Not sure the right place for this thread but figured here made sense. I have been following the discussions in various race threads about PCM 18 AI and thought I would post some stats.
One area of discussion has been poor performance of the pure puncheurs as Mountain stat becomes important for extended climing. In one thread Tamijo theorized that the "PCM 18 is too realistic for the man game DB".
I thought that was an interesting and probably valid. And since I compiled all the PCM DBs from 2011 and 2018 into a single table a while back I decided to compare that to the Man Game Db.
I don't think it is a surprise that the Man game DB doesn't mirror reality but if that difference is significant then you wouldn't expect the AI to reward the lack of reality. Or said another way a reality-based AI shouldn't reward unrealistic stat combos.
Each point in the graphic below represents a combination of MT and HI stat for a rider season. Limited to those with HI 77+. The size of the point indicates the number of rider seasons, so bigger points are more important. The Man game data is offset by 0.5 on MT so you can compare (so 77.5 = 77).
I think this shows that the man game MT stats for high HI riders are too low and the high HI low MT rider is way over-represented. So to Tamijo's point if those riders don't really exist IRL then the AI shouldn't have them winning Pais Vasco. To put a number to it, in the PCM DB only 6% of riders with Hill >77 had MT <70, in Man game that is 38%.
This isn't to say managers with these riders shouldn't be upset about the erosion of their investment. But I don't think it is a flaw in the AI.
For the record the only riders in the PCM DB with hill > 77 MT < 70 are below (with the seasons this applied). I don't think these riders are typically expected to do well in mid-mountain races like Pais Vasco, PN or Catalonia. Also while I think there are fewer of these in more recent seasons when the AI punishes them, there still weren't that many prior to the switch in the application of the MT/HI stat.
One other statistic. I have seen some concern about the "randomness" of sprints. In response, I think Fabianski raised a good point about the number of good sprinters in the DB as a cause. To put this in perspective I counted the number of 80+ sprinters in the entire PCM DB each year. It looks like this.
Riders w. > 80 Sprint
2011
6
2012
3
2013
3
2014
7
2015
6
2016
5
2017
6
2018
7
Given this when there are 25 80+ sprinters in a single race I think it is hard to expect the AI to handle it. Although perhaps random outcomes make sense when you have such a large number of good sprinters.
This isn't a criticism of the man game DB, recognize that is a result of years of decisions and growth. Just think we have to be careful in criticizing the AI for factors that may result from the nature of the DB.
Ulrich Ulriksen wrote:
This isn't a criticism of the man game DB, recognize that is a result of years of decisions and growth. Just think we have to be careful in criticizing the AI for factors that may result from the nature of the DB.
I'm not sure if anyone is directly criticizing the game's AI but rather just the way it works with the Man-Game DB. But I also think frustrations come from the facts that we could have tried to forseen this a bit more idk.
If we could have redone the off season I think we should have done this: (Hindsight is 20-20 of course and this isn't a critism, but rather an idea for the plan) ~also going to be using personal examples but I'm sure I'm not the only one who experienced these things and I also know theres a lot of season left so please don't take any of this definitive~
Option 1: Keep the Man-Game DB exactly the same including OVL calculation/renewal calculation b/c we aren't exactly sure how the AI is gonna work, but announce we are switching games at least 1 season ahead of time. That way people investing in riders would get a better inkling on their riders' long-term value.
For example: I never would have paid the ridiculous price I paid for Coppel last off-season knowing he wouldn't be a top 20 rider this year, which he most likely would have been had we not switched games as seen by his riding last season. At the time I definitely thought it was worth it considering he was barely decling but his lack of mountain is crippling in this new game, which I would have been able to forsee or even plan around but given the investment (1Mil+High wages) put in on him losing him for nothing this off-season was not an option to me at least (maybe that's just on me).
Option 2: Since we didn't give a full season's heads up, super in depth testing to re-do the OVL calculation/renewal calculation. I could be completely wrong, but it seemed to me that renewals were not affected at all by the fact that the game was changing.
For example: Knowing we were switching games, I bought PCM 18 and played it a bit myself and got the vibe that Coppel wouldn't be as effective in the new game which led me to low-ball him in renewals (offered him 400k, which I thought made sense to me as his real value now was as a pure TimeTrialist). So I'm not saying that he should have taken that offer specifically, but I was a bit salty and tried to understand and was given the reasoning that he could climb a bit, which although true doesn't matter much in this game considering his super low MTN stat. That said though I did end up going after other cheap-ish puncheurs w/ low mountain stat as well so I can't say I 100% expected him (Coppel) to be completely useless in the hills.
Overall I don't think the issues stem from the game itself, and I think your point about the "randomness" of sprints is definitely valid. There's just too many sprinters that are in the 80s in the important stats for there to be a lot of consistency, which I'm totally okay with. But I feel like we should have potentially tested that out more and curved sprinters' renewal costs as well to value them less than in previous years.
Again, I know all this is 20-20 in hindsight, but next time we switch games I really think the process should be taken care of extremely carefully either in terms of curving renewals correctly or just giving a super-advance notice to everyone and have the managers have to deal with it themselves.
I will also say I apologize if I come off as complaining, definitely don't want to come off that way and it just happens that I think I'll probably be getting the short end of the stick (which is okay, someone was bound to). I also will totally confess to not being super in the loop at times during the man-game so if there's things I missed like the renewals changing a such, please let me know. Definitely not trying to create any false narratives or come at anyone at all.
I would agree testing was insufficient (or the results weren't communicated well enough), but with the huge delay we already had there really was no other option. You win some (Senni! <3), you lose some (pretty much all my sprinters). At this point all we can do is finish the season and hope for the best.
I also agree with the theory the MG DB seems to be the problem instead of AI. One dimensional riders are punished (which is a good thing on its own!), but the hill specialists are in trouble. For the sprints I 100% agree the problem may be too many 80+ sprinters in the game.
And then there's problems we just created like putting all cobbles to 5 stars, but they can be fixed easily next season.
I too would like to stress I'm not criticising anyone, but these are definitely things we need to consider for next season.
Ulrich Ulriksen wrote:
For the record the only riders in the PCM DB with hill > 77 MT < 70 are below (with the seasons this applied). I don't think these riders are typically expected to do well in mid-mountain races like Pais Vasco, PN or Catalonia. Also while I think there are fewer of these in more recent seasons when the AI punishes them, there still weren't that many prior to the switch in the application of the MT/HI stat.
Gilbert
11 to 15
Iglinskiy
12 to 13
Gasparotto
12 to 15
Gerrans
12 to 17
Simon
18
Van Avermaet
11
Geschke
14
Leukemans
11
Albasini
14 to 15
Matthews
15
Valgren
16
You forgot about my rider, Bystrøm.
79 hill and mt < 70
My suggesting would be to make and update to the database - effective from April races - to save the season for most "hilly" managers
Where riders would get an MO update relative to their OVL/HILL values so that good riders especially those with high HILL, would be at least able to compete on a reasonable level. This would not only compensate riders price tag, but would make the PCM18 AI show how good it really is, providing for some interesting and nice reports.
I believe that the sprint AI is just as good as in PCM15, just a different way of handling it, but all i all similar (or better) results. We must not forget that PCM15 was far from perfect.
Edited by Tamijo on 10-01-2020 15:12
How can that possibly be made fair? I have Koretzky so obviously I would like to see him get more MO, but would he be as strong in relation to before? And what about Coquard? Will he get better MO too? Because last season he did fine in bumpy races.
When we knew we were changing to PCM18, we knew that MO HL was going to work differently, and that it would hurt low MO punchers. And that the MG DB was full of uniquly stated riders that most likely would not be as good on the new game version.
Trying to change things now, mid-season, will only end badly. I guarantee we'd get the adjustments wrong.
We'd also be massively disadvantaging those managers who planned accordingly. Either those who worked hard to get/make/keep balance MO punchers, to move away from that terrain.
It's part of the reason i entirely dumped my climbers and punchers, to go after more predicatble sprints, tt's and classics (that turned out well...).
______
What Ulrich data shows is kind of unrelated to the AI. It is a strong showing of what we've all known for several years, the MG DB is massively over inflated, to the point now where i think PCM can no longer run the game effectively.
PCM15 AI was struggling, but just about making it work because of the way the stat matrix worked in the game at the time. The huge shift from 15 to 18 in the way the game fundamentally works and runs is now kind of just broken. I will say it still seems to be working decently for Cobbled races, and for TT's (areas in the DB that are either less inflated, or where stat inflationg does not affect the games AI). I'm seeing flashes of the improved PCM18 AI as it tries to play thngs out more realistically, but ultimately it is falling short more often that not.
Whether that is solely down to PCM18, or whether the DB finally hit the tipping point of inflation, i can't say (would need to run the 2019 season on 15, and the 2018 season on 18 to figure that out.
But either way we need a serious and deep converstation about what version is used for 2020, or about tackling the inflation problem. Real radical idea, maybe it's time for a soft-reset on the game and wipe the DB back to something more in line with the normal Daily/RSM DB a la when the MG started... very radical suggestion but.....
Tamijo wrote:
My suggesting would be to make and update to the database - effective from April races - to save the season for most "hilly" managers
Even as someone super disadvantages by the mtn-hill thing, i think this isn't really a good idea.
I think the only course of action currently i probably to just plan the future better as TMM said in terms of looking at the inflation and such while also running tests on the DB and AI in general. It's unfortunately just tough luck for us Hill-Managers tbh
In general I actually think the AI in PCM 2018 is pretty solid and the races have been pretty dope with all the time-gaps which is more realistic. But more realistic doesnt necessarily mean good for the Man-Game of course
Tamijo wrote:
My suggesting would be to make and update to the database - effective from April races - to save the season for most "hilly" managers
Where riders would get an MO update relative to their OVL/HILL values so that good riders especially those with high HILL, would be at least able to compete on a reasonable level. This would not only compensate riders price tag, but would make the PCM18 AI show how good it really is, providing for some interesting and nice reports.
Editing the DB midseason is the worst thing we can do - as much as e.g. my team could be hurt by this edition. There is absolutely no way to make it fair for everyone or even most. For now there is nothing we can do - except watching the situation and discussing what we will do next season.
@ollfahrd: I dont think testing was not communicated properly because imo it is up to the individual manager to talk to guys with the game. I do agree that there might not be enough testing done with the MG DB.
PCM 16-17 were skipped because the games were deemed to be absolutely not compatible with the MG DB iirc. There was talk about this version being better because of lowered thresholds for Mo/hi use and strong AI that would still recognize pure puncheurs as favourites on hill stages and thus help them to good results. It's still too early in the season to see if that is indeed the case with the cases weve seen so far being mostly outliers (due to too hard profiles or whatever) or if that was a massive missevaluation based on only knowing AI with the normal AI.
As much as i hate on the game so far, we must not forget that the game was indeed able to let a rider like Skujins finish top 10 in Grand Duche or Badaling so we should at least give the game the benefit of the doubt for now until we see how the ardennes shake out. If he continues like that in other pure hill classics then we can at least continue to see hill as a main stat which absolutely is necessary imo to continue the MG with this game version. Eventhough i would like to keep the idea of returning to pcm15 in the back of the mind if the rest of the season disappoints in that part.
Spoiler
skujins only used as the one example where i knew the results and didnt have to look things up
Well my team is not affected as I have a MO/SPRINT combo, so I'm fine, but we see absurd results where riders like Coppel loses 20 min of a semi-hard hill stage, and can expect the same from Bobridge, Claeys, Ginanni, Di Maggio in similar races later if we continue without changes. that is results that does not justify the price tag on those riders. As a result we won't have any fun this season and every discussion tread will be full of "game related" critics, mostly justified.
I think that will be awful and also very sad for those investing time into those reports.
what about Coquard? Will he get better MO too?
Yes off course he would, anyone with a relative high OVL compared to MO would benefit, as the general issue is that a MO at 60 would be ok in most stages at PCM15, but in PCM18 the developers realized that no modern top riders (whatever type) is useless when it comes to climbing.
Agree with what Manx has said. I've been against moving to PCM 18 and obviously people put in the testing and somewhere along the line it was determined that PCM 18 would be alright for the MG, but I always had strong feelings that it was not going to work out well with the stat matrix and the new MO/HIL AI.
With that said, I planned for it, and atleast tried to kind of prepare for it. To make any midseason changes is first of all completely unfair, and quite a knee-jerk reaction to a problem im not sure anyone will be able to fix in a way that is fair to everyone, i.e apart from changing OVLs theres not much we can do for now.
Maybe in the off-season we can introduce a form of training similar to how if the riders stat is a certain amount below their ovl they can train for a discount fee, but instead make the relation between Hill and MO. And make it a max of 2 upgrades as well.
I remember for EPIC, at some point we switched from PCM12 to 13 or something and the earlier version had no working RES stat. It was there, but it didn't influence the game. No one had trained it so far because it had no effect, so I gave everyone a small boost. Maybe we can do something similar? If the gap between MO and Hi is 6-10, add 1MO, 11-15 add 2, 15+ add 3.
Sprint might be fixed in time with less top riders being added to the game. MO is the biggest issue imo.
there's an 'easier' solution. 'easier' because it takes more time but it's probably better.
And that is to adjust the stages, not the stats. Essentially make everything steeper, flat stages, hill stages, mountain stages, just higher and longer peaks.
edit: we all knew this was gonna happen, but I thought that stages were edited already, or were they not?
quadsas wrote:
there's an 'easier' solution. 'easier' because it takes more time but it's probably better.
And that is to adjust the stages, not the stats. Essentially make everything steeper, flat stages, hill stages, mountain stages, just higher and longer peaks.
What? Make them higher and longer woundent that make things worse, not to mention that a DB update would be quite quick, updating every stage would be several 100's of workhours. fixing a stage even the slightest takes time, you change a single Hill and sprints move, KoM move, then rebuild citys,
A few stages was changes this year (preparing for 2020) just to fix bad sprints and even a few extreme climbs and it always takes time, with 100s of stages would be impossible unless 20+ managers would be a part of it, something i doubt.
quadsas wrote:
there's an 'easier' solution. 'easier' because it takes more time but it's probably better.
And that is to adjust the stages, not the stats. Essentially make everything steeper, flat stages, hill stages, mountain stages, just higher and longer peaks.
What? Make them higher and longer woundent that make things worse, not to mention that a DB update would be quite quick, updating every stage would be several 100's of workhours. fixing a stage even the slightest takes time, you change a single Hill and sprints move, KoM move, then rebuild citys,
A few stages was changes this year (preparing for 2020) just to fix bad sprints and even a few extreme climbs and it always takes time, with 100s of stages would be impossible unless 20+ managers would be a part of it, something i doubt.
from playing a lot of PCM18 I felt like steeper hills actually help punchers and not the climbers. Do you not remember what someone did with Strade Bianche? They just made everything harder and it worked much better, not a massive bunch sprint finish
quadsas wrote:
there's an 'easier' solution. 'easier' because it takes more time but it's probably better.
And that is to adjust the stages, not the stats. Essentially make everything steeper, flat stages, hill stages, mountain stages, just higher and longer peaks.
What? Make them higher and longer woundent that make things worse, not to mention that a DB update would be quite quick, updating every stage would be several 100's of workhours. fixing a stage even the slightest takes time, you change a single Hill and sprints move, KoM move, then rebuild citys,
A few stages was changes this year (preparing for 2020) just to fix bad sprints and even a few extreme climbs and it always takes time, with 100s of stages would be impossible unless 20+ managers would be a part of it, something i doubt.
from playing a lot of PCM18 I felt like steeper hills actually help punchers and not the climbers. Do you not remember what someone did with Strade Bianche? They just made everything harder and it worked much better, not a massive bunch sprint finish
That might be right in strade and help punchers over sprinters, but in mg we mainly got series of too long climbs, that will favour MO, because they were set to "hill" even if they are (also in PCM15) actual semi-mountain stages (approx 50/50) you can't set it to Hill anymore so they will kill low MO punchers every time. As we have seen many times. Same with hard cob staged, if the rider has no MO it will be tough to pass many longer climbs as you can only use Hill when riding close to max, the pack only does than for a limited time especially if also suffering on the cobs.
So in regards to what changes if any should be made, the way the PCM.daily DB adjusted to the change in mechanics was an adjustment to RES. Based on a riders MO/HI ratio, the riders with a high MO, low HI got a reduction to their res, while the high HI, low MO got a boost. This worked out really well as it gives some of the better puncheurs enough energy to survive smaller climbs with the climbers but not enough to be competitive on mountain routes.
IIRC we did an adjustment of +- 0.33 RES per 1 difference:
IE a rider with 71 MO, 80 HL, 75 RS would end up with 78 RS
rider with 84 MO, 74 HL, 75 RS ends up at 72 RS (rounding was done normally).
I did check if this would exceed 85 anywhere and it wouldn't, Kiril Yatsevich would be at exactly 85 RES, Floris Gerts would be 2nd with 83. The biggest boost comes to the riders that excel at both (like Spilak, who would only lose 1 RS compared to the 2 or 3 lost by other top climbers). And concerns about the fairness of it.
![imgur.com/KlJUqDx.jpg[/img[img]]https://imgur.com/yUygrQ.jpg](https://imgur.com/KlJUqDx.jpg[/img[img]]https://imgur.com/yUygrQ.jpg)
