My thoughts on how to improve the roster AI management

jerrycapo · 06-14-2006, 01:30 PM

I think I have realized the problem with the AI roster management. The biggest reason the AI moves players all over the place is because the AI does not use the date to make its roster moves. It sees March 1st the same as April 3rd, June 6th, July 11th, July 31st, and October 21st.

Please allow me to explain, March 1st is Spring Training this is where the AI should be looking at what holes it has to fill, it should do this by first placing their most expensive contracts into a starting position. That is the most expensive, not necessarily the best (I will explain how the AI should handle bad contracts at another time. Then the AI should assign players according to ability. If you have a position filled by a guy with below average ability, the AI should bring in some competition (ie sign average players and below to minor league contracts and invite to spring training).

April 3rd, Opening Day. Teams should fill out their rosters first by big contracts then by whomever “wins” the position in Spring Training. This may be an oversimplification but it is the gist. Spring training stats should be the deciding factor between individuals in the same talent range. These rosters should stay intact at all levels from MLB to low A. Changes should be made due to injury, but for the most part they should not be changed.

June 6th is the amateur draft. (I know in game it is June 1st). At this point the AI should look at their rosters and determine if the players are adequately performing. It should also look at the minors and see if any players are red hot and should be moved up. The AI should make the moves accordingly. After the moves are made the AI should only be moving players due to injury until the next stage.

July 11th, the All Star Game. This is the point where teams should be looking at whether or not they are in contention. If they are out and as they fall out of contention, they should be looking to trade any potential free agents that will not re-sign with the team. Minor league players should be getting more attention now for teams out of contention. Veterans should be desired more for teams that are in contention to fill positions where players are underperforming.

July, 31st, the trading deadline. As this date gets closer, teams out of contention should be trying harder to move anyone that is a potential free agent or arbitration eligible that they do not plan to sign for the following season. Players who are not generating any interest are either cut or “given” away to another team via waivers before the August 31st waiver deadline.

Obviously September 1st, the game handles as the teams expand their rosters. Again teams out of contention should place an emphasis on getting their young players playing time. Teams contending for a playoff spot, should be using primarily veterans and reserves in blowout games, teams that have clinched should be starting veterans and substituting as the games go on to keep vets fresh but rested for the playoffs.

Finally October 21st, the World Series. Teams not involved in the big game should be firing managers, gms, coaches, and scouts, with hiring to take place sometime after the World Series.

The problem as I can tell with the AI is it treats every day the same and has a formula it uses to check and see if players should be moved. It would seem this is done everyday. Since the game does recognize that players need playing time to be able to grow, the same players are shuffled through the various leagues to get playing time. My guess is they are also valued pretty close to each other. The following is a scenario that I believe is happening for four shortstops

Pos Name Avg Rating Potential
SS Wright 45/80 55/80
SS Power 48/80 52/80
SS Jones 47/80 53/80
SS Young 46/80 54/80

SS Wright starts the year at AAA (Richmond), Power at AA (Mississippi), Jones at A (Myrtle Beach), Young at A (Rome). On May 1st Wright has a bad day, Power has an average day, Jones a good day, and Young a great day:

May 1
Wright 0-4 3 K’s
Power 1-4 1 K, 1B
Jones 2-4 1B, HR, 1RBI
Young 3-3 BB, 2HR, 7 RBI

So when you add stats as part of the equation Wright gets sent down to AA Power up to AAA. Jones up to AA and Young up to A Myrtle beach. So the next day Power struggles at AAA, Jones has a good day, Wright is still struggling at AA, and Young has a good day:

May 2
Power 0-4 1 K
Wright 1-4 3 K
Jones 2-4 2B, HR, 3 RBI
Young 3-4 1B, 3B, HR, 2 RBI

So Power goes down to AA, Jones moves up to AAA, Wright goes down to A MB, and Young moves up to AA. It seems the AI does this every day and adjusts its rosters accordingly.

This is just food for thought. Thought I would share what I think I am seeing and a possible way to fix it. Any other ideas would be appreciated. Thanks.

gordyhulten · 06-14-2006, 01:35 PM

Excellent thoughts. I don't know about the feasibility of implementing them, but excellent, constructive thoughts.

jerrycapo · 06-15-2006, 01:58 PM

Quote:

Originally Posted by gordyhulten

Excellent thoughts. I don't know about the feasibility of implementing them, but excellent, constructive thoughts.

I agree. This may take a re-write of the roster handling logic. I don't think it is a simple teak, but I think the roster handling needs to be more of an event driven procedure, than an everyday set of checks that it seems to perform curently. Using an event driven logic should be the best way for the AI to keep its competitiveness while providing realism at the same time.

Killebrew · 06-15-2006, 03:15 PM

Nice idea's, but the AI could not use the specific dates, it would have to calculate these key points in the season based on the schedule the league was using. Having some AI roster management/trade strategy/waiver functions kicked off at these key points in the season seems like a great idea. In fact, I'm pretty sure the game is already kicking off some similar roster management functions now (e.g. the expansion of spring training rosters and the return to 25 man rosters for opening day). For my fictional league some of these AI roster functions don't always make sense, and I wonder if the game is basing some of these AI roster decisions on hard coded dates rather than calculating the key points of my leagues unique schedule (a 42 game season that is over by July 1st

).

jerrycapo · 06-15-2006, 04:16 PM

I just used the specific dates to illustrate how MLB works. I would imagine the game could use the same logic it uses for the simming to important events. IE Sim until 1st Year Player draft. The logic seems to be there for the AI to recognize key events and it can use that to determine which point of the season it is at.

06-14-2006, 01:30 PM	#1
jerrycapo Major Leagues Join Date: Feb 2002 Posts: 411	My thoughts on how to improve the roster AI management I think I have realized the problem with the AI roster management. The biggest reason the AI moves players all over the place is because the AI does not use the date to make its roster moves. It sees March 1st the same as April 3rd, June 6th, July 11th, July 31st, and October 21st. Please allow me to explain, March 1st is Spring Training this is where the AI should be looking at what holes it has to fill, it should do this by first placing their most expensive contracts into a starting position. That is the most expensive, not necessarily the best (I will explain how the AI should handle bad contracts at another time. Then the AI should assign players according to ability. If you have a position filled by a guy with below average ability, the AI should bring in some competition (ie sign average players and below to minor league contracts and invite to spring training). April 3rd, Opening Day. Teams should fill out their rosters first by big contracts then by whomever “wins” the position in Spring Training. This may be an oversimplification but it is the gist. Spring training stats should be the deciding factor between individuals in the same talent range. These rosters should stay intact at all levels from MLB to low A. Changes should be made due to injury, but for the most part they should not be changed. June 6th is the amateur draft. (I know in game it is June 1st). At this point the AI should look at their rosters and determine if the players are adequately performing. It should also look at the minors and see if any players are red hot and should be moved up. The AI should make the moves accordingly. After the moves are made the AI should only be moving players due to injury until the next stage. July 11th, the All Star Game. This is the point where teams should be looking at whether or not they are in contention. If they are out and as they fall out of contention, they should be looking to trade any potential free agents that will not re-sign with the team. Minor league players should be getting more attention now for teams out of contention. Veterans should be desired more for teams that are in contention to fill positions where players are underperforming. July, 31st, the trading deadline. As this date gets closer, teams out of contention should be trying harder to move anyone that is a potential free agent or arbitration eligible that they do not plan to sign for the following season. Players who are not generating any interest are either cut or “given” away to another team via waivers before the August 31st waiver deadline. Obviously September 1st, the game handles as the teams expand their rosters. Again teams out of contention should place an emphasis on getting their young players playing time. Teams contending for a playoff spot, should be using primarily veterans and reserves in blowout games, teams that have clinched should be starting veterans and substituting as the games go on to keep vets fresh but rested for the playoffs. Finally October 21st, the World Series. Teams not involved in the big game should be firing managers, gms, coaches, and scouts, with hiring to take place sometime after the World Series. The problem as I can tell with the AI is it treats every day the same and has a formula it uses to check and see if players should be moved. It would seem this is done everyday. Since the game does recognize that players need playing time to be able to grow, the same players are shuffled through the various leagues to get playing time. My guess is they are also valued pretty close to each other. The following is a scenario that I believe is happening for four shortstops Pos Name Avg Rating Potential SS Wright 45/80 55/80 SS Power 48/80 52/80 SS Jones 47/80 53/80 SS Young 46/80 54/80 SS Wright starts the year at AAA (Richmond), Power at AA (Mississippi), Jones at A (Myrtle Beach), Young at A (Rome). On May 1st Wright has a bad day, Power has an average day, Jones a good day, and Young a great day: May 1 Wright 0-4 3 K’s Power 1-4 1 K, 1B Jones 2-4 1B, HR, 1RBI Young 3-3 BB, 2HR, 7 RBI So when you add stats as part of the equation Wright gets sent down to AA Power up to AAA. Jones up to AA and Young up to A Myrtle beach. So the next day Power struggles at AAA, Jones has a good day, Wright is still struggling at AA, and Young has a good day: May 2 Power 0-4 1 K Wright 1-4 3 K Jones 2-4 2B, HR, 3 RBI Young 3-4 1B, 3B, HR, 2 RBI So Power goes down to AA, Jones moves up to AAA, Wright goes down to A MB, and Young moves up to AA. It seems the AI does this every day and adjusts its rosters accordingly. This is just food for thought. Thought I would share what I think I am seeing and a possible way to fix it. Any other ideas would be appreciated. Thanks. __________________ All the best, Jerry

06-14-2006, 01:35 PM	#2
gordyhulten All Star Starter Join Date: Jan 2003 Location: Champaign, IL Posts: 1,175	Excellent thoughts. I don't know about the feasibility of implementing them, but excellent, constructive thoughts. __________________ Gordy Hulten Owner / General Manager Red Willow Roadrunners -- Kennel Series Champions: 1951, 1959, 1964, 1965, 1972, 1975, 1980, 1982, 1983 Dog Days Baseball - "The World's Best Online OOTP League" Creator inactive: Republican League - OOTP 2009 Dynasty inactive: Republican League Dynasty - Version 2.0 inactive: Republican League Dynasty

06-15-2006, 03:15 PM	#4
Killebrew Hall Of Famer Join Date: Dec 2001 Posts: 3,326	Nice idea's, but the AI could not use the specific dates, it would have to calculate these key points in the season based on the schedule the league was using. Having some AI roster management/trade strategy/waiver functions kicked off at these key points in the season seems like a great idea. In fact, I'm pretty sure the game is already kicking off some similar roster management functions now (e.g. the expansion of spring training rosters and the return to 25 man rosters for opening day). For my fictional league some of these AI roster functions don't always make sense, and I wonder if the game is basing some of these AI roster decisions on hard coded dates rather than calculating the key points of my leagues unique schedule (a 42 game season that is over by July 1st).

06-15-2006, 04:16 PM	#5
jerrycapo Major Leagues Join Date: Feb 2002 Posts: 411	I just used the specific dates to illustrate how MLB works. I would imagine the game could use the same logic it uses for the simming to important events. IE Sim until 1st Year Player draft. The logic seems to be there for the AI to recognize key events and it can use that to determine which point of the season it is at. __________________ All the best, Jerry