AI Trading Mindsets

Tom Martell · 04-22-2003, 04:56 PM

I know I've mentioned this before, but given all the ongoing complaints about the trading AI...

Implement 3 distinct modes for the for each team. Create a simple algorithm based on current prospect list (there already is the team system rankings, so teams know how good they are relative to other teams), current record and divisional standing, and current major league talent that sends the team between the different mode.

1) Pennant Race - the team will attempt to acquire major league at the expense of its minor league system to fill holes in the lineup or rotation.

2) Contending - the team has no trade modifiers - it won't be anxious to deal away its prospects or its major leaguers because it is hoping to both contend now and build for the future.

3) Rebuilding - AI takes the view that major leaguers are worthless and eat money. the team is anxious to dump high salary players and acquire top minor league talent, intent on making a run in 2-3 years.

Nothing is more ridiculous than being able to trade an old, declining star for top tier talent from teams that are in last place and 35 games out of first. They should have no interest whatsoever.

WLight · 04-22-2003, 07:33 PM

I think this is a great idea and a lot of support for it has already been expressed.

However, as you note, there are systemic problems with the trade AI in 5.1 that many players feel are of paramount importance. Until those basic issues are addressed I think the implementation of a trade 'feature' such as you suggest will have to wait.

Sweed · 04-22-2003, 07:47 PM

Quote:

Originally posted by Tom Martell
I know I've mentioned this before, but given all the ongoing complaints about the trading AI...

Implement 3 distinct modes for the for each team. Create a simple algorithm based on current prospect list (there already is the team system rankings, so teams know how good they are relative to other teams), current record and divisional standing, and current major league talent that sends the team between the different mode.

1) Pennant Race - the team will attempt to acquire major league at the expense of its minor league system to fill holes in the lineup or rotation.

2) Contending - the team has no trade modifiers - it won't be anxious to deal away its prospects or its major leaguers because it is hoping to both contend now and build for the future.

3) Rebuilding - AI takes the view that major leaguers are worthless and eat money. the team is anxious to dump high salary players and acquire top minor league talent, intent on making a run in 2-3 years.

Nothing is more ridiculous than being able to trade an old, declining star for top tier talent from teams that are in last place and 35 games out of first. They should have no interest whatsoever.

You should post your ideas in the sticky trade thread discussion so it doesn't fall off the board. Some good ideas here that have been expressed in different ways. I would sure like to see something like this. Maybe this could help "fix" some of the systmic problems in the current version.

Bears5122 · 04-22-2003, 08:08 PM

Good post, I agree 100%

ctorg · 04-23-2003, 09:50 AM

I've been posting similar ideas for awhile, but you've simplified it and made a version much easier to implement. I think it's excellent.

Posted by WLight:

Quote:

However, as you note, there are systemic problems with the trade AI in 5.1 that many players feel are of paramount importance. Until those basic issues are addressed I think the implementation of a trade 'feature' such as you suggest will have to wait.

People will complain about the trade AI no matter what. Some think it's too easy, and if it is made harder, others think it's too hard. I think the above idea would make the trade AI more intelligent and more palatable to everyone on both sides. Therefore, I think it is an important step and not something that should wait.

Henry · 04-23-2003, 10:29 AM

I could go along with this

It would add more depth to the equation and filter the trade possibilties to a more realistic result.

Another thing needed, however, is a better understanding of the "star" system. Many of us are basing the "good/bad" of a trade based on stars - and I remember Markus specifically telling us stars can be wrong as well. If that's the case, we need to understand a little how the AI balances trades so we're not off on a tangent trying to figure out why we think it doesn't make sense...

Henry

bsbll_n_mn · 04-23-2003, 11:48 AM

come on, crap trades happen in baseball every day

ex.

Chuck Knoblauch for Eric Milton and Cristian Guzman (i think that was it)

ctorg · 04-23-2003, 11:57 AM

Quote:

Originally posted by bsbll_n_mn
come on, crap trades happen in baseball every day

ex.

Chuck Knoblauch for Eric Milton and Cristian Guzman (i think that was it)

Remember, everyone thought the Yankees got a steal when they landed Knoblauch. In retrospect, it was the other way around.

Sure, crap trades happen all the time, but one should not be able to consistently swindle other teams.

akw4572 · 04-23-2003, 12:17 PM

i like the trading AI settings, would be good for vers. 6. I am sure this would be a major overhaul.

ghulten · 04-23-2003, 01:41 PM

One word of caution. Not every team always accurately understands which of those three phases they're currently in. A great real-life example of this is the Mets, who act like they're a Contending or Pennant Race team but in actuality should be in a rebuilding mode. Another similar example is the Diamondbacks.

A slightly different example might be the Giants, whom many people thought were in a rebuilding mode this winter, promoting young pitchers and letting some high-priced guys go, but their management was actually applying a Pennant Race strategy.

While the three categories you've got listed would seem to be an improvement, I think they might pigeon-hole teams too much. And teams should be capable of making mistakes recogizing what they should be doing, just like real MLB teams. I can see the complaints already: "This AI is broken. The White Sox are obviously a Contending team, but the computer is acting like they're rebuilding."

I think implementing this might open up a whole new can of worms, and needs to be thought through all the way.

Henry · 04-23-2003, 01:49 PM

Quote:

Originally posted by ghulten
One word of caution. Not every team always accurately understands which of those three phases they're currently in. A great real-life example of this is the Mets, who act like they're a Contending or Pennant Race team but in actuality should be in a rebuilding mode. Another similar example is the Diamondbacks.

A slightly different example might be the Giants, whom many people thought were in a rebuilding mode this winter, promoting young pitchers and letting some high-priced guys go, but their management was actually applying a Pennant Race strategy.

While the three categories you've got listed would seem to be an improvement, I think they might pigeon-hole teams too much. And teams should be capable of making mistakes recogizing what they should be doing, just like real MLB teams. I can see the complaints already: "This AI is broken. The White Sox are obviously a Contending team, but the computer is acting like they're rebuilding."

I think implementing this might open up a whole new can of worms, and needs to be thought through all the way.

I don't disagree, however this is a 1st cut

In the final cut, I would suspect you need 5 levels...

1. Rebuilding (liberal trading approach looking for prospects only)

2. Average trading interest looking for younger players, but may be interested in the "right" veteran - low cost

3. Contending (generally conservative about trading unless they get an offer they can't refuse)

4. Average trading interest looking for veteran players, but may be interested in the "right" prospect - 5 star

5. Pennant Race (liberal trading approach looking for veterans only)

The above would give you enough variance, I believe, to avoid any "pidgeonholing"

Henry

ghulten · 04-23-2003, 01:57 PM

Right, but I think there should still be an allowance for teams to mistakenly categorize themselves. Which will lead right back to the same complaints about how the AI isn't omniscient enough.

I mean, if you listen to Dave Littlefield of the Pirates right now, he thinks he's contending. He's done a good job patching some holes with cheap filler, but is he really contending? If he trades prospects for a veteran in July, where does that put him?

If a situation were to happen like that in one of our fictional leagues, people would be complaining incessantly about crappy AI. I think having the computer make self-assessment mistakes is very much a necessary part of any AI. However, right now I don't the AI does ANY self-assessment, and that could be improved.

I just want people to think this all the way through before suggesting it/asking for it. If this is implemented, I think GMs should make mistakes, sometimes horrible, and sometimes often. Hell, look at the Mets. They have no idea where they are in the success cycle, and haven't had a clue since the 2000 WS ended.

EDIT: I guess I don't want to quibble with the actual categories used, but with the idea of categories in general and how the AI fits itself into categories.

ctorg · 04-23-2003, 02:33 PM

Very interesting posts, both Henry and ghulten.

I do think that the AI could very easily make mistakes using the system proposed by Henry. It could think (due to its scout) that it's a contender, when it's really not. Like the Mets.

I think that the mistake is much more often made in thinking that the team is better than it is than the other way around. There aren't too many teams that, given a decent payroll, will start rebuilding when they're in a pennant race, but there are plenty who continue to insist they're in one when they're not (my poor, poor Mets).

Perhaps having an additional scout/coach type, a GM character for instance, would be good for this.

Another Mike D · 04-23-2003, 03:32 PM

Quote:

Originally posted by bsbll_n_mn
come on, crap trades happen in baseball every day

ex.

Chuck Knoblauch for Eric Milton and Cristian Guzman (i think that was it)

This trade was not bad for the Yanks....knobby played well for a bit, scored lots of runs, helped them win WS...Milton has been mediocre and has never lived up to his potential and Guzman is OK, but the Yanks had no where to put him. But, the point of this thread was that a trade like that might happen, but it shouldn't be the last place twinkies trading away prospects for knobby. It happens way too much in OOTP....regardless if the players are fair, the trades just tend to not make sense in terms of team directions.

ghulten · 04-23-2003, 03:38 PM

I posted this as well in the Trade AI thread that's stickied, but:

Please think this through all the way. How is the AI going to evaluate it's current situation? Is the AI going to be good enough to recognize whether it's a contender or a rebuilder? In real life, many team fail to recognize where they are in the "success cycle" (think NY Mets) and insist on acting like contenders when they should be rebuilding. This is being discussed in a little more detail here but needs to be remembered. REAL LIFE GMS SCREW UP all the time while evaluating thier own teams and their chances of contending. I don't know if it's realistic to expect Markus to program an AI that thinks more intelligently than Steve Phillips or Dave Littlefield do (or maybe not ;-)

When these AI GMs screw up their self-evaluation and make trades that are inappropriate for their team's current position, aren't we right back where we started? With teams making trades that make no sense for them and their situation? Such mistakes are realistic, so how do we address that?

Tom Martell · 04-23-2003, 04:10 PM

The options I mentioned don't have to be implemented as 3 distinct settings - it could easily be a spectrum that the AI for each team moves along, deciding where it falls. Also, with manager mode the concept of an owner is introduced. This could easily be expanded to determine where the owner wants the team taken. As teh Mets keep trying to contend, this could be because the owner refuses to authorize a rebuilding phase.

Henry · 04-23-2003, 04:12 PM

Quote:

Originally posted by ghulten
When these AI GMs screw up their self-evaluation and make trades that are inappropriate for their team's current position, aren't we right back where we started? With teams making trades that make no sense for them and their situation? Such mistakes are realistic, so how do we address that?

I think we give the AI every reason to call the right shot - knowing it will still screw it up now and then. That approach is defendable against those that say it's "screwed up". If you do nothing to attempt the right decision, all you've got to hang your hat on is "it's too hard to program" - something that doesn't seem to go over well LOL.

If there was a GM character (as proposed) that made an analysis of what the team "should" be doing (basically placing it in one of the five catagories) - then you've got the "attempt". Heck, you could even have the game write up a preseason report on how each team sees itself (then have fun saying what a dummy it was later in the season!).

The human player will be doing the same thing but not through code (of course) and he maybe has the same ability to screw up - BUT - now the teams have a "character" when it comes to trades. Some will want the young guys, some veterans - some will think your offer stinks - and others will jump at it

Henry

ghulten · 04-23-2003, 04:19 PM

Henry,

I'm OK with that. But in the end, I just see this as programming a roleplaying AI that's going to end up with the same results as we have now. It seems like a tremendous change to something that's already fairly balanced just to get an in-game justification for something that already occurs.

Why not just imagine the same justification now? I suspect the results will be less frustrating... :-)

I don't want you to think I'm advocating giving up on improving the trading AI; it's one of those things that can always get better. I'm just really wary of this notion of "personalities" for AI teams when real-life MLB teams are so schizophrenic every year. Because to program it realistically, it should screw up 25-50-75 percent of the time, and all you're really doing is creating an in-game justification for the screw-ups. Does that make any sense at all?

WLight · 04-23-2003, 04:37 PM

Regarding having AI teams screw up in self-characterization, I think most of the real-life mistakes in that regard (examples given being the Mets and Pirates this year) happen early in the season or in the offseason. I think it'd be hard to have the AI duplicate the current folly of the Mets organization spending loads of money every offseason thinking they're going to contend when we all know their squad is crap and the additions they make are generally ill-advised. But if the game is built to make an assessment each year it shouldn't be too hard. Maybe keep most teams in the average range to begin the year, then have them start making more (or less) aggressive moves based on their payroll, record, etc. as the season progresses. If a team is really hopeless (or really strong) you could have it look back a few seasons to see if they're the Tigers or the Yankees so it doesn't have to start over with self-assessment every year.

Regarding the viability of the Pirates trading prospects for vets in July, I think that would depend on what the NL Central looks like at that time.

ghulten · 04-23-2003, 04:40 PM

Assumption: We want AI teams to have "personalities" or "plans" so that their trading will make more sense.

To illustrate the problem with this approach another way: We're all pretty enlighten baseball fans around here; we understand what OPS and WARP are, and we think that we know what is best for EVERY team in MLB right now. But how many teams in MLB right now would you say recognize what best for themselves? How may current MLB teams would you say have a good plan or a recognizable personality?

I'll try to put together a list, along with what I think they're easily recognizable "plan" or "personality" is:
1. Yankees: If it's good, it must be expensive. If it's expensive, it must be good. We want to be good. Ergo, we want to acquire everything expensive.
2. A's: Pitch counts and college arms for pitching; working counts and power for hitting.
3. Indians: Acquire as many talents that will be ready in 2005 as possible. Discard everything that won't help us win in 2005.
4. Twins: Win now with the talent we've developed from within before they get too expensive to keep.
5. Angels: make contact, make contact, make contact, and play defense. (I'm not even sure this counts, but)
6. Giants: put together as solid a supporting cast as possible for Bonds for around $65m.
7. Padres: See Indians, as related to their new ballpark, but not to such an extreme.
8. Atlanta: pitching, pitching, pitching, and two doses of Leo Mazzone.

That's eight teams out of 32 in real MLB that I would say have a readily identifiable personality that permeates their organization and guides everything they do. And yes, I'm sure that someone with more local knowledge than me will claim that there local team has some plan or personality; I just get all my information from Baseball Prospectus 2003, and you're probably more right than I am.

But somewhere around 24 of 32 teams just muddle along and hope they get lucky, at least that's how I see it. I just think any discussion of "personalities" for AI teams in OOTP needs to take into consideration the amazing shifting priorities of two-thirds of the teams on any given day.

Your thoughts?

04-22-2003, 04:56 PM	#1
Tom Martell Minors (Rookie Ball) Join Date: Mar 2002 Location: Stanford, CA Posts: 24	AI Trading Mindsets I know I've mentioned this before, but given all the ongoing complaints about the trading AI... Implement 3 distinct modes for the for each team. Create a simple algorithm based on current prospect list (there already is the team system rankings, so teams know how good they are relative to other teams), current record and divisional standing, and current major league talent that sends the team between the different mode. 1) Pennant Race - the team will attempt to acquire major league at the expense of its minor league system to fill holes in the lineup or rotation. 2) Contending - the team has no trade modifiers - it won't be anxious to deal away its prospects or its major leaguers because it is hoping to both contend now and build for the future. 3) Rebuilding - AI takes the view that major leaguers are worthless and eat money. the team is anxious to dump high salary players and acquire top minor league talent, intent on making a run in 2-3 years. Nothing is more ridiculous than being able to trade an old, declining star for top tier talent from teams that are in last place and 35 games out of first. They should have no interest whatsoever.

04-23-2003, 01:57 PM	#12
ghulten Minors (Double A) Join Date: Jan 2003 Posts: 135	Right, but I think there should still be an allowance for teams to mistakenly categorize themselves. Which will lead right back to the same complaints about how the AI isn't omniscient enough. I mean, if you listen to Dave Littlefield of the Pirates right now, he thinks he's contending. He's done a good job patching some holes with cheap filler, but is he really contending? If he trades prospects for a veteran in July, where does that put him? If a situation were to happen like that in one of our fictional leagues, people would be complaining incessantly about crappy AI. I think having the computer make self-assessment mistakes is very much a necessary part of any AI. However, right now I don't the AI does ANY self-assessment, and that could be improved. I just want people to think this all the way through before suggesting it/asking for it. If this is implemented, I think GMs should make mistakes, sometimes horrible, and sometimes often. Hell, look at the Mets. They have no idea where they are in the success cycle, and haven't had a clue since the 2000 WS ended. EDIT: I guess I don't want to quibble with the actual categories used, but with the idea of categories in general and how the AI fits itself into categories. Last edited by ghulten; 04-23-2003 at 02:02 PM.

04-23-2003, 02:33 PM	#13
ctorg Global Moderator Join Date: Nov 2002 Location: Queens, NY Posts: 9,848	Very interesting posts, both Henry and ghulten. I do think that the AI could very easily make mistakes using the system proposed by Henry. It could think (due to its scout) that it's a contender, when it's really not. Like the Mets. I think that the mistake is much more often made in thinking that the team is better than it is than the other way around. There aren't too many teams that, given a decent payroll, will start rebuilding when they're in a pennant race, but there are plenty who continue to insist they're in one when they're not (my poor, poor Mets). Perhaps having an additional scout/coach type, a GM character for instance, would be good for this. __________________ My music "When the trees blow back and forth, that's what makes the wind." - Steven Wright Fjord emena pancreas thorax fornicate marmalade morpheme proteolysis smaxa cabana offal srue vitriol grope hallelujah lentils

04-22-2003, 07:33 PM	#2
WLight Major Leagues Join Date: Jan 2003 Location: CA Posts: 314	I think this is a great idea and a lot of support for it has already been expressed. However, as you note, there are systemic problems with the trade AI in 5.1 that many players feel are of paramount importance. Until those basic issues are addressed I think the implementation of a trade 'feature' such as you suggest will have to wait.

04-22-2003, 08:08 PM	#4
Bears5122 All Star Reserve Join Date: Mar 2003 Posts: 847	Good post, I agree 100%

04-23-2003, 10:29 AM	#6
Henry Hall of Fame Join Date: Dec 2001 Posts: 6,498	I could go along with this It would add more depth to the equation and filter the trade possibilties to a more realistic result. Another thing needed, however, is a better understanding of the "star" system. Many of us are basing the "good/bad" of a trade based on stars - and I remember Markus specifically telling us stars can be wrong as well. If that's the case, we need to understand a little how the AI balances trades so we're not off on a tangent trying to figure out why we think it doesn't make sense... Henry

04-23-2003, 11:48 AM	#7
bsbll_n_mn Registered User Join Date: Apr 2003 Posts: 593	come on, crap trades happen in baseball every day ex. Chuck Knoblauch for Eric Milton and Cristian Guzman (i think that was it)

04-23-2003, 12:17 PM	#9
akw4572 Hall Of Famer Join Date: Jul 2002 Location: Kentucky Posts: 2,601	i like the trading AI settings, would be good for vers. 6. I am sure this would be a major overhaul.

04-23-2003, 01:41 PM	#10
ghulten Minors (Double A) Join Date: Jan 2003 Posts: 135	One word of caution. Not every team always accurately understands which of those three phases they're currently in. A great real-life example of this is the Mets, who act like they're a Contending or Pennant Race team but in actuality should be in a rebuilding mode. Another similar example is the Diamondbacks. A slightly different example might be the Giants, whom many people thought were in a rebuilding mode this winter, promoting young pitchers and letting some high-priced guys go, but their management was actually applying a Pennant Race strategy. While the three categories you've got listed would seem to be an improvement, I think they might pigeon-hole teams too much. And teams should be capable of making mistakes recogizing what they should be doing, just like real MLB teams. I can see the complaints already: "This AI is broken. The White Sox are obviously a Contending team, but the computer is acting like they're rebuilding." I think implementing this might open up a whole new can of worms, and needs to be thought through all the way.

04-23-2003, 03:38 PM	#15
ghulten Minors (Double A) Join Date: Jan 2003 Posts: 135	I posted this as well in the Trade AI thread that's stickied, but: Please think this through all the way. How is the AI going to evaluate it's current situation? Is the AI going to be good enough to recognize whether it's a contender or a rebuilder? In real life, many team fail to recognize where they are in the "success cycle" (think NY Mets) and insist on acting like contenders when they should be rebuilding. This is being discussed in a little more detail here but needs to be remembered. REAL LIFE GMS SCREW UP all the time while evaluating thier own teams and their chances of contending. I don't know if it's realistic to expect Markus to program an AI that thinks more intelligently than Steve Phillips or Dave Littlefield do (or maybe not ;-) When these AI GMs screw up their self-evaluation and make trades that are inappropriate for their team's current position, aren't we right back where we started? With teams making trades that make no sense for them and their situation? Such mistakes are realistic, so how do we address that?

04-23-2003, 04:10 PM	#16
Tom Martell Minors (Rookie Ball) Join Date: Mar 2002 Location: Stanford, CA Posts: 24	The options I mentioned don't have to be implemented as 3 distinct settings - it could easily be a spectrum that the AI for each team moves along, deciding where it falls. Also, with manager mode the concept of an owner is introduced. This could easily be expanded to determine where the owner wants the team taken. As teh Mets keep trying to contend, this could be because the owner refuses to authorize a rebuilding phase.

04-23-2003, 04:19 PM	#18
ghulten Minors (Double A) Join Date: Jan 2003 Posts: 135	Henry, I'm OK with that. But in the end, I just see this as programming a roleplaying AI that's going to end up with the same results as we have now. It seems like a tremendous change to something that's already fairly balanced just to get an in-game justification for something that already occurs. Why not just imagine the same justification now? I suspect the results will be less frustrating... :-) I don't want you to think I'm advocating giving up on improving the trading AI; it's one of those things that can always get better. I'm just really wary of this notion of "personalities" for AI teams when real-life MLB teams are so schizophrenic every year. Because to program it realistically, it should screw up 25-50-75 percent of the time, and all you're really doing is creating an in-game justification for the screw-ups. Does that make any sense at all?

04-23-2003, 04:37 PM	#19
WLight Major Leagues Join Date: Jan 2003 Location: CA Posts: 314	Regarding having AI teams screw up in self-characterization, I think most of the real-life mistakes in that regard (examples given being the Mets and Pirates this year) happen early in the season or in the offseason. I think it'd be hard to have the AI duplicate the current folly of the Mets organization spending loads of money every offseason thinking they're going to contend when we all know their squad is crap and the additions they make are generally ill-advised. But if the game is built to make an assessment each year it shouldn't be too hard. Maybe keep most teams in the average range to begin the year, then have them start making more (or less) aggressive moves based on their payroll, record, etc. as the season progresses. If a team is really hopeless (or really strong) you could have it look back a few seasons to see if they're the Tigers or the Yankees so it doesn't have to start over with self-assessment every year. Regarding the viability of the Pirates trading prospects for vets in July, I think that would depend on what the NL Central looks like at that time.

04-23-2003, 04:40 PM	#20
ghulten Minors (Double A) Join Date: Jan 2003 Posts: 135	Assumption: We want AI teams to have "personalities" or "plans" so that their trading will make more sense. To illustrate the problem with this approach another way: We're all pretty enlighten baseball fans around here; we understand what OPS and WARP are, and we think that we know what is best for EVERY team in MLB right now. But how many teams in MLB right now would you say recognize what best for themselves? How may current MLB teams would you say have a good plan or a recognizable personality? I'll try to put together a list, along with what I think they're easily recognizable "plan" or "personality" is: 1. Yankees: If it's good, it must be expensive. If it's expensive, it must be good. We want to be good. Ergo, we want to acquire everything expensive. 2. A's: Pitch counts and college arms for pitching; working counts and power for hitting. 3. Indians: Acquire as many talents that will be ready in 2005 as possible. Discard everything that won't help us win in 2005. 4. Twins: Win now with the talent we've developed from within before they get too expensive to keep. 5. Angels: make contact, make contact, make contact, and play defense. (I'm not even sure this counts, but) 6. Giants: put together as solid a supporting cast as possible for Bonds for around $65m. 7. Padres: See Indians, as related to their new ballpark, but not to such an extreme. 8. Atlanta: pitching, pitching, pitching, and two doses of Leo Mazzone. That's eight teams out of 32 in real MLB that I would say have a readily identifiable personality that permeates their organization and guides everything they do. And yes, I'm sure that someone with more local knowledge than me will claim that there local team has some plan or personality; I just get all my information from Baseball Prospectus 2003, and you're probably more right than I am. But somewhere around 24 of 32 teams just muddle along and hope they get lucky, at least that's how I see it. I just think any discussion of "personalities" for AI teams in OOTP needs to take into consideration the amazing shifting priorities of two-thirds of the teams on any given day. Your thoughts?