than hill specialist.
Hors-ligne
Avatar de l’utilisateur

3emeType

Cyanide Team

  • Messages: 122
  • Inscrit le: 23 Fév 2009, 16:34

Re: Bug Hillstat

Message21 Juin 2017, 10:05

balaban a écrit:At 85 effort is it 100% the mountain stat?
At 86 effort is it 99% the mountain stat and 1% hill? or is it 100% hill? A what effort is it 100% hill, only at 99 effort?


- effort 85%: 100% mountain
- effort 99% or attack/sprint: 100% hill
- the more you approach 99%, the more hill is used over mountain.

balaban a écrit:Tailwind would make things easier and the gaps smaller because it makes rides faster so the climbs take less time to ride and so it's beneficial for the hill specialists (short climb specialists)


With tailwind, the race is harder on flat zones because you must ride faster without more protection from air resistance (on climbs there is less air resistance due to the low speed so tailwind has less impact). So everybody is more tired. So if you have a better flat stat and a better endurance, you are helped.

balaban a écrit:the endurance stat only has an effect in the last kms (unless you've changed that this year)


This changed. Last year I think. Now: more endurance -> green energy decreases slower.

balaban a écrit:Flat, resistance, sprint and acceleration can't explain the gap because the hill specialists like Gilbert have higher flat, sprint, acceleration and resistance than Dayer Quintana in that test


But the mountain stat is important on Liege if the race is fast. It's a whole.



You're videos don't work. It's a shame because your screenshots are interesting. I'll make some tests myself.
Do you have replays ? They contain more information. I prefer. I can move the camera where I want.

suigfdhuibsyret a écrit:Look, it's fine to make a mistake but please don't act like it works like it should.


I will have no problem admitting our error as soon as I have seen it. Meanwhile, I'm still finding it. Thank you for helping me.

balaban a écrit:He doesn't want the pcm 2015 system of setting a ratio for the stage.
He wants to be able to set different ratios for different climbs in the same stage.

That way you can set different ratios for the small climbs and for the long climbs.


So the speed of a rider on the first meter of a climb would depend on the length of the climb? This would not be very realistic :)
The performance of riders depend on the speed of the climb, not on its length: if the Mont Ventoux is climbed very slowly, Gilbert can win thanks to his punch (his HIL stat).

fragglerock a écrit:Is not very logical that climb riders arrive at the final of the stage with more energy than hill specialist.


What's a hill specialist? What can he do better than climbers? For us, a hill specialist is a rider that can ride very fast when he attacks in climbs (or make a very high effort).
Before the end of the stage, if there has been no attack, this is perfectly normal that, if there are lots of climbs, he loses more energy than good climbers.
Hors-ligne

suigfdhuibsyret

  • Messages: 11
  • Inscrit le: 05 Août 2013, 23:25

Re: [AI] Bug Hillstat

Message21 Juin 2017, 10:56

Oh sorry, i put the videos on private. They should work now. I dont have replays though. I'll think about them next time. Thx for replying
Hors-ligne

balaban

  • Messages: 20
  • Inscrit le: 21 Juin 2013, 08:53

Re: Bug Hillstat

Message21 Juin 2017, 11:27

3emeType a écrit:- effort 85%: 100% mountain
- effort 99% or attack/sprint: 100% hill
- the more you approach 99%, the more hill is used over mountain.


Thanks

3emeType a écrit:So the speed of a rider on the first meter of a climb would depend on the length of the climb? This would not be very realistic :)
The performance of riders depend on the speed of the climb, not on its length: if the Mont Ventoux is climbed very slowly, Gilbert can win thanks to his punch (his HIL stat).


I agree. I was just explaining what paul23 was suggesting.
Hors-ligne
Avatar de l’utilisateur

3emeType

Cyanide Team

  • Messages: 122
  • Inscrit le: 23 Fév 2009, 16:34

Re: [AI] Bug Hillstat

Message21 Juin 2017, 11:32

suigfdhuibsyret a écrit:Oh sorry, i put the videos on private. They should work now. I dont have replays though. I'll think about them next time. Thx for replying


Thx.

What I see here is more an exploit with trains than a problem with hill stat.
It looks like you've found a way of escaping with your team thanks to the train. Once you do that, you use your team to take time, and not your hill stat.
AI should better follow your train.

Another AI problem: on Plouay, Gilbert should have attacked you on the last climb. By doing that he would have won.

And with a correct sprint stat for modified Quintana D. (70 is impossible for a climber that has no punch. A climber can't be more powerful on flat than in climb, so you should give him 50 in sprint), you would have lose also. So, in both races, modified Quintana D. would have lose.

So we have work to do to improve the AI :)
Thanks. Very useful videos.
PrécédentSuivant

Retour vers Pro Cycling Manager 2017

Qui est en ligne ?

Utilisateur(s) parcourant actuellement ce forum : Google [Bot] et 3 invité(s)

cron