AI Evaluation Settings

[PSUColonel]

well, the more stats heavy I go, I see some strange players put on waivers at times...however, leaning towards stats also seems to allow the AI to better utilize players in it's rotation/bullpen/lineups. Also, when going more stats heavy, it seems a bit easier to fleece the AI in trades.


So, the closer to 50/50 the better I feel, but there are advantages to both 55/25/15/5 & 45/30/30/5 IMO....I have sort of ruled out 40/30/20/10 just because it's 10 points off to 50 for ratings.

The seemingly obvious one to use might be 50/30/15/5...I think it works well, but it also really weakens the previous years stats. You might say the same about 55/25/15/5...but that's not really the case since you are lowering the current year's stats.

50/30/15/5......this year is worth 2x last years

55/25/15/5 ......this year is worth only 1.6x last years.....but in this case, ratings are 2.2x more than this year's stats.

[Orcin]

Originally Posted by PSUColonel

So, the closer to 50/50 the better I feel, but there are advantages to both 55/25/15/5 & 45/30/30/5 IMO....I have sort of ruled out 40/30/20/10 just because it's 10 points off to 50 for ratings.

I mistyped in my earlier post. I am using 30/40/20/10 in this historical league. The reason that I tried such a low percent to rating is related to historical minors. I wanted the AI to keep as many regular MLB players in the majors as possible instead of replacing them with over-rated players from the minor league database. It has worked for the most part and I haven't seen any goofy trades or waiver moves.

I would agree with you that at least 50% ratings is better for fictional and especially modern MLB (which trends toward fictional very quickly). I use 50/30/15/5 for that. Any of your other suggestions could be better; I just haven't tried them.

[NoOne]

Originally Posted by endgame

Found it. FWIW (this is not a stats-only suggestion, btw)

<refer to orginal post for quoted passage>

belated, but... the game does redistribute current year while sample is on the small side. obviously, with the quote you posted it gives you control of how you it is done.

i really like this idea, i may have to start dinking around a bit. i'd probably keep ratings static and affect the 3 years distribution at specific # of games into schedule.

[homerj]

What about something closer to like 50/45/5/0 ? Is

[PSUColonel]

Originally Posted by homerj

What about something closer to like 50/45/5/0 ? Is

If you wanted to accomplish something like that, I'd probably go 50/30/20/0

[endgame]

Argh.... to each his own. This conversation always stirs me up. Without rigorous examination, the difference between these 5's and 10's is likely indiscernible, or marginal at most. Beyond that, it's just gut and overweighting confirmation bias, IMO. This isn't Madden sliders. It's evaluations by variable data-driven entities.

Somewhere, Deja Bru suggests adjustments in tandem with star ratings (or numerical, it doesn't matter) and determining the personal view and judgment' that suits you personally. That approach is just as likely to arrive at a reasonable evaluation that will offer you consistency, along with a few surprises and disappointments.

[homerj]

Originally Posted by endgame

Argh.... to each his own. This conversation always stirs me up. Without rigorous examination, the difference between these 5's and 10's is likely indiscernible, or marginal at most. Beyond that, it's just gut and overweighting confirmation bias, IMO. This isn't Madden sliders. It's evaluations by variable data-driven entities.

Yes, if it's one thing OOTP lets you do to your own peril is dive head first down the rabbit hole of editing 100 different numbers. I barely touch the default numbers anymore, as I felt I was spending too much time tweaking numbers, and not enough time playing baseball.

[Orcin]

Originally Posted by endgame

Argh.... to each his own. This conversation always stirs me up. Without rigorous examination, the difference between these 5's and 10's is likely indiscernible, or marginal at most. Beyond that, it's just gut and overweighting confirmation bias, IMO. This isn't Madden sliders. It's evaluations by variable data-driven entities.

However, Matt Arnold did say that the default setting for AI evaluation in the MLB game is different from the historical and fictional games in order to influence the AI behavior. So it obviously does influence the AI behavior, and likely in a way that is discernible - otherwise, why bother?

[endgame]

Originally Posted by Orcin

However, Matt Arnold did say that the default setting for AI evaluation in the MLB game is different from the historical and fictional games in order to influence the AI behavior. So it obviously does influence the AI behavior, and likely in a way that is discernible - otherwise, why bother?

Oh, no doubt it does, and those comparisons differ more greatly than the 5 point distinctions being discussed here, more often than not. And in the MLB case, they're there to assist in directing the selection of the initial roster and lineups, but fast forward a few weeks and, my guess- and that's all it is -is the teams are already making different decisions based on acquisitions and farm systems. It's only controllable on the front end and for so long. Edit: I'll add that the rosters in the MLB setup are also 'tuned' with edited ratings to further assist that tandem effort. Evaluations alone don't suffice.

Does it have an impact? Surely. But I hold firm that without extensive and data recording under exact and controlled studies, the minor differences between one setting and another, e.g. 50/30/10/10 or 45/25/20/10 are just fancy.

[PSUColonel]

Originally Posted by endgame

Argh.... to each his own. This conversation always stirs me up. Without rigorous examination, the difference between these 5's and 10's is likely indiscernible, or marginal at most. Beyond that, it's just gut and overweighting confirmation bias, IMO. This isn't Madden sliders. It's evaluations by variable data-driven entities.

Somewhere, Deja Bru suggests adjustments in tandem with star ratings (or numerical, it doesn't matter) and determining the personal view and judgment' that suits you personally. That approach is just as likely to arrive at a reasonable evaluation that will offer you consistency, along with a few surprises and disappointments.

You a very likely correct...I guess I am just on a quest for perfection, or at least what I perceive to be perfection....but in all honesty I do think there are some slight nuances if ratings are say 10 points apart. E.g. 45 & 55 but not so much 5 points apart

[Mat]

If you are experienced with the game, by all means, play around with it.

IMO, the key is to pick a setting (default does the trick just fine) and stick with it.

[Orcin]

Originally Posted by endgame

Does it have an impact? Surely. But I hold firm that without extensive and data recording under exact and controlled studies, the minor differences between one setting and another, e.g. 50/30/10/10 or 45/25/20/10 are just fancy.

I did some extensive testing today, and I absolutely agree with your statement. It was very hard to find any significant differences from moving the settings by 5% one way or another.

The rest of your post got me thinking about trying a current MLB league with the three magic boxes (player ratings relative to selected league, ratings based on AI eval not pure, and ratings based on all players not positions) checked (instead of the default unchecked) to see just how much different it is. The AI should act more quickly to improve its team in theory and this may lead to a better (although less true to life) experience. I will still use the default weights (65/20/10/5) to start.

[endgame]

Thanks, Orc. Did I miss it, or how are you controlling team focus?

[Orcin]

Originally Posted by endgame

Thanks, Orc. Did I miss it, or how are you controlling team focus?

I'm not. I am just thinking that the AI will be quicker to demote players that are not performing and more likely to shuffle its lineups. We'll see. I haven't done this in a while because I have been stuck on historical, so it should be fun anyway.

[joejccva71]

If I already started my 1st season with the default (65, 20,10,5) is it too late to change it to 50,30,15,5 ? Do I have to wait until the offseason?

I'm about 55 games in.

Also, do you guys make the change in Game Settings \ Global Settings - "Overall rating based on AI evaluation, not pure ratings" ? Or do you leave it unchecked?

[Orcin]

Originally Posted by joejccva71

Also, do you guys make the change in Game Settings \ Global Settings - "Overall rating based on AI evaluation, not pure ratings" ? Or do you leave it unchecked?

If you leave this box unchecked, you might as well leave the AI evaluation settings at default because the AI won't use them. If you check this box, the AI uses the weights. Unchecked, it uses ratings only.

[joejccva71]

Originally Posted by Orcin

If you leave this box unchecked, you might as well leave the AI evaluation settings at default because the AI won't use them. If you check this box, the AI uses the weights. Unchecked, it uses ratings only.

I understand.

So do you think I should edit the weights mid-season (60 games in now) or should I wait until off-season?

[Orcin]

I would do it on the first day of the offseason. Recalc ratings and re-run scouting after you change them.

[Mat]

This I'm pretty sure is incorrect.

Ai evaluation tick box only effects human scouting.

Percentages are always in effect for AI scout ratings.

[Orcin]

Originally Posted by Mat

This I'm pretty sure is incorrect.

Ai evaluation tick box only effects human scouting.

Percentages are always in effect for AI scout ratings.

Ok, I am confused. Sorry for giving the wrong advice. Maybe I am getting too old for this. I will just write it off as a bad day yesterday and move on.

It is really interesting to see the ratings with stats included vs. pure ratings. Some of the evaluations make a lot more sense this way.