Trade AI Discussion Master Thread (trying to gather everything in one place)

[Bill]

Originally posted by Henry
Let's look at the situation... Trade screen comes up and a list of players is presented by the other team.... you see a 5-star player listed and immediately claim the AI is flawed because you consider what your offering is junk.

The fallasy of this approach is we are measuring the AI's position based on what our own would be. That's the problem with the entire trading debate. Everything we complain about is based on what WE would do. Granted, if every GM was ME then every trade would be to my expectation.

In real life, every GM is NOT a clone of every other GM. Each has a different personality, a different set of priorities, a different knowledge level of baseball, a different team to work with, a different salary structure, and then some are also just plain stupid. As a result, the trade will not always look to you like something you would have done.

Looking at the specific TRADE is not the way to determine if the AI is flawed. Looking at the results is. If, over the course of many seasons, a majority of players of this game can rip off the other teams through trades and consistently win - then the issue is the trade AI needs to be tougher. If, on the other hand, only 5-10% of the players of this game can do that - and another 5-10% can't ever win - then I would call the AI just about right.

Looking at each trade is the wrong way to analyze this... looking at the long term results IS the way to analyze this. To spend too much time on the balance of a single trade is to waste our time in solving the real problem..... if there is one.


Henry

The above was written by Henry in another thread...

There have been some good discussions going on about the Trading AI in this and other threads. I happen to agree with what is said above, Has anyone tested the long-term implications of these trades being made? We just may be trying to solve a problem that is not a problem. I am not saying that the Trading AI should not be improved upon.

MrWorkate had some excellent suggestions as did wipethatsmirk...

I have also read several times that what one person thinks is a bad trade, someone else thinks is a decent trade. When there is a differing of opinion, that in itself tells me that overall the Trade AI is very satisfactory.

My main question is what are the long-term results of these trades as Henry points out. I have not tested this enough to come up with any conclusions.

[hukarez]

Originally posted by Tycobbler
I get the same thing in my game, but i like to look at it like this. I don't mind teams offering their oldsters for my young studs. Even though i won't make a move like this, i tend to look at it as the team sending me their old guy is either trying to 1. make room for young kids. 2. Trying to clear up some money to resign their own guys. 3. Testing the market to see what they can get for these guys.
Those are just a few quick ideas i like to believe to keep the game as interesting and fun as possible.

While I can't provide any examples, I can recall times when I tried to reciprocate the good gesture with other teams, via offering one of my players in decline for a decengt prospect or two. A simple 3 star and 2.5 star prospect, for a 2.5 star oldie. I was promptly shot down. Of course, most of the trades that I seem to engage in just aren't viable for me...sometimes I do wish there would be an option where I'd be able to state that I'm not interested in receiving players on their decline, and ready to retire at the end of the season or so. I guess I need to re-think my approach to the whole A.I. thing...but I have to admit, I was a bit unsettled by it.

[mtw]

After now trying into the hundreds of trades to see the working of the AI, the only consistently identifiable flaw I see is the value placed on 1-star and 2-star veteran players. I can usually offer a 1 or 2 veteran star player at any position and get a 3 star or better prospect back, and sometimes even a better player at the same position, especially if there is a need. If there is a need, I cna still get this value even if the player I trade will not start for the AI team.

This, along with observations about other trades involving higher value players I "proposed", has led me to these two general conclusions, which are somewhat related.

1) Slightly too much value to veteran players as opposed to prospects. By example, for AI teams with "starting pitcher" as a need will often trade me good (i.e. young and with some ratings) 4-star SP prospects, even if the SP I trade is no better than other pitchers in the rotation, and often even if the player would not make the rotation. That's a little too much value for a veteran in any circumstance IMO.

2) Somewhat too much emphasis on need as opposed to value. Ideally, the AI would see what it's depth and lineup will look like pre- and post-trade. The difference in the quality of the depth/lineup/rotation should be the value of the trade to the AI.

By example, a 1-star 3rd string 3B should have almost no value, and nothing better than a two star prospect should be offered, and only that if the the 3B is making a very small amount of money and will make the major league roster. A backup should be valued less than a starter, and 3rd in the depth chart (or for pitchers, mop up reliever) should have very little value, unless that player will be the 1) primary LH or RH pinch hitter or 2) primary pinch runner, in which case the player should be slightly more valuable.

Again, I am very impressed with the AI, and think it is only small adjustments away from being as good as possible in trade balancing.

Once the balancing issues are addressed, I would like to see trade scenarios implemented, which would include things like:

1) In playoff hunt--slightly more interest in upgrading major league talent, especially position player starters and in the bullpen. Slightly more willing to trade prospects and less willing to trade it's starters.

2) Out of playoff hunt--slight emphasis on getting prospects and ridding itself of impending free agents and older veteran players; less likely to add salary.

3) Financial problems--slight emphasis on dumping salary with emphasis on getting prospects; very unlikely to add salary.

4) If this is possible--"Change of Scenery" trades--players who are underperforming compared to expectations--this would take some statistical analysis, but if a player has been underperforming compared to where he "should" be (taking into account last season and perhaps park effects), the AI is somewhat more likely to trade that player.


BTW- I like this "sticky" thread, and would like to see more such threads for other major game aspects, to create a central repository for ideas and feedback on such as player development, AI roster management, and played game (PbP, AI strategy).

[dougaiton]

This is my first post, so I'm sorry if my point seems stupid!

I think the trade AI works very well, but there is one particular thing that I feel might be tightened, and its about boring-old stability. I recently completed a trade with the Giants who were 10 games ahead in their division race, approaching the trade deadline. I then completed a trade with them for 2 of their starters and their ace in return for 4 of my players and a prospect. All well and good in the offseason, or if both teams were out of contention. The Giants, in real-world terms, would putting themselves at considerable risk by trading a third of their succesful team (especially two months before the playoffs).

Stability (and in some cases stagancy) is a major part of baseball - while much is made in the amount of players who leave teams after World Series wins, but mostly these are role players. After a succesful season, perhaps teams would make a major effort to tie in their players, even going above market value (look at the Yankees' loyalty to Sojo, Knoblauch and Brosius throughout their championships, even when they offesively tailed off). Perhaps it could be linked to 'fan interest'.

Sorry about the gripe, which is a minor one in otherwise a very well rounded AI. But society fears change - maybe pennant contenders should too!

[Another Mike D]

I'm going to chime in a bit here. I've noticed that the computer will strip itself at various positions it "doesn't need" to fill that one position they do need. For example they will need a 2B and trade away a star OF and 3B to get that 2B, then I check on their trade screen again and now they need OF and 3B. It would be like the Yankees needing a 3B, so they trade Jeter and Bernie Williams to get Scott Rolen (just an example, relax).

Also, from what I understand, despite occasional posts to the contrary, the computer AI doesn't "rebuild" or "go for it" at all. I've seen teams 25 games out at the deadline acquire high priced veterans for youngsters and 1st place teams trading away their cleanup hitter for a few mediocre guys. The game seems to place little emphasis on standings and overall ages when trading amongst themselves and with me. It is a huge improvement from v2 and v3 where teams would stockpile guys they don't need, but now I think the focus has shifted too much on what they need, so much that you can strip their team to help them fill a need..

Lastly, I hear everyone hear hungup on stupid trades made in real life and that is valid...but the system of trading is *usually* consistent. Teams that are 25 games out almost never acquire high priced veterans and teams in 1st place almost never trade their superstars...

[earlg]

If this has been mentioned before I apolize but one thing I would like to see changed is how a trade is proposed. Right now you add a player at a time and the computer gives you an answer. If the computer accepts it you can then add a player and see if it would still accept the trade. I would rather put all the players I want involved and then submit the trade. If the AI accepts it you have the choice of making the trade or pulling the offer.But if you pull the offer it should make the AI mad for offering the deal and then having second thoughts. I like a lot of the things that mrworkrate mentioned for dealing with other teams.

[Erithtotl]

Lets look at the following issues:

1) AI teams in OOTP5 do NOT consider the current state of compeditiveness of their team in trades. The same trades are made whether they are 20 games out or 20 games in first place. If they do take this into account, it is a very, very small amount.

2) AI teams do not have 'plans' in any sense of the word like real world teams do. By that I mean that they do not go into 'rebuilding' mode, or 'contention' mode, or 'almost there' mode, or 'decline mode' like most real teams. Their moves are based entirely on evaluating player value and their current roster.

3) AI teams do not have 'personalities'. They don't build around pitching, or defense, or their ballpark, or OPS, or whatever.

By accepting these issues, we can look at some of the other assertations that people have made. The argument 'teams do dumb things in real life' is true. But the AI is already working at a huge disadvantage to real life GMs because it lacks the sense of 'state' and planning that a real GM would. To further curse it with the boneheaded moves of a Cam Bonifey is not fair.

Its clear that the AI does not value prospects the way real teams do. Yes, the farther from the majors, the less valuable a prospect is, but the AI needs to recognize players who are 'replacement level' talent. A 33 year old 2 star starting pitcher with a 4.9 career ERA is 'replacement level'. He is not worth giving up any appreciable talent for. Yes, in real life there are GMs that will make this kind of mistake (Kenny Williams trading Kip Wells and Josh Fogg for Todd Ritchie, for example), but since the human player is going to be smarter than ALL the AI GMs just by default, making some AI GMs do dumb things in addition to their other disadvantages doesn't make a fun game.

I think the AI between 5.00 and 5.01 was the best we've seen in OOTP5. 5.00 was by far the hardest trading AI OOTP has had, but it had a few flaws. Prospects were impossible to trade, and impossible to trade for. The AI in 5.01 had a lot of improvements. You could trade a 4 star prospect you didn't need for a 2.5 star vet you did. You could trade a 4 star vet for a couple 4 star prospects. But it did have some holes. You could trade a 5 star 35 year old vet for 5 - 5 star prospects sometimes. Or sometimes you could trade a 2.5 star vet for a 4.5 star prospect, when the AI didn't really need your 2.5 star vet.

ASsuming the 3 items I listed at the top cannot make it into OOTP5.x, then I would ask for an AI that had the difficulty of 5.00 but with the ability to trade away and trade for prospects, but at a reduced level from 5.01. It is vetter the AI err on the side of caution (and difficulty). Or even better, have a trade difficulty AI toggle. Simply adjust it and it weights the AI players more valuable than the human's players. This alone might help a great deal.

[J P Falcon]

One additional aspect that cannot be ignored is how trading is closely tied to a teams finances...as the seasons progress in my solo league, the AI trading becomes tighter and appears to work better than in the early seasons...one reason is that the AI teams have a hard time keeping their finances in a positive balance...in 2011 I have 23 AI teams with a negative cash balance....3 teams have a positive balance of 1.9 million or less, 1 team less than 2.9 mil, 1 team less than 3.9 mil and 2 over 4mil. You start seeing "We cannot afford... or so and so just eats $$$" responses due to the lack of cash. Those early trades, where the AI gives up prospects (whether they pan out or not) for older expensive players tends to haunt them financially later on. Releasing expensive players who refuse to go to the minors (some of which were traded for) also affects the AI fianances. Henry's points are very good and I alluded in an earlier post that the fault may be in my perception of the trades. However when he mentions the long term effect of trades, the financial hole that an AI team puts itself in is one of those consequences. Though finances might be considered by some to digress from the main point of this thread, I think they are tied to the long term financial problems of the AI teams. They take on too much salary in the early years, then are strapped for cash later on.

[cfeedback]

One solution that I've always thought would work well would be to create a trade strategy setting that a human GM could set and that AI teams would try to set (somewhat) intelligently for themselves...examples:

Win At All Costs
Contention
Steady/Neutral
Rebuilding
Dump Salary
Etc...

This does not seem like it would be anymore difficult to program than the computer setting it's own prices/team strategies, etc. Once this was done, it would be easy to tell the AI not to trade high salary players to a team in the "Dump Salary" mode or vets to a team that's rebuilding. Just my $0.02 on the subject.

also...

2) AI teams do not have 'plans' in any sense of the word like real world teams do. By that I mean that they do not go into 'rebuilding' mode, or 'contention' mode, or 'almost there' mode, or 'decline mode' like most real teams. Their moves are based entirely on evaluating player value and their current roster.

ASsuming the 3 items I listed at the top cannot make it into OOTP5.x, then I would ask for an AI that had the difficulty of 5.00 but with the ability to trade away and trade for prospects, but at a reduced level from 5.01. It is vetter the AI err on the side of caution (and difficulty). Or even better, have a trade difficulty AI toggle. Simply adjust it and it weights the AI players more valuable than the human's players. This alone might help a great deal.

I'm hoping that assumption is wrong...I don't follow this board very closely, but I'm always extremely pleased with any OOTP patch that's released...I always marvel at the amount of things that are improved upon. IANAP but a team plan/philosophy type setting does not seem like it would be difficult to impliment to me...keep it simple, just give the AI a few settings, just enough to give it some guidance but not too many that it can't pick the right one. If the AI can't decide that it needs to be set to rebuilding mode when it's 30 games out in June then we're probably screwed.

[anpham]

When you first take a look at trading in OOTP or if you’ve played other games that have player trading then it looks rather simple and it can be. You add some players to a list and if your total value is equal to or greater than the players you’re asking for then the other team should approve it.

For instance let’s say you want to add in future draft picks as an option in trading. Let’s say there are 100 picks in the draft and the 1st pick is worth 100 points, 2nd pick is worth 99, last pick worth 1 point, and so on. To get the 1st pick in the draft you would have to give up at the very least a 50th and 51st pick. I doubt anyone would trade away the first pick for two middle round picks since the earlier picks are extremely weighted but I’m trying to keep it simple. This makes your offer to the other team 101 for their 100 so a computer team will see your offer as being better and it’ll give you the thumbs up. If you’ve played Madden NFL 2003 then you know what I’m talking about. Writing this code would take very little time and the computer could run it in milliseconds. Now when I’m trading with OOTP I see that it doesn’t take milliseconds which means it isn’t just a few lines of code. Even a second to process a query is considered a long time on today’s computers. Now if you try and do a trade that requires the computer to think, for example trading for a superstar on another team, then it has a tendency to crash the game when you add the fifth player. If you trade for a junky player then the fifth player bug doesn’t appear because the computer has already approved the trade on the first person on your list and exits the code immediately. If the computer hasn’t approved the trade by the first player then it has to validate each player to see if the approval total has been reached and when it reaches the fifth person it’s done more calculations than it can handle and causes the entire game to crash. This is either because the math is wrong and put the computer into a loop, or because it just wasn’t able to handle all the error checking with 5 players. But I’m assuming the math isn’t wrong since it still works up to the fourth person which technically means it should work with hundreds of people. So what we have I believe is a ton of code.

I see people keep asking for more slots, more options, full league trade features, etc and all I see are more lines of code added to something that already has a huge amount of code in it already. Some of these suggestions obviously wouldn’t cause much stress on the computer since it isn’t just one central function but some of these ideas will involve integration with the main one. However it does seem that the current state of the trade AI may have reached a certain limit when it comes to being compatible with today’s PCs. It’s actually at a point where certain features might have to be slimmed down a little to prevent the amount of CPU that is used on a trade query so it doesn’t cause any crashes at all. When I say slimmed down I mean the trade AI should be less smart. A lot of people complain about being able to trade junk for top prospects but that is a problem caused by adding too many elements to try and give each team a different value to each player you offer them. What this means is two teams have similar 5 star outfielders. However one team is having salary cap problems while the other is not. Because of this the trade package you offer the two teams are obviously going to be different. One could consist of high priced veterans while the other one is full of prospects. However this is just one scenario. Think of all the different scenarios that could possibly arise out of a sport like baseball. If you go with the general 1 + 2 = 3 formula then you can’t do the trade I just mentioned. If you do go with that 1 + 2 = 3 formula then you won’t have any top prospects being undervalued since their value will be constant and won’t vary from team to team. But I’m pretty sure that most people like the idea of each team trying to take on a different personality when it comes to trading so it’s going to open what seems like major flaws at times. These flaws can be patched when found but some flaws are not always visible until someone mentions it. But if you want to patch a flaw then it requires more code and possibly more CPU resources. And sometimes patches have a tendency to create new flaws in other areas. It can be a vicious endless cycle.

Let’s say you’re playing Madden NFL 2003. The score is tied and it’s late in the 3rd quarter (slightly different because each quarter is 5:00). The opponent has just crossed the 50 yard line and is about 10 yards from field goal range. It’s fourth down and less than a yard to go. If you go for it and make it then you’re in business, but if you don’t cross the first down marker then you just gave your opponent real good field position because you could have pinned them below the 20 with a punt. Obviously a real live game is going to be different but we have to think in terms of how a computer has to approach this scenario because it isn’t human and it has to be consistent on how it responds to certain game situations. This situation did happen to me before so I’m not making this up. It’s the third quarter and I just stopped them on third down but the third quarter has less than 10 seconds remaining. They come out with their punting formation and get ready to punt but the quarter ends before they could get the punt snap off. So the fourth quarter begins and I assume they’re still going to punt but now they decide to go for it. Because of this I’m forced to call a timeout so I can get my punt return unit off the field and put a real defense in their to stop them.

Did the computer really change its mind over the course of 10 seconds on whether or not to punt? No it didn’t. All that happened was a line of code that altered the computer’s play calling because one variable changed from quarter=3 to quarter=4. The computer can’t think for itself. The programmer has to tell the computer what to do in almost every situation or write code that generally solves a type of problem. So if you think of how many different trade scenarios are possible then it’s nearly impossible to cover them all without adding too much code into the trade AI.

Madden NFL added what looks like minimal code to me that just says. If it’s the third quarter you punt it. If it’s the fourth quarter you go for it. Even if there’s only one second separating the third and fourth quarter you still punt it. A real coach would just see the one second and say we’re going for it because there really isn’t much of a difference. But a computer has to draw the line somewhere and so it decides to draw the line at the fourth quarter.

From what I’ve seen of OOTP so far it isn’t a third quarter or fourth quarter situation. It’s about the total amount of time remaining in the game altogether. Madden NFL chose the slimmed down version (slimmed down but at least they covered it) instead of a more situational approach that involves calculating the total time remaining, their 4th down conversion rate, the crowd noise, or basically every variable available. Will most consumers notice this difference? Not really. But it does sort of demonstrate that extra code was added to the program that made it seem like the computer “changed” its mind. Did the programmers of Madden NFL 2003 intentionally write this code hoping that the consumer was going to notice the change of strategy from the 3rd to the 4th quarter? No because it doesn’t happen often enough while playing to warrant any time being spent on it. It just so happened that they just wrote general code on how to handle this situation and decided to just draw the line in the 4th and they happened to have quarter clock run out on them before they could get the punt off. It was basically an accident that looked cool because it looked like the computer changed it’s mind and went for it like in real life. But if you’re trying to write a program that does something where you’ve figured out one hundred ways of responding when there are a thousand of things to respond to then obviously it will never add up to 100%. If you give it the 1 + 2 = 3 way then that’s all you need and it will cover all one thousand scenarios. But if you try to attempt to code an AI then the computer will have to select the functions available that result in the closest match. Sometimes it matches up perfectly, other times adequate, occasionally way off, but most of the time it is more that sufficient. In terms of sufficient I mean I have seen what I can get out of the trade AI so far so I understand why it behaves the way it does sometimes.

I do believe that extra code is being added to the trade AI as of 5.10 with some room to spare but it is quite possible that eventually there could be a “salary cap” in effect so that it doesn’t use as much CPU. I’m not exactly sure on this since I don’t know how it truly works but just seeing the trade screen crash occasionally leads me to believe that it is already at a point where the whole thing needs to be looked at before adding new things to it. One of those moments where a programmer now has all the ideas in front of him because of threads like these and wonders if he should just re-code the entire thing now that he knows what people want. Would I like to see things added? Of course. Most of the ideas in here would make it better. Is it pretty good the way it is? Yes it is and most people here think the same as well. But realistically I’m not the one doing the programming so I’m not aware of what is feasible with the current code. This thread was created for the purpose of sharing ideas and from what I’ve see there aren’t really any major gripes, mostly suggestions. The fifth player crash could be an isolated event and there could in fact be plenty of room to spare to add more trade code. But whenever I see that “Thinking” screen appear and that pause for several seconds, it does kind of make me wonder why it would take that long to go through each of a team’s players (let’s say 60) and find just one player that this team would be willing to do a one for one swap with since that’s what the computer does initially when you ask for one of their players. Without seeing the code I can only guess that it is doing quite a bit.

My idea in my previous post was pretty simple. It didn’t add anything to the trade AI code itself and wouldn’t bog it down one bit. It was basically an after-trade effect on the prospects which would balance out any chance of there being trades where one team was continually coming out ahead every time. People who would complain about trading their 2-star veteran for a 5-star prospect would find out later that all they got was a 3-star rookie performing as a 1-star over and over again. It is more of a band-aid than a true solution but technically it does happen in real life so it is somewhat legitimate. Some people want new features and a smarter AI engine while others just want things to be fixed. Most people want both and I do too. However I do think there should also be a little more emphasis on the discussions where people have noticed something that was completely off and possible ways to fix it without requiring the computer to think as much while at the same time providing solutions that would help the program be useable to as many people as possible.

[euphoria13]

Quote:
"My suggestions for trade improvements are:
1. Value young potential stars more highly
2. Have a league-wide "Shop Player" function
3. Include draft picks (assuming that they can be evaluated properly)
4. What lynchjm24 said"


I wouldn't include trading draft picks. That isn't even legal in the major leagues.

[dougaiton]

Just a thought about the ways that these systems of team agendas ('rebuilding', 'contender') may work in connection with the scouting system.

Presumably the scouting system means that a 'fair' scout will grade a player in a different way (presumably better or worse) than an outstanding scout (given the premise that this scout will be closer to a value-free truth). What if we were to add our own agendas to these valuations? If their was an 'Assign instruction' role for the scout, you could then instruct him to be interested in hitting etc. (which presumably would be of use when simming seasons) over pitching, perhaps in a new general rating column (not prospect or overall, but scout's). Thus, an aspiring Rockies GM could realise that the days of signing pitchers like Hampton and Neagle are over, and bring in some bats.

If this was a possibility, certain other variables could be judged differently - for example, the scouts could rate a 4 star prospect as a better player than a three star outfielder. Same with age, and such like. Thus, the scouts would come to skewed valuations as regards our preferences.

Transfer this to computer GMs, and their scouts, and they could evalaute trades dependent on their own settings. how you would get these guys to select their 'setting' is a subject for someone smarter than me!

As I said, this is just a stream-of-consciousness witter about something that occured to me, and may be within the possibilities. Any thoughts? I understand its probably totally unfeasible and deeply flawed, and offering ideas for the best-programmed baseball computer game in history is like telling Da Vinci he should have used pastels, but hey - that's what this is for!

[locust9]

Not sure if this has been addressed but it's plagued this otherwise great series in all its incarnations. Basically the example is this: asymmetry in player ratings usually means they will overperform, and symmetry means they underperform.

For example, there's a player in my current league who is a 10 average and an 11 in HR, but he continually underperforms players who are at like 5 av and 8 hr, and his stats are usually on par with players far, far below his caliber. The nearest I can figure, the two stats (for power and av, in this case) are sort of competing with one another. I recall this being a problem when we were working on a baseball simulation a long, long time ago.

This happens over and over and over again. I've simmed something like 500 years of ootp, and it's probably the thing that makes me craziest. Are there any fixes for it? Anyone even notice this at all except for me? Gimme a sign. .

2nd - I think the hall of fame should really be able to admit any field that the player season stats are sorted by. The old standbys for the HoF in rl baseball aren't the best filters for quality OO players. I'd prefer to have the lifetime .900 ops, for instance, instead of a .280 average, or a lifetime .400 obp and lifetime .500 slg. Something that gives the stat freaks out there their due.

Thanks for listening, and thanks for the best baseball sim ever made.

loc
http://www.the-elemental.net

[Smellyface]

Originally posted by locust9
[B]Not sure if this has been addressed but it's plagued this otherwise great series in all its incarnations. Basically the example is this: asymmetry in player ratings usually means they will overperform, and symmetry means they underperform.

For example, there's a player in my current league who is a 10 average and an 11 in HR, but he continually underperforms players who are at like 5 av and 8 hr, and his stats are usually on par with players far, far below his caliber.

Nope, I personally have never noticed this. I have had many guys who are 8s or 9s at avg and 10 or better at homers and perform amazingly. I have also seen some who don't do as good as you would think they should, but this is the same as it is for the guys who are 5 at avg and 8 at HR. Some guys outperform their numbers and some guys never do quite as well as you think they should. This is just my observation from many many years simmed, and I like it.

[wipethatsmirk]

Posted this in the regular forum, then realized this was the best place for it. Ignore if you've read it already

A cautionary tale regarding the new trade interface…

Just for the heck of it, I wanted to see what I could get for my single-A catcher James Vong, a fictional player created to fill the minors. The saga follows.

C James Vong (20 yrs, 3/4/3, 1 blue star, undrafted)
Traded to Texas for
SP Eric Good (22 yrs, 0/2/4/5, 3 ½ blue stars, 34th round pick, 1009th overall)
Traded to Montreal for
MR Blaine Neal (24 yrs, 6/7/8/5, 4 ½ blue stars, 15th round pick, 424th overall)
Traded to Chicago (N) for
MR Joe Valentine (23 yrs, 6/7/9/4, 5 blue stars, 10th round pick, 277th overall)
Traded to New York (A) for
1B Justin Morneau (21 yrs, 4/4/4, 5 blue stars, 13th round pick, 387th overall)
Traded to Tampa Bay for
3B Joe Crede (24 yrs, 5/9/1, 3 ½ gold stars, 15th round pick, 429th overall)
Traded to Montreal for
CL Billy Wagner (31 yrs, 9/8/7/6, 5 gold stars, 8th round pick, 214th overall)
Traded to Florida for
SP Bartolo Colon (29 yrs, 6/6//6/5, 3 ½ gold stars, 4th round pick, 92nd overall)
Traded to New York (A) for
CF Carlos Beltran (25 yrs, 5/6/5, 4 gold stars, 1st round pick, 27th overall)
Traded to Colorado for
RF Bobby Abreu (29 yrs, 7/6/8, 4 gold stars, 1st round pick, 18th overall)
Traded to Minnesota for
RF Sammy Sosa (34 yrs, 7/12/7, 4 ½ gold stars, 1st round pick, 30th overall)

(Using OOTP 5.10, Total Minors 1.1, and of course a fantasy draft)

Yes, I’d rather have Beltran or Abreu than Sosa, but I decided to end the chain with Sosa to get more people to read this

The new “Players that would make the deal work” window, while helpful in some ways, makes the above sort of thing too easy to do. The point is, I wish I weren’t able to do this, and I’d like to see the loophole closed somehow.

Hoping this will stir up some more discussion about the trade AI…

[TwinMonster]

I have to say I like the idea of implementing the rebuilding, going for it all, etc. for the AI teams. If nothing else it would add direction and interest to the game. I rather like being able to strip other teams of their prospects, but than I just plain hate losing games. I have so many prospects now that I have to trade them away. Ultimately, I end up developing them, keeping the ones I want, and then trading the extras away for another round of youngsters.

[suurimonster]

[i]
SP Eric Good (22 yrs, 0/2/4/5, 3 ½ blue stars, 34th round pick, 1009th overall)
Traded to Montreal for
MR Blaine Neal (24 yrs, 6/7/8/5, 4 ½ blue stars, 15th round pick, 424th overall)
[/B]

If you forget about the names, then this is the only trade that looks crazy...at first, but maybe they really needed an SP and there are usually plenty of quality MRs around.

The real problem this highlights is the computer will always be much less active in the trading market than a human. The computer will never make 8 trades in a row, as it is simulating real-life baseball. You will make it, as you are playing 'screw the AI'.

I think the best option to help with this, as others have suggested, is to punish GMs who are trying to abuse the system e.g. players refusing to deal with you after they see that you are just trading people away after 1 day in the team.

[marc420]

In reading this thread, I've thought of one thing I haven't seen mentioned yet. The one thing I might think about changing in this system is the part where a team immediately agrees to a deal when they get a good offer.

To me, this sounds a little fishy in real life, and it could provide a way of balancing out the AI trade system.

Instead of agreeing immediately to the trade, teams would instead signal that they think this is a decent idea for a trade, but not agree to it on that day. Then team would offer the players they are offering to your team to the other AI teams, just to see what they could get.

Picture this. If a real-life GM is trying to trade a veteran from a non-contending team just before the trade deadline, he isn't just talking to one team. At some point he's probably talked to every team. Then its probably developed into a short list of a few teams that are interested. Then he's probably talking to each team, trying to see what each will offer. At some point he may have several trades arranged, but may try to get each team to add one more player just to see if they will. Then as the trade deadline approaches, he'll pick what he thinks is the best offer he's seen and make that trade.

So I'm thinking the AI teams could do the same. Trading should be a process that takes place over several game days. Now if you make an offer that is just incredibly wonderful and the AI thinks its a steal, they may immediately accept. Otherwise, they'll negotiate with you to work out a tentative agreement. But then they'll shop the players in that tentative agreement around the league.

So there's an above example where an AI team had offered say Bobby Abreu for a 1 star minor league catcher. So what I'm thinking is once that tentative deal is agreed to, the AI team takes the time to start calling other AI GM's. Basically saying, "Hey, we are thinking about trading Abreu. Would you be interested, and who might you offer?" Then when the AI gets the responses from the other teams, the AI would compare what they'd offered to the 1star minor league catcher you'd offered. If that was still the best deal, then they might be ready to conclude the deal at that point. If the other AI teams were offereing more than you were for Abreu, then the AI GM you are dealing with would come back on a following day and tell you that you need to increase your offer, or they'll trade Abreu to another team.

This way, the various AI teams would act as a check on each other against stupid trades. And the game would resemble more of the marketplace it should. If a team is willing to trade a player, what they are willing to accept for that player would depend more on the value the various GM's in the league put on the player, not just on one AI's calculation.

Actually, where this really struck me was in the Free Agent market in OOPT4. My jaw almost hit the floor the first time I was negotiating with a free agent, and when I hit a price he liked he immediately signed with me. Any real-life agent that did that should be fired for incompetence. The agent's response should have been something like "That's a very interesting offer. We'll get back to you." Then the agent would contact other GM's to see if any wanted to top my offer. If not, the agent might call me back and agree to sign. If someone topped my offer,then the agent would not immediately sign with that team either, but would call me back and see if I wanted to raise my offer.

Both an AI agent for a Free Agent player, and an AI GM considering making a trade should be very happy to get a bidding war going to raise the price of a deal.

And if this got really good, both would be willing to bluff at times.

In programming terms, the AI GM would set a floor value for a player. Below this level, the AI GM would not consider trading a player, as the AI GM feels that keeping the player on the team is a better option than any deal offered below this level.

The AI GM would also set an upper level. Any trade offer above this level would indeed be immediately accepted, as the GM who offered a deal this good appears to be a complete idiot and the AI decides to seal the deal before he changes his mind. (Note: at this point, the other GM, if its an AI should say something like "this is interesting, we'll get back to you." Then shop the deal around).

Between those two levels, the final value of make-up of the trade would depend on what the market allows. That would happen over several game days as teams make various trade offers and counter offers to each other. Only when a team feels like it has the best offers on the table that it is going to get should the AI GM decide which offer to take.

That sort of system should provide a much more realistic feel than a single AI trying to decide if a trade is good or not. What the single AI should be doing is comparing different offers for a player (or players) and deciding which is best for its team.

[Crapshoot]

marc420,
that's a pretty good and thorough idea you've laid out there.
I agree with you for the most part, but Im not sure its always accurate that it is done this way...

[Night2000]

I just got a chance to install this patch and I gotta say Markus
did one hell of a patch. Yeah sure the trade Ai needs some tweaks. However Markus has built a nice foundation with the much improved trading interface! At least I think so.
One thing maybe we should try to keep in perspective is, yeah sure the AI may still undervalue prospects as opposed to vets.
However just because the AI is willing to part with a 4 blue star
prospect for a 2 gold star vet doesn't automatically make it a bad or one sided trade. Seems as soon as the prospects turn into vets their ratings drop off and they start out with a gold star or two. Sometimes they will develop to full potential or just become mediocre. As in all sports the vet is already proven, while the prospect shows promise. So I'm sure Markus is on the right track
with the new trading AI, and it just needs to be tweaked.
Many of you have come up with some great ideas to tweak the trading. However I will still/always disagree that the trading should be hard just to make it challenging. It should just try to be realistic as possible. Sometimes you make a good/trade sometimes the AI makes a good trade. Not no one sided nonsense. There has to be a balance, which I believe Markus is almost achieving with this latest fix.