The AI and Trading

[endgame]

A better variety of responses would be a worthy objective of change.

[Déjà Bru]

Originally Posted by endgame

A better variety of responses would be a worthy objective of change.

Heh, the position of this post at the top of a new page was most unfortunate, eg. I thought you were complaining about the quality of respondents in this thread!

You were, I believe, replying in support of this post immediately preceding it but at the bottom of the previous (standard length) page:

Originally Posted by babs42

Maybe a possible solution to the problem would be a diversity of answers from the AI. Instead of "I'd never accept!", perhaps something like, "This trade is fair on paper, but we don't want to give up this player unless we're overwhelmed" would be better. Or different outlooks based on how the season is going beyond what we already have, like "we want to win but we aren't willing to give up top prospects for a rental." If this is possible it would create a more realistic trading experience and lead to better communication between AI and player.

[dkgo]

Originally Posted by steady teddy

Look at the trade that just occurred between the Cubs & the White Sox. I know nothing about the prospects the White Sox received in the trade and time will only tell how they pan out because after all, prospects are nothing more than prospects. But that's not my point.

Chances are if you had control of the Cubs (or ANY team for that matter) in OOTP 18 and tried to acquire a pitcher of Quintana's caliber, the AI would not only ask for 2-3 quality prospects, but also the best player on your team. It would be fine as you say if the Trade AI merely wanted players you don't want to let go. But in 99% of the trades, the AI wants the Sun , the Stars, the Moon, and everything in between.

I happened to trade for quintana pre-season as the brewers on the challenge mode settings. I sent

Brett Phillips (4*, #75 prospect)
Jake Gatewood (3.5* decent prospect)
Junior Guerra (useless ML SP)
Mario Feliciano (1*, pretty useless)

[Greenbay]

The computer can not replicate the human mind. It has no emotion, no gut feeling, no prejudices. It simply analyzes probabilities and gives us results. We want (expect) the game to mirror real life but it can't. Trading in OOTP can be very tuff (and frustrating). It should be. But we can't expect it to mirror MLB trades.
I do a lot of trading. I use the "make this work now" button a lot. Sometimes they ask for another player that I am willing to give up. Other times (probably most times) they want too much, so I walk away. There's always another team to try to work a deal with.

[NoOne]

its not about mimicing any individual from RL... it's trying to look at motivations and reasons for making a decision and making it quantifiable for a program to use.

somethings will be lost in translation, but for the most part it operates on the basic motivations from real life... just as in real life, some decisions are terrible, lol.

positional depth and other qunatifiable weaknesses most definitely influence what you can get in return... you can trace a cause/reason for everything the ai does, and that's a good thing.

just as expecting too much for an RP, i think most need a re-evaluation of how they perceive the MLB trading environment -- especially if their only reasoning is - "i watch alot of baseball." or that they've been around for a long time... because most, if not all peole, are not linear-regression machines... they have a hard time remembering 2 weeks ago, let alone figuring out what 'normal' is.

trades in real life are difficult to get done... sometimes months or even longer, lol. mutual motivation and complimentary needs isn't that common and very difficult to find.

[steady teddy]

I regret even mentioning the Quintana trade. It deflected from the OT. Adding to the variety aspect, something I think we'd like to see is something that we'll never see and that's a bit more fairness in the trade AI. But it won't happen for the reasons already mentioned: It's a computer simulation.

Again, that's why I usually play with fictional rosters so when make I bad trades I don't worry how realistic the trade value really is.

[jlech1805]

I assume that the trade difficulty essentially changes the surplus value requirements for the AI to make a trade. I'd like to keep this feature, but wish OOTP added another.

As soon as you reach terms of a potential trade the AI would then shop each player in the package around the league. If they can get a better deal(s) elsewhere they'd request for you to increase your offer.

Assuming each AI team has slightly different evaluations for each player and different values / needs for that player (win curve, positional depth) it could add some realism to the trade market.

What would be nice about this type of logic is the AI could do it for all transactions. Instead of waiving or DFA'ing a player instantly the AI should first shop around that player.

[Cobby]

A couple thoughts about trading...

So, after building up my team to a dynasty, I turned over control to the AI and watched with some fascination as the AI destroyed the dynasty.

Oddly, although the trades the AI made were not what I would have done, none of them seemed to me to be completely terrible individually. But there was no sense of strategy behind them. The biggest mistake the AI makes, I think, are in trades *not* made. Basically, the AI held on to aging players too long instead of trading them for prospects. So eventually, the prospect pool gets depleted and the former dynasty sinks into mediocrity.

The AI doesn't seem to have a sense for anticipating even just a few years into the future. One of the first trades the AI made after I turned over control was to trade a nearly-ready pitching prospect for an older shortstop. It wasn't a completely terrible trade - we did have a weakness at shortstop and our rotation was strong. But we also had three 30+ year old pitchers in the rotation. That young pitcher wasn't needed at the moment but was going to be vital in a few years. There weren't many young pitchers available either, and I had worked hard to get him.

Which brings me to my second point. Nearly every trade I make as a human GM, my assistant GM tells me I'm getting a bad deal and I should ask for more. Yet I end up building a dynasty while still seeming to losing every trade on points according to the AI evaluation.

Now, possibly the difficulty wasn't with the AI in general but with that particular GM. All of the GMs have ratings for how they evaluate players - like how much they favor prospects vs veterans, or how much the appreciate player loyalty, or offense vs. defense or hitting vs pitching, etc. But, even so - I think that ratings like that ought not to override common sense.

As an aside - one thing that might be cool to implement would be that when an assistant GM is working with a legendary human gm, that any time the human gm makes a trade that the assistant disagrees with, then some of these GM parameters get tweaked in such a way to make the assistant gm more likely to approve of the trade. So, there could be a sort of mentoring relationship and then when the assistant gm quits and goes to another team, he's evaluating trades now more like the human would have.

One could apply this gm parameter tweaking also to winning vs losing with computer teams too. GMs whose teams are deteriorating could start changing their parameters until their teams start improving or something like that... Or even when an AI gm trades with another more highly rated AI gm. If both are agreeing with the trade then the maybe the less successful gm is being taken advantage of and could tweak his parameters to make him less likely to accept futer trades like that...

Anyway...

[NoOne]

Originally Posted by endgame

<snip by noone>
Enough for now. Ramble any more and you'll mistake me for NoOne. Well, the overuse of commas, added punctuation, and the complete absence of lol's should give me away, right?

umm... where are the type-o's? i have standards, c'mon!

sorry, i need to kill some time, just having fun below:

i may use extra commas, but likely not as bad as you you probably think. most people don't know what a direct address is, for example, or an interjetion. :P plus, if it's phrases, i put them in there to seperate thoughts. not gramatically correct.... but, they are phrases so who cares? (that should be all correct uses of commas, but i won't say that confidently about anythign that follows)

added punctuation? i do a "..." if i double up it's a type-o

you know what i don't do for informal writing? i don't proof-read. my spelling has gone downhill but compared to most, i have excellent grammar when it counts. both parents = teachers, 1 math 1 english.. i was screwed if i didn't get good grades from the toughest teachers available so, anything related to that i won't argue much

i can't get over the fact that it's 1 space now instead of 2 spaces after a "." ... wtf, why did that change other than arbitrary nonsense? consistency is nearest to godliness, and not cleanliness... although no one likes a stinky person unless it's part of your culture.

grammatical rules are important for consistency, but they are not "natural law" so to speak. arbitrarily choosen stuff for the sake of uniformity, which is the important part.

i took a english major weed-out class on accident... tons of writing. but, i could poop out a 2 page essay in 30mins or less when i was in college. 10-15 page research papers took a bit longer, but only a few hours if i had the research readily available. good working memory, i can organize it all efore i put anythign to paper.

spell check takes care of most of my mistakes...

[endgame]

Originally Posted by endgame

Enough for now. Ramble any more and you'll mistake me for NoOne. Well, the overuse of commas, added punctuation, and the complete absence of lol's should give me away, right?

Having a little fun, too. Do note, the bolded above was directed at my own personal proclivities. In this case, being prevalent enough to "give me away" as the writer.

The only other comment I might make about your post/response is using English major, followed by weed seems such a natural combination. You activated my nostalgia.

[Bears5122]

Originally Posted by babs42

OOTP is an excellent game, probably one of the best sports games out there. However, when it comes to trading, the AI needs serious work. Take this example I ran (Hard difficulty, which is the most realistic according to a decent amount of threads I found, Neutral preference). I am playing as the Yankees, and I am attempting to trade with the Cubs. We both have winning records, and they are a "Win Now" team. I wanted to see what I could get for Dellin Betances. Dellin has a 2.43 ERA with 42 K's in 30 innings, and a FIP of about 3. He has 2 1/2 years left before free agency. The Cubs have Wade Davis, who is pitching to a 4.55 ERA in his walk year. For one of baseball's best relievers, the Cubs should be willing to give up at least one top prospect, possibly 2. When attempting to trade with them for three of their top prospects (just one at a time, not bundled together), the AI stated we were way off in negotiations, with my assistant GM stating we were giving up way too much. Whenever I attempted to make the trade work, the AI wanted one of my best players, in addition to Betances, for one top prospect. Then I attempted to trade Dellin for Davis straight, which would be an easy win on paper for the Cubs, and yet they thought that trade would be bad for them. Clearly, the AI either wants to fleece, or lets you fleece them. I have no idea how to fix the problem, and I understand that it's much easier to point out a problem than to offer a solution, but I think the more we talk about it the better chance it can get fixed.

I don't think it's a given that you'd get 1 or 2 top prospects for Betances. He's still a middle reliever and teams giving up top prospects for them is a new thing. You're expecting an Andrew Miller type haul for a player who isn't close to Miller. What other middle relievers have gotten 1 or 2 top prospects of late?

As for trading straight up, it depends on how the AI evaluates. Davis likely has better ratings, and his stats last season and the year before are better than Betances. So unless they are valuing the current season at a huge number (75 or higher), Davis is going to be seen as the better pitcher. You're putting a lot of your value on less than half a season of work.

[PSUColonel]

I would like to know the difference between average and hard. Is it just that on hard the AI demands more, or does it in fact take "need" into consideration as well?


EDIT: the reason I ask, is because I haven't seen any real difference in terms of what the AI will accept when I hit "make this work now".

[Markus Heinsohn]

Trade AI is probably the most difficult thing to develop, and even if it is "perfect", many users will still complain because they attach a certain value to certain player names they know. The AI does not care about names, all it does is care about ratings & stats and basic stuff like age and contracts.

When the user offers a trade to the AI, it evaluates its own team both before and after the trade in several ways (for example overall current value, or quality of farm system, future budget etc) and then weighs these evaluations based on the team mode (e.g. win now) and the GMs personality. And in the end you have two values as a result of some incredibly complex calculations (which are probably not 100% bug-free because of their complexity), and if X (value after trade) is bigger than Y (value before trade) the AI will accept the trade. If not, it will decline.

We tweak these evaluations every year, and they get better every year. But keep in mind, it will always be a comprimise, because each 1% of additional "smartness" costs a) a ton of simulation speed and b) a ton of development time.

[CBL-Commish]

I know I'm dredging up an old thread, but there are a million trade threads and didn't want to make a new one. I'm not really complaining about the trading difficulty or AI, I'm just a little bemused by it. I understand that trading has long been a complaint, literally since OOTP began, and it's deviously hard to program something that mimics the back-and-forth of humans negotiating.

But here's my situation. Offseason. My team is pretty strong. I have settings for difficulty average, favor prospects, ratings set for 30%, current year 50%, previous 15%, two years ago 5%.

Every other day of the offseason I get an offer like this:

The other team wants to send me a starting pitcher coming off an 11-15 season with a 5.40 ERA. In return they want my starting third baseman, a prospect with 4-star potential, and my setup man who has had four consecutive years with 60+ innings and an ERA under 2.50. And they want me to pay most of the salaries of my guys going over.

I can't remember the last trade I accepted. I did make one recently based on an AI proposal, but only after it was modified very substantially and I took on a bunch of salary.

If I didn't know any better I'd assume the AI was constantly proposing absurd trades in the hopes that I'd accidentally hit the accept button. Like a phishing attempt.

I almost want to change the trade setting to Hard or Very Hard just to see what the outer limits of absurd are. Average already has me shipping three times the star value out, and paying salary.

[Calvert98]

Originally Posted by CBL-Commish

I know I'm dredging up an old thread, but there are a million trade threads and didn't want to make a new one. I'm not really complaining about the trading difficulty or AI, I'm just a little bemused by it. I understand that trading has long been a complaint, literally since OOTP began, and it's deviously hard to program something that mimics the back-and-forth of humans negotiating.

But here's my situation. Offseason. My team is pretty strong. I have settings for difficulty average, favor prospects, ratings set for 30%, current year 50%, previous 15%, two years ago 5%.

Every other day of the offseason I get an offer like this:

The other team wants to send me a starting pitcher coming off an 11-15 season with a 5.40 ERA. In return they want my starting third baseman, a prospect with 4-star potential, and my setup man who has had four consecutive years with 60+ innings and an ERA under 2.50. And they want me to pay most of the salaries of my guys going over.

I can't remember the last trade I accepted. I did make one recently based on an AI proposal, but only after it was modified very substantially and I took on a bunch of salary.

If I didn't know any better I'd assume the AI was constantly proposing absurd trades in the hopes that I'd accidentally hit the accept button. Like a phishing attempt.

I almost want to change the trade setting to Hard or Very Hard just to see what the outer limits of absurd are. Average already has me shipping three times the star value out, and paying salary.

Hey CBL,

I haven't really seen this that much, but I have seen it.

Here are my settings: "Out-of-the-box" Ratings, Trade Difficulty "Very Hard or Hard (can't recall), Favor Prospects and No Stars (helps with the Fog-of-War).

Yes, I've seen some goofy trade requests of course, and many of them are not that great, but some of them are fairly decent.

I have noticed an improvement, and every year OOTPD works on it from what I gather; to that, I think I heard that in OOTP 19, they have been specifically giving the Trade AI more attention.

[DelAbbot]

In Challenge mode, the AI trading is good as long as you keep draft-pick-trading disabled. You pretty much have to pay 2x of "value" to get AI to accept a trade of a good prospect. Sometimes, it's 3x "value" for a top 100 prospect. The AI is asking for an obvious overpay to accept, but at least it makes it a challenge.

[Juggernt]

Originally Posted by endgame

a better variety of responses would be a worthy objective of change.

no way!

[majesty95]

I would be in favor of a setting between normal and hard as well. I'm not sure if that is able to happen with the coding time Markus acknowledged trading takes.

I will say, playing historicals on normal, neutral, I do see some things that are strange. I've tried trading off good players at the deadline when not in contention and received no offers or they want me to add a top level player with my vet for a mid level prospect. I've also had instances where I've tried getting rid of a mid-range player and nobody was willing to offer anything.

One thing I wondered when reading Markus' response was about the calculation of whether the AI team is better after the trade and how much of a role that plays. For example, if I offer my 4* prospect for the AI's mid-level vet when they're rebuilding, why do they also ask me for another 3* plus prospect? Do they value their current team maybe too much or the prospects too little?

I think that's been the biggest thing for me. Rebuilding teams don't seem to get rid of the players they can and often seem to ask too much for them. If they're judging whether the prospects improve their current team or how much they are hurting their current team by trading the player they don't need, is that playing a role? That could also be what causes some of the imbalance. I have no idea if that's true but it crossed my mind when reading that.