player eval AI settings

[PSUColonel]

Originally Posted by Sweed

Thanks. While I was typing my response you posted this. You said in one sentence what I tried to say in a few paragraphs

Hey, that’s what I had to do in broadcasting…write pithy 30-40 second news copy. LOL

[NoOne]

Not sure if noted, but it does account for lack of sample early in 'current' year. That could override any extreme 100% current year setting to some extent, i'd bet.

ratings are definitely more correlated to results in general, but invariably lag behind reality at end of careers too.

Any 1 year of stats is highly volatile - the opposite of well-correlated. 3 yrs worth starts to become clearer, but i'd wager a high end scout and beefy budget is till equal or better, but when this occurs, they are more often falling off the age cliff too, so to speak... their talent levels out 26-30, but by 29-30, when this info is best, they are just about to lose talent. So, now it is telling you more about what they were than what they are.

In the past OOTP has said that the real-world current player modes start with a different set to maintain known names in starting positions longer. It generally favors ratings more than stats compared to other game mode defaults.

So, you could assume that to some extent and possibly limited to a common sense range of values that adding a larger percentage to stats pushes them out sooner.

Makes sense given the characteristics of each given above, too.

when does each lag? when does each get ahead of curve? stats will lead you late in a player's career but early on the ratings are far, far better... intrinsically due to how the game works. they are afterall the inputs for cause and effect of the simulation where as stats merely result and sample size dictates confidence you can have in them.

[oomm]

Ok, appreciate the detailed explanations. No better input than that derived from gameplay experience, so will try 4/25 and see how it goes.

[oomm]

Originally Posted by NoOne

Not sure if noted, but it does account for lack of sample early in 'current' year. That could override any extreme 100% current year setting to some extent, i'd bet.

ratings are definitely more correlated to results in general, but invariably lag behind reality at end of careers too.

Any 1 year of stats is highly volatile - the opposite of well-correlated. 3 yrs worth starts to become clearer, but i'd wager a high end scout and beefy budget is till equal or better, but when this occurs, they are more often falling off the age cliff too, so to speak... their talent levels out 26-30, but by 29-30, when this info is best, they are just about to lose talent. So, now it is telling you more about what they were than what they are.

In the past OOTP has said that the real-world current player modes start with a different set to maintain known names in starting positions longer. It generally favors ratings more than stats compared to other game mode defaults.

So, you could assume that to some extent and possibly limited to a common sense range of values that adding a larger percentage to stats pushes them out sooner.

Makes sense given the characteristics of each given above, too.

when does each lag? when does each get ahead of curve? stats will lead you late in a player's career but early on the ratings are far, far better... intrinsically due to how the game works. they are afterall the inputs for cause and effect of the simulation where as stats merely result and sample size dictates confidence you can have in them.

You don't think 4/25 turns it off? If, with player eval settings one can impose order of emphasis from say most to least, then it follows that without player eval settings one could not impose order of emphasis. If the player eval settings are used in a way so as to not impose order of emphasis isn't that another way of saying one is neutralizing player eval settings, thus turning them 'off'...........?

I'm going to try 4/25 regardless but am not getting the logic others are insisting on.

I have chosen this emoticon to express my perplexity

[RubeBaker]

50/25/15/10

I used the default AI ratings for years, but this year switched to 50/25/15/10 and I'm seeing better results, more realism than before.

[Brad K]

This thread isn't as informative as it could be because people often aren't specifying whether they're playing historical or something else. The defaults are different.

[Sweed]

Originally Posted by oomm

You don't think 4/25 turns it off? If, with player eval settings one can impose order of emphasis from say most to least, then it follows that without player eval settings one could not impose order of emphasis. If the player eval settings are used in a way so as to not impose order of emphasis isn't that another way of saying one is neutralizing player eval settings, thus turning them 'off'...........?

I'm going to try 4/25 regardless but am not getting the logic others are insisting on.

I have chosen this emoticon to express my perplexity

I think we are talking past each other semantically. You regard it as "turning off" where I and PSU are saying leveling is "turning on" the best way for the AI to proceed.

I look at it this way... ratings\current\previous\2 yrs ago

55\20\15\10 means no matter what the current ratings drive how the AI will make it's decision. Some may like it that way and that's fine but... That is what can lead to the Reggie Jackson example I posted. IE reigning MVP, with 20 HR at All-Star break released because the AI scout determined he had a ratings reduction. In this scenario stats can only be 45% of the decision making process for the AI. Then there is the other side of the coin. Reggie is hitting .235 with 8 HR at the AS break but gets an extension with a nice raise because his ratings increased.

Now go..
45/30/15/10 and the ratings can only be 45% of the decision making process. Now you get the guy outperforming his ratings, due to some luck, in his FA year. Now Reggie Jackson, in the same All-Star example, with declining ratings, but good stats gets an extension instead of released.

with 25\25\25\25 as PSU noted no single category is emphasized over the others. The AI sees the whole picture on a level field. Nothing is turned off. Everything is weighed the same. This is what, in theory, leads the AI to make the most realistic roster decisions.

[Reed]

Originally Posted by Sweed

with 25\25\25\25 as PSU noted no single category is emphasized over the others. The AI sees the whole picture on a level field. Nothing is turned off. Everything is weighed the same. This is what, in theory, leads the AI to make the most realistic roster decisions.

I think I disagree. If I have a 36 year oldplayer I am going to pay a lot more attention to his current stats versus his stats from last year and his 2 year old stats mean almost nothing. Same thing with a 22 year old youngster.
I look at ratings as being a combination of input from my scout, trainer, hitting/pitching coach. I would weigh their input much more than 2 year old or even last year stats, maybe more important than current stats. There is no one correct answer.

[Sweed]

Originally Posted by Reed

I think I disagree. If I have a 36 year oldplayer I am going to pay a lot more attention to his current stats versus his stats from last year and his 2 year old stats mean almost nothing. Same thing with a 22 year old youngster.
I look at ratings as being a combination of input from my scout, trainer, hitting/pitching coach. I would weigh their input much more than 2 year old or even last year stats, maybe more important than current stats. There is no one correct answer.

But this is about how we get the AI to evaluate the player. It's easy for the human to take into account whatever he wants on a case by case basis. Not so easy for the AI. I suppose one could argue the developer should add an "age" column to the equation and you could have a few evaluation setups based on age. So yes, as in my Harmon Killebrew in 1973 example earlier in the thread we want a small window of stats. But until you can make multiple evaluation settings we are stuck with a "one for all" model.

The AI GMs, as noted, also have an input into how this all fits together. Perhaps their veteran\young player preference already does this to an extent? If not it might be a place for the developers to start? Of course if they do they also need to find a way to make it work and not break what we already have

So yeah, in the end players on each end of the age scale will probably have the most problems. But once you weigh heavily one way or the other you open the "bad transaction" from the AI possibilities window wider. At least IMHO.

Great thing is we can all set it to what makes us happy. Or at least avoids the most undesirable outcomes. It is a balance and as we have all said time after time...

There is no right answer

[PSUColonel]

Originally Posted by Reed

I think I disagree. If I have a 36 year oldplayer I am going to pay a lot more attention to his current stats versus his stats from last year and his 2 year old stats mean almost nothing. Same thing with a 22 year old youngster.
I look at ratings as being a combination of input from my scout, trainer, hitting/pitching coach. I would weigh their input much more than 2 year old or even last year stats, maybe more important than current stats. There is no one correct answer.

You have to remember some AI staff will value ratings over stats…others will value stats over ratings. From what I can see….things are already weighted under the hood. Markus has told me in the past that even if you put 0 value into ratings, the AI still has to use ratings to a certain extent.

Also. AI staff will never value 2 years a go too highly…this at 25/25/25

[Mariner and Giants Fan]

2020 and how it was impacted by the pandemic. I would reduce the percentage weight for 2020 to between 0-5% because of the small sample size not being a good indicator compared with the full seasons around it. What do you all think?

[PSUColonel]

Originally Posted by Mariner and Giants Fan

2020 and how it was impacted by the pandemic. I would reduce the percentage weight for 2020 to between 0-5% because of the small sample size not being a good indicator compared with the full seasons around it. What do you all think?

No

[Mariner and Giants Fan]

Originally Posted by PSUColonel

No

Defend your position please.

[oomm]

Originally Posted by Sweed

Great thing is we can all set it to what makes us happy. Or at least avoids the most undesirable outcomes. It is a balance and as we have all said time after time...

There is no right answer

I disagree, in essence and in practice. Of course there is no "right answer" if there is not a serious question. But once there is a good question there is at least a good answer and from good answer there is always a better answer.



Am experimenting with 4/25 and immediately spied contract a contract signing the value of which did not seem to reflect the stats (overpriced). Could be coincidental of course. I previously had ratings weight set at 5.

[oomm]

Originally Posted by PSUColonel

You have to remember some AI staff will value ratings over stats…others will value stats over ratings. From what I can see….things are already weighted under the hood. Markus has told me in the past that even if you put 0 value into ratings, the AI still has to use ratings to a certain extent.

Also. AI staff will never value 2 years a go too highly…this at 25/25/25

Did he say why? Again, my point has been that ratings distill all performance, theoretically. This hasnt been pointed out yet but ratings are for humans, not the game. The game uses data, not 'ratings'. Ratings allow humans to take a snapshot of game data - either in context of player performance, or game manager bias and how they might interact.



Where am I wrong?

[oomm]

Originally Posted by Sweed

I think we are talking past each other semantically. You regard it as "turning off" where I and PSU are saying leveling is "turning on" the best way for the AI to proceed.

I look at it this way... ratings\current\previous\2 yrs ago

55\20\15\10 means no matter what the current ratings drive how the AI will make it's decision. Some may like it that way and that's fine but... That is what can lead to the Reggie Jackson example I posted. IE reigning MVP, with 20 HR at All-Star break released because the AI scout determined he had a ratings reduction. In this scenario stats can only be 45% of the decision making process for the AI. Then there is the other side of the coin. Reggie is hitting .235 with 8 HR at the AS break but gets an extension with a nice raise because his ratings increased.

Now go..
45/30/15/10 and the ratings can only be 45% of the decision making process. Now you get the guy outperforming his ratings, due to some luck, in his FA year. Now Reggie Jackson, in the same All-Star example, with declining ratings, but good stats gets an extension instead of released.

with 25\25\25\25 as PSU noted no single category is emphasized over the others. The AI sees the whole picture on a level field. Nothing is turned off. Everything is weighed the same. This is what, in theory, leads the AI to make the most realistic roster decisions.

We can only know this if we know what data are assessed under 'ratings'. Logic suggests all data inputs not included within the most recent 3 seasons. But even if we do know to a large degree what data goes where we cannot know how effective the programming is. So ultimately it must come down to trial and error.



For those skeptical of the opinion that 4/25 turns off or neutralizes eval settings, you at least have to admit that 4/25 seems non-intuitive, ie not the first values you would consider using. Even the default settings back that up.

[RchW]

Originally Posted by oomm

Did he say why? Again, my point has been that ratings distill all performance, theoretically. This hasnt been pointed out yet but ratings are for humans, not the game. The game uses data, not 'ratings'. Ratings allow humans to take a snapshot of game data - either in context of player performance, or game manager bias and how they might interact.



Where am I wrong?

I don't understand the bold. What is this "data" if it is not the internal 1-250 ratings, "under the hood"?

[Sweed]

Originally Posted by oomm

Did he say why? Again, my point has been that ratings distill all performance, theoretically. This hasnt been pointed out yet but ratings are for humans, not the game. The game uses data, not 'ratings'. Ratings allow humans to take a snapshot of game data - either in context of player performance, or game manager bias and how they might interact.



Where am I wrong?

I think the "why?" is because the game is ratings dependent.

The AI used to only look at ratings in older versions. Then Markus added the ability to have the AI also consider stats. Scouting reports, until recently, were based only on ratings. IIRC correctly it's only been since maybe v20 that scouts considered stats in their evaluations (an option you can turn on or off).

An OOTP player is created with ratings for different skills on a 0-250 scale. The human can display that scale however they want (2-8, 0-20, 0-100 etc.). All of this with scouting turned on is colored with a fog of war for both the human and AI. Or one can turn off scouting and play with 100% accurate ratings. Underneath those ratings are plugged into the game engine and all the variables within. How do these ratings not equal data?

The game has to start somewhere with numbers to enter into the game engine's formulas, no? Sure there are tons of variables but at it's simplest, using only two ratings for the example, don't we have a batter with 215 contact on the "underneath scale" facing a pitcher with 175 stuff?

[Sweed]

Originally Posted by RchW

I don't understand the bold. What is this "data" if it is not the internal 1-250 ratings, "under the hood"?

Thanks.

I miss that button

[Sweed]

Originally Posted by oomm

I disagree, in essence and in practice. Of course there is no "right answer" if there is not a serious question. But once there is a good question there is at least a good answer and from good answer there is always a better answer.



Am experimenting with 4/25 and immediately spied contract a contract signing the value of which did not seem to reflect the stats (overpriced). Could be coincidental of course. I previously had ratings weight set at 5.

But better answer for who? Certainly with all the variables my "better" is different than your "better". That is the point of "no right answer". Which by the way is a strength of OOTP.



And you'll see more. The idea is to lessen these situations while increasing realistic ones. Or might be better to say "when I dig in I can see why the AI did X,Y, or Z". There could be many reasons for this and could occupy a whole thread by itself but one consideration when looking at transactions is..
Unless you are playing without scouts you are seeing the player in question through your scouts eyes. The team that made the deal may have a different picture of who this player is. And that can also be modified by what their GM has for preferences too. Or could just be a crap AI deal

Originally Posted by oomm

We can only know this if we know what data are assessed under 'ratings'. Logic suggests all data inputs not included within the most recent 3 seasons. But even if we do know to a large degree what data goes where we cannot know how effective the programming is. So ultimately it must come down to trial and error.



For those skeptical of the opinion that 4/25 turns off or neutralizes eval settings, you at least have to admit that 4/25 seems non-intuitive, ie not the first values you would consider using. Even the default settings back that up.

Intuitive? No it's not. Congrats you win

When I first read SMJ's theory on 4-25's I thought "huh"

As above, it's the "better" answer for me. It may very well not be for you