ai stats weight

[endgame]

Interesting element. The exactness of a bar's setting is not so . . . exact . . . and there is no bar for Player Ratings.

::Note: This is the display currently under a 100/0/0/0 setting, so....stats mean 'something' even though both the eval setting and the stats themselves are at zero. Personally, love that it's just outside understanding- as scouting should be -and that it still inspires discussion.

[jimmysthebestcop]

Originally Posted by endgame

Interesting element. The exactness of a bar's setting is not so . . . exact . . . and there is no bar for Player Ratings.

::Note: This is the display currently under a 100/0/0/0 setting, so....stats mean 'something' even though both the eval setting and the stats themselves are at zero. Personally, love that it's just outside understanding- as scouting should be -and that it still inspires discussion.

All game mechanics should be known to the human player. There shouldn't be a mystery over game mechanics. That makes no sense and I haven't seen anything like that in any other game.

Now if the game wants to create artificial "fogs of war" around certain mechanics that is fine. But all mechanics should be properly detailed and described to the player.

Guessing about what check boxes do or different settings is kind of ridiculous.

[ThePride87]

Originally Posted by endgame

Interesting element. The exactness of a bar's setting is not so . . . exact . . . and there is no bar for Player Ratings.

::Note: This is the display currently under a 100/0/0/0 setting, so....stats mean 'something' even though both the eval setting and the stats themselves are at zero. Personally, love that it's just outside understanding- as scouting should be -and that it still inspires discussion.

Yep, even if you have 100% allocated to one of the categories, several staff members will still have a small number in current stats (and sometimes last year's too). Another thing I noticed, even if you don't change the weights at all but still press "recalculate gm tendencies based on these weights", staff won't necessarily keep the exact same settings, you might notice their numbers move slightly. Not sure why, though it's not drastic enough for me to worry....just won't press that button more than once.

As you've said before, there's a lot more that goes into these numbers than just staff settings, so I need to be careful not to speak in such absolutes or use terms like "that's his 'pure' settings"....I don't have enough proof to speak with such confidence. But I am very intrigued by SirMichael's point of view, the concept makes a lot of sense. I can easily accept certain fog of war's in this game like the randomness of development or the fact everything has a probability attached to it, but I've always been annoyed by how little we actually know about these AI eval settings and how to find what you're looking for. I prefer that no bias shapes a GM's or Scout's functions here, and my hope is SirMichael's 25/25/25/25 takes you in that direction.

[SirMichaelJordan]

Never mind, it automatically adjusted.

[luckymann]

Originally Posted by mjryan65

I use 100/0/0/0 for my NCAA league, because there just aren't enough stats in a short season and players develop so quickly that stats from previous years are useless. However, the settings seem to confuse the scouting calculations and I get Power Potentials in the 180s (in the 20-80 scale) and Stuff Potentials in the 150s. If anyone has found a good setup for NCAA (development rate, aging rate, min/max ages, scouting mix), I would love to try it.

I'm bumping this because I'm interested to know if anyone has any thoughts on it.

[Timofmars]

This thread is quite confusing to me. I don't get where some of your conclusions are coming from. Here's the assumptions I would make.

Higher importance of ratings should mean teams make decisions based on what the ratings say the player should do. So a poorly performing player will still be valued the same compared to if he were performing well. I would think this would make the AI as strong as possible since the true ratings should be driving the performance of the player, and differences in actual results from that expected performance are just luck. Also you shouldn't be able to get a discount on a star during a bad year, or trade away your mediocre guy on a hot streak for a premium if the AI doesn't look at stats.

But some caveats regarding the above are:

1. The ratings the AI sees probably depends on their scouting quality.

2. Maybe some combination of attributes can lead to better or more valuable performances than a different combination of attributes even with both cases having the same overall rating. In other words, maybe even 100% accurate overall rating doesn't tell you the true value of the expected performance, and so including a comparison of stats in the evaluation might lead to more accurate determinations of player value in some cases, as the AI will favor players that do better, even if it's not clear why they do better.


On the other hand, having greater dependence on stats does the opposite. Although ratings will probably still be used to some extent even with it set to 0%, since it seems that any stats that don't have a large enough sample size will be prorated in how much it is used in the AI evaluation. So for example, maybe at the beginning of a new season, there are no current stats, so the percentage of evaluation that comes from that will be 0%. Although I don't know if those missing stats will default to the AI using ratings as the replacement, or if the evaluation will just ignore the current year factor, making the other factors more important proportional to what you had set in the game settings.

I would also think that the setting to include AI evaluation in ratings would just affect what ratings you, the human, get. I think it just lets you see what the average AI manager sees (if they had your scouting). Maybe you could use it to see who is overvalued by the AI in your team, etc. Or may it could give you a good idea of what the AI sees when it is making some decisions you think seem strange. Maybe setting scouting accuracy to 100% and turning on AI evaluation in overall ratings would help test out how different AI evaluation settings really change things.



I think the optimal AI in terms of being as competitive as possible would be if the AI used ratings as much as possible, minus some amount based on the accuracy level of their scouting, where less accuracy means they rely more on stats. After all, it should be possible to determine a probable error range interval for stats depending on the size of the sample, and if you could determine an error range for your scouting too, it should be possible to weight each of these based on error range and combine them into a best guess of the true player rating. Though this is something for the game programmers to do, not something players have much control over except in maybe making stats matter more when scouting accuracy is turned down, or making it 100% ratings when scouting is 100% accurate.



I also wonder if the ratings from stats are prorated based on handedness too. Like if I play right handed platoon batter only where he has the advantage in the L/R matchup, will the AI think he is better than he really is? Or will it take all of the ABs against LHPs and use that only in determining his value against the proportion of the league that is LHPs, while giving no opinion on his value vs RHPs, and so defaults to using his scouted ratings vs RHPs for that if there is not enough statistical data.

[ThePride87]

You are correct in your thought-process. Teams relying on stats makes them worse, because results are not as important as process.

[PSUColonel]

This is simply one of the greatest discussion threads EVER.

Great Article to supplement this discussion.

https://blogs.fangraphs.com/the-stat...-stats-debate/

[PSUColonel]

I have really been fooling around with these again (surprise, surprise) and I so desperately want to like 25/25/25/25...and I do to a certain point. I agree they do a great job of being "unbiased" for lack of a better word, and evaluate a player wholey they way it should be done.

If you are using the MLB setup...I am actually going to go right ahead and advocate these settings.

The only issue I see, is if you have international leagues enabled. The AI will use states from leagues such as ***, KBO & LMB and attempt (I think from what I can tell here) to compare "apples and oranges" without knowing the difference. So I decided to test some more and came to the conclusion, that yes, ratings do need to be the largest part of the evaluation...but not so overwhelmingly as I first thought.

I think a "staggered" system probably is best, and that after all of this discussion and testing...we probably had the best mix 10 years ago, when 40/30/20/10 was the overwhelming favorite AI evaluation.

40 percent ratings and 60 percent stats...yet the ratings get the lion's share of any one evaluation category. A pretty good mix actually...and the AI will better be able to compare those "apples & oranges" that are international and independent leagues. It will also take aging into more realistic consideration, as to realize ratings drops, but not act with such a knee jerk reaction that it ends up being unrealistic. Same with contracts.

SO , anyway that is my final answer, yes. lol. This has likely been one of the most hotly debated subjects on these forums over the years. As always, it's still a "do what you like" type option...but in looking for the most realistic AI behavior possible, I think it's very funny to have ended up, right where I started. Make you think huh?

[wodi]

Originally Posted by PSUColonel

I have really been fooling around with these again (surprise, surprise) and I so desperately want to like 25/25/25/25...and I do to a certain point. I agree they do a great job of being "unbiased" for lack of a better word, and evaluate a player wholey they way it should be done.

If you are using the MLB setup...I am actually going to go right ahead and advocate these settings.

The only issue I see, is if you have international leagues enabled. The AI will use states from leagues such as ***, KBO & LMB and attempt (I think from what I can tell here) to compare "apples and oranges" without knowing the difference. So I decided to test some more and came to the conclusion, that yes, ratings do need to be the largest part of the evaluation...but not so overwhelmingly as I first thought.

I think a "staggered" system probably is best, and that after all of this discussion and testing...we probably had the best mix 10 years ago, when 40/30/20/10 was the overwhelming favorite AI evaluation.

40 percent ratings and 60 percent stats...yet the ratings get the lion's share of any one evaluation category. A pretty good mix actually...and the AI will better be able to compare those "apples & oranges" that are international and independent leagues. It will also take aging into more realistic consideration, as to realize ratings drops, but not act with such a knee jerk reaction that it ends up being unrealistic. Same with contracts.

SO , anyway that is my final answer, yes. lol. This has likely been one of the most hotly debated subjects on these forums over the years. As always, it's still a "do what you like" type option...but in looking for the most realistic AI behavior possible, I think it's very funny to have ended up, right where I started. Make you think huh?

I’ve always found the AI too be too slow to promote top prospects, in what you have seen, what is the best setup to have the AI disregard middling talents and actually give playing time to top prospects?