Major Problems With Historical Roster AI

Charlie Hough · 08-06-2007, 11:31 PM

There is some serious trouble with the roster AI in historical leagues. The AI GMs are consistently placing excellent players in the minor leagues to start a new league, and they are sometimes releasing players as well.

I've run a number of tests, and there seems to be no rhyme or reason why the AI makes horrible decisions. For example, in a 1987 league that I just created, the Oakland A's GM put Jose Canseco on its AAA roster. The Atlanta Braves GM placed Glenn Hubbard, Andres Thomas, and Rafael Ramirez in AAA, even though they're all starters.

The Texas Rangers put five top players in AAA, including Bobby Witt, Mitch Williams, Steve Buchele, Curtis Wilkerson, and Don Slaught.

I've tried creating several leagues in different years, and I've tried various settings for ratings weights for player evaluations. The end result is that the AI always finds a way to put top players in the minors while using deadbeats on the 25-man roster.

Does anyone know how to prevent this other than taking over every team and painstakingly correcting these mistakes and running all the franchises myself?

Kelric · 08-07-2007, 12:22 AM

What are the guys rated as they're getting demoted compared to the guys getting promoted over them?

Charlie Hough · 08-07-2007, 11:30 PM

The major guys that are getting demoted are usually better rated than those who are on the active rosters. There may be certain ratings that are higher or lower, but the overall quality of these players is better than the ones who have no business being on the active roster.

However, it shouldn't matter what they're rated when I place ZERO percentage emphasis on ratings and set the AI to place 100% emphasis on last season's stats.

Kelric · 08-08-2007, 12:46 AM

Ah, well then... yeah... I got nothing.

No Pepper · 08-08-2007, 08:26 AM

Quote:

Originally Posted by Charlie Hough

However, it shouldn't matter what they're rated when I place ZERO percentage emphasis on ratings and set the AI to place 100% emphasis on last season's stats.

I wonder if the last season stat emphasis function uses simmed stats and not imported stats in a first season or subsequent ones. That might explain why the game still uses "ratings" to jumble your roster because there are "no stats" to use.

Tekneek · 08-08-2007, 10:52 AM

If the game uses in-game stats to evaluate players, and brand new rookies have nothing for in-game stats, then it makes sense that they would not start in the big leagues over other guys who have in-game stats.

Charlie Hough · 08-08-2007, 06:04 PM

These are not rookies. These are players that are part of the entire roster set for the starting season.

The results are better when you have imported the full history when creating your league. If there is no history, then there aren't any stats for the game to use. However, even if there is no history, there is still a major difference between leaving the default emphasis and setting the emphasis to be 100% on the current or previous years' stats. Somehow the game is distinguishing between these settings even though there may be no stats to use.

Also, whether it's stats or ratings, there is no excuse whatsoever for Jose Canseco being put in the minors for 1987. He had great stats in 1986 AND his ratings are better than virtually anyone who would be put on the active roster ahead of him. So there seems to be a problem, even when using ratings as the emphasis.

The solution has been doing an import of all history up to the beginning season for the league. Even then, there are players who are put in the minors when they shouldn't be. But it's not as extreme as the problem when you don't have any historical stats. In that case, the game seems to become confused and does not correctly assess players for their stats OR their ratings. That's how you end up with someone like Canseco in the minors. But when you import the complete history before your season, most of this is fixed.

pstrickert · 08-08-2007, 07:03 PM

CH, you might try PMing Lighthousekeeper. He tested this part of the game (stats-based AI) quite a bit during beta testing.

pstrickert · 08-08-2007, 07:13 PM

I hope it's OK to quote Markus from the beta forums. He made the following comments in response to Lighthousekeeper. Lighthousekeeper was testing a league in the year 2010. I think it's fairly easy to infer the questions that elicited each comment.

Keep in mind that if no stats are available, a proper % of the ratings will be used instead. So if you have stats % of 2 years back set to 10, and there are no stats for that player, these 10% gets assigned to the ratings weight.

Two years back means in this case 2008. The guy had only 391 AB, but for the stats to count 100%, he needs about 600 AB. So, in this case about 60% stats are used, and 40% ratings. 40% of these perfect ratings is enough to make him an 80 overall, since this guy is a 2B. So, the calculation is not messed up.

And don't forget, minor league stats are used as well...

Lets say you have current year stats set to 100%. And the hitter this year has only 100 PA so far. The minimum PA for the stats in order to count entirely is 500 now. So the percentages get adjusted, the current year stats are used for 20% now, and the previous year stats 80%. Now, if the guy had just 250 PA last year, it is 20% current year, 40% last year, 40% 2 years ago. And if that guy did not play 2 years ago, it is 20% current, 40% last and 40% ratings.

Kelric · 08-08-2007, 07:56 PM

See, that sort of stuff would be great to have put out there somewhere other than the beta forum. Good to know. Thanks.

No Pepper · 08-09-2007, 01:47 PM

Although I had a small hunch that's the way it worked, "using current year stats" or any such setting is misleading without that info [from Markus]. Thanks a million pstrickert! I totally second Kelric's plea of making this sort of stuff public.

Charlie Hough · 08-09-2007, 09:56 PM

This is good information, but there is still a problem, even when ratings are used in lieu of stats. There are numerous cases of players with better ratings ending up in the minors for no good reason.

pstrickert · 08-09-2007, 10:12 PM

I'm guessing that Markus would tell you to trust the OOTP AI. It picks the line-up the real-life manager SHOULD have picked, if the real-life manager had been smarter.

I exaggerate.

Nevertheless, Markus is not persuaded by anecdotal evidence. He needs hard numbers and screenshots. Can you post the specifics please (ratings, stats, etc.)? And remember: the AI does not recognize names. (I'm sure you've heard that before.)

DougWyatt · 08-11-2007, 04:53 AM

pstric

Do you know if those min. at bats are hard numbers ?

I ask because I run a league with a very short schedule (26 games) - and if it's a 500-600 AB; then I'm now thinking the AI will never take stats into consideration

.

pstrickert · 08-11-2007, 11:46 AM

That's a good question. I'm sorry, but I don't know. I'll post the Q on the beta board, though.

pstrickert · 08-12-2007, 11:22 PM

I imported the '87 season and took a look. I don't see nearly as many problems as you do, CharlieHough. (I used the 3-year recalc, BTW, and left the AI stats/ratings evaluation at default.) Hopefully, Markus will refine this part of the game for the next version. I know Garlon has created a new formula for ratings pitchers. If implemented, it should result in familiar names showing up in familiar places. That, IMO, will make historicals more realistic.

pstrickert · 08-12-2007, 11:23 PM

Quote:

Originally Posted by DougWyatt

pstric

Do you know if those min. at bats are hard numbers ?

I ask because I run a league with a very short schedule (26 games) - and if it's a 500-600 AB; then I'm now thinking the AI will never take stats into consideration

.

Still no reply on the beta board. Sorry.

pstrickert · 08-20-2007, 09:17 AM

From Markus: Yes, the 500 PA minimum is adjusted based on schedule length.

DougWyatt · 08-21-2007, 11:07 AM

Quote:

Originally Posted by pstrickert

From Markus: Yes, the 500 PA minimum is adjusted based on schedule length.

Good stuff. Thanks

08-06-2007, 11:31 PM	#1
Charlie Hough Hall Of Famer Join Date: May 2006 Posts: 3,625	Major Problems With Historical Roster AI There is some serious trouble with the roster AI in historical leagues. The AI GMs are consistently placing excellent players in the minor leagues to start a new league, and they are sometimes releasing players as well. I've run a number of tests, and there seems to be no rhyme or reason why the AI makes horrible decisions. For example, in a 1987 league that I just created, the Oakland A's GM put Jose Canseco on its AAA roster. The Atlanta Braves GM placed Glenn Hubbard, Andres Thomas, and Rafael Ramirez in AAA, even though they're all starters. The Texas Rangers put five top players in AAA, including Bobby Witt, Mitch Williams, Steve Buchele, Curtis Wilkerson, and Don Slaught. I've tried creating several leagues in different years, and I've tried various settings for ratings weights for player evaluations. The end result is that the AI always finds a way to put top players in the minors while using deadbeats on the 25-man roster. Does anyone know how to prevent this other than taking over every team and painstakingly correcting these mistakes and running all the franchises myself?

08-09-2007, 01:47 PM	#11
No Pepper All Star Starter Join Date: Jan 2003 Location: Milwaukee, WI Posts: 1,553	Although I had a small hunch that's the way it worked, "using current year stats" or any such setting is misleading without that info [from Markus]. Thanks a million pstrickert! I totally second Kelric's plea of making this sort of stuff public. __________________ Uniforms compatible with OOTP23/24 Historical Major League Baseball 1901-current Historical Major League Baseball 1871-1900 Historical Federal League Historical Negro Leagues

08-07-2007, 12:22 AM	#2
Kelric Hall Of Famer Join Date: Mar 2002 Location: Greater Boston Area Posts: 3,992	What are the guys rated as they're getting demoted compared to the guys getting promoted over them?

08-07-2007, 11:30 PM	#3
Charlie Hough Hall Of Famer Join Date: May 2006 Posts: 3,625	The major guys that are getting demoted are usually better rated than those who are on the active rosters. There may be certain ratings that are higher or lower, but the overall quality of these players is better than the ones who have no business being on the active roster. However, it shouldn't matter what they're rated when I place ZERO percentage emphasis on ratings and set the AI to place 100% emphasis on last season's stats.

08-08-2007, 12:46 AM	#4
Kelric Hall Of Famer Join Date: Mar 2002 Location: Greater Boston Area Posts: 3,992	Ah, well then... yeah... I got nothing.

08-08-2007, 10:52 AM	#6
Tekneek Major Leagues Join Date: Mar 2002 Location: GA, USA Posts: 453	If the game uses in-game stats to evaluate players, and brand new rookies have nothing for in-game stats, then it makes sense that they would not start in the big leagues over other guys who have in-game stats.

08-08-2007, 06:04 PM	#7
Charlie Hough Hall Of Famer Join Date: May 2006 Posts: 3,625	These are not rookies. These are players that are part of the entire roster set for the starting season. The results are better when you have imported the full history when creating your league. If there is no history, then there aren't any stats for the game to use. However, even if there is no history, there is still a major difference between leaving the default emphasis and setting the emphasis to be 100% on the current or previous years' stats. Somehow the game is distinguishing between these settings even though there may be no stats to use. Also, whether it's stats or ratings, there is no excuse whatsoever for Jose Canseco being put in the minors for 1987. He had great stats in 1986 AND his ratings are better than virtually anyone who would be put on the active roster ahead of him. So there seems to be a problem, even when using ratings as the emphasis. The solution has been doing an import of all history up to the beginning season for the league. Even then, there are players who are put in the minors when they shouldn't be. But it's not as extreme as the problem when you don't have any historical stats. In that case, the game seems to become confused and does not correctly assess players for their stats OR their ratings. That's how you end up with someone like Canseco in the minors. But when you import the complete history before your season, most of this is fixed.

08-08-2007, 07:03 PM	#8
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	CH, you might try PMing Lighthousekeeper. He tested this part of the game (stats-based AI) quite a bit during beta testing.

08-08-2007, 07:13 PM	#9
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	I hope it's OK to quote Markus from the beta forums. He made the following comments in response to Lighthousekeeper. Lighthousekeeper was testing a league in the year 2010. I think it's fairly easy to infer the questions that elicited each comment. Keep in mind that if no stats are available, a proper % of the ratings will be used instead. So if you have stats % of 2 years back set to 10, and there are no stats for that player, these 10% gets assigned to the ratings weight. Two years back means in this case 2008. The guy had only 391 AB, but for the stats to count 100%, he needs about 600 AB. So, in this case about 60% stats are used, and 40% ratings. 40% of these perfect ratings is enough to make him an 80 overall, since this guy is a 2B. So, the calculation is not messed up. And don't forget, minor league stats are used as well... Lets say you have current year stats set to 100%. And the hitter this year has only 100 PA so far. The minimum PA for the stats in order to count entirely is 500 now. So the percentages get adjusted, the current year stats are used for 20% now, and the previous year stats 80%. Now, if the guy had just 250 PA last year, it is 20% current year, 40% last year, 40% 2 years ago. And if that guy did not play 2 years ago, it is 20% current, 40% last and 40% ratings.

08-08-2007, 07:56 PM	#10
Kelric Hall Of Famer Join Date: Mar 2002 Location: Greater Boston Area Posts: 3,992	See, that sort of stuff would be great to have put out there somewhere other than the beta forum. Good to know. Thanks.

08-09-2007, 09:56 PM	#12
Charlie Hough Hall Of Famer Join Date: May 2006 Posts: 3,625	This is good information, but there is still a problem, even when ratings are used in lieu of stats. There are numerous cases of players with better ratings ending up in the minors for no good reason.

08-09-2007, 10:12 PM	#13
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	I'm guessing that Markus would tell you to trust the OOTP AI. It picks the line-up the real-life manager SHOULD have picked, if the real-life manager had been smarter. I exaggerate. Nevertheless, Markus is not persuaded by anecdotal evidence. He needs hard numbers and screenshots. Can you post the specifics please (ratings, stats, etc.)? And remember: the AI does not recognize names. (I'm sure you've heard that before.)

08-11-2007, 04:53 AM	#14
DougWyatt All Star Starter Join Date: Dec 2003 Location: Michigan Posts: 1,117	pstric Do you know if those min. at bats are hard numbers ? I ask because I run a league with a very short schedule (26 games) - and if it's a 500-600 AB; then I'm now thinking the AI will never take stats into consideration .

08-11-2007, 11:46 AM	#15
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	That's a good question. I'm sorry, but I don't know. I'll post the Q on the beta board, though.

08-12-2007, 11:22 PM	#16
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	I imported the '87 season and took a look. I don't see nearly as many problems as you do, CharlieHough. (I used the 3-year recalc, BTW, and left the AI stats/ratings evaluation at default.) Hopefully, Markus will refine this part of the game for the next version. I know Garlon has created a new formula for ratings pitchers. If implemented, it should result in familiar names showing up in familiar places. That, IMO, will make historicals more realistic.

08-20-2007, 09:17 AM	#18
pstrickert Hall Of Famer Join Date: Dec 2005 Posts: 15,741	From Markus: Yes, the 500 PA minimum is adjusted based on schedule length.