[SittingDuck]

Excellent explanation. Thanks a lot for that. Very clear.

[OldFatGuy]

OK, now at least I do understand why changing league totals is "counter-intuitive", or I guess I should say now I see exactly why lowering the HR amount in league totals actually increases Home Runs.

But I'm still at a complete loss over the big picture.

I mean, what I'm seeing is there must be, I dunno, 100 different forumulas for the 100 different possible outcomes? I mean, I admit, I don't know how many possible outcomes there are, but there must be many.

So, there's a formula similar to what you just decscribed to give the "odds" of a home run for Manny, then there's another formula that gives the "odds" of a strikeout for Manny, then one for a single to right, a single to left, a groundball to short. Is this the way its working?

Ok, if so, then you're left with about a hundred different forumalas giving a hundred different "odds" on the 100 different possible outcomes of each at bat.

Then what?

A new "Superformula" adds up all of these "odds"; and then compares them to what?

I.E. let's say Manny's final HR/AB # vs the Unit is something like 0.0532.

Ok, then a similar foruma comes up with the odds of a double? a Triple? A strikeout?

OK, so now I'm left with lots of different odds of a particular outcome per AB of Manny vs Randy.

That still doesn't answer whether A particular outcome is derived totally random based off of those odds, or whether something else is calculated at some point in the process too. Some sort of mathematical "counter-weight" that tries to ensure that the season's final count of outcomes (let's say Home Runs) is close to what the "expected" number of home runs should be.

Or, is it entirely random based on the ratings, and the results of all of those formulas that the ratings of the batter and pitcher produce?

I mean, I've always expected it was this way. But when I see what I've seen now on more than one occasion, it caused me to post. Then I read responses to my posts indicating others are seeing it. Don't you understand how it then becomes inevitable to then wonder? I'm as aware of anyone of the fallacy of judging things on too small of a sample size. But when you see the size begin to increase, well you begin to wonder.

Imagine just ONE case IRL of a manager bunting the 8 hole hitter to get to a weak hitting pitcher and having him succeed. I think we could at least agree we could imagine it, though we'd certainly question the viability of the manager in question remaining a manager.

But if you saw it three times wouldn't you at least wonder? And then after I saw it at least three times and posted, others responded saying they had seen it. Doesn't the repetition of these 100-1 long shots at least stir your curiosity just a bit as to whether there might be something else going on?

[RonCo]

Since I have no insight into his algorithms, I'll just say that it's possible that Markus has written a very sophisticated algorithm to force a team to make a bad decision specifically in order to keep scoring to some limit.

But I think it more likely--far, far more likely--that his AI algorithm just isn't robust enough to take into account a two-three move chain of events. I'm purely guessing that in this case it makes a decision to bunt because Markus thinks that's a decent strategy, then after the bunt is resolved his AI makes a decision on whether to keep the SP in the game.

The game AI is a continual source of angst, and I think it always will be. This is because it's really, really hard (IMHO) to write an AI that will make the right calls. The "right call" is difficult to define, better yet make a piece of software that determines it.

If I see three people posting that they see a single bad decision, it makes me wonder. On the ohter hand, OOTP owners play thousands and thousands of games in a week. Real baseball teams play only a hundred or so. Three occurrences in 5,000 games is hard to get too excited about. I note that no one gets into the forum and says "Wow! My #8 hitter bunted, then the AI pinch-hit for may pitcher! What a great call by the AI!"

Given the sheer number of OOTP games in a day, the repetition of a 100-1 shot had better happen quite a number of times or else the game's engine is not truly random.

At the end of the day, though, I agree with the idea that the AI has many bad logic flaws. I just don't see any evidence that makes me think it's because Markus is trying to shoehorn in results to match some pre-defined outcome.

I do, however, see compelling evidence that his algorithms are based on setting random probablilty and using the computer's ever-present random number generator.

Please note that I'm not intending to rant on Markus. I know he trulywants his AI to be great. But I've been on record before as saying that it always will have these flaws because Markus is not a 100-person psychological study group. It is my opinion that he can't possibly be expected to write a piece of code that will think like Davey Johnson or Earl Weaver or Sparky Anderson or, for that matter, like me--baseball is a game far more complex than chess, and it took major think-tanks decades to write chess programs that could compete with humans.

[RonCo]

Originally Posted by OldFatGuy

That still doesn't answer whether A particular outcome is derived totally random based off of those odds, or whether something else is calculated at some point in the process too. Some sort of mathematical "counter-weight" that tries to ensure that the season's final count of outcomes (let's say Home Runs) is close to what the "expected" number of home runs should be.

Or, is it entirely random based on the ratings, and the results of all of those formulas that the ratings of the batter and pitcher produce?

And, just to be clear...

Big changes to League Totals cause big changes to output in ways that are predictable by the Log5 equations. This is pretty solid evidence that the game has not clamped output to any specific level. There does not appear to be any counter-weight to ensure numbers fit into an overall "expected" number. If there is, it is completely ineffective.

[RonCo]

Similarly, if you keep league totals the same and change all the player's ratings up or down, the game output also changes in ways that are predictable with the Log5 equations.

Or if you keep the league totals and player ratings the same, but change all the ballpark factors, you will see predictable changes in league-wide outcomes.

All of these also suggest there is no clamp or counter-weight.

[SittingDuck]

Pretty good series of posts, RonCo.

[OldFatGuy]

Originally Posted by RonCo

And, just to be clear...

Big changes to League Totals cause big changes to output in ways that are predictable by the Log5 equations. This is pretty solid evidence that the game has not clamped output to any specific level. There does not appear to be any counter-weight to ensure numbers fit into an overall "expected" number. If there is, it is completely ineffective.

Yeah, see that was my whole point from the beginning. I was questioning whether there was some other "element" going into the equation other than the standard odds of a particular at-bat. When I read your lengthy, (and very, very good and informative posts about the log5 and league totals and how they interact), it helped me understand how the particular odds are calculated, but it still didn't answer my overriding question was whether or not there was another factor in final outcome.

Like I said, I always suspected not. And I understand AI is weak, always have. And like I said further if I had seen this once only, it wouldn't have caused concern. But I've seen it at least three times myself, and each and every time the AI pulls this stupid trick, the pitcher got a hit.

EVERY TIME. I'm not making that up. So, while I now see you've answered my question (no, you don't believe there's some other calculation involved once an at-bats odds are determined), I hope you'll understand why at least I was asking. If you played them all out like I do, and saw the AI make stupid moves (which it does all the time) but then in this one particular stupid move have it succeed EVERY TIME you've seen it, wouldn't you at least question it too?

And, like I said, after I posted, well read the first two or three responses. I'm not the only one. It really got me questioning.

Yes, my biggest concern for the game has always been AI. And I wish that would be worked on at the cost of everything else. But, when this was happening and it got me wondering about the possibility of this possible "internal regulator" if you will, then that became my biggest concern. If some "internal regulator" governing calculations was keeping up with league totals on the fly and "adjusting" odds to make final stats be within some pre-set range, then that's not a baseball sim I want any part of.

[satchel]

RonCo, thank you for your intelligent posts and for all your hard work.

[SittingDuck]

Originally Posted by OldFatGuy

Yes, my biggest concern for the game has always been AI. And I wish that would be worked on at the cost of everything else.

I agree, but now I realize that Markus basically views AI (IMO) as a fairly undesirable thing to code. I think many developers do. He'll push it a little, but I think he finds it more beneficial to code in various features than devote much time to the AI. I am against that - a significant number of owners seem to feel the same way. But it is down to what Markus does, and so I work at times (harder than I'd like) to just accept the current state of the game.

The interesting thing is - and again it's my opinion - if someone developed a baseball sim that had an outstanding AI with only a portion of what OOTP presents otherwise (or any other baseball sim for that matter), then I think a HUGE proportion of baseball text simmers would swing themselves right into that game, at the detriment of OOTP and other baseball sims. That's how important many feel a top-notch AI is. Baseball is a thinking man's game. How can a great AI not top the list as most important?

[Cryomaniac]

Originally Posted by SittingDuck

I agree, but now I realize that Markus basically views AI (IMO) as a fairly undesirable thing to code. I think many developers do.

That's because AI is an ass to code.

[SittingDuck]

Definitely. Which is why wargames where a designer has put a heavy emphasis on creating a good AI get such rave reviews.

[OldFatGuy]

Originally Posted by SittingDuck

The interesting thing is - and again it's my opinion - if someone developed a baseball sim that had an outstanding AI with only a portion of what OOTP presents otherwise (or any other baseball sim for that matter), then I think a HUGE proportion of baseball text simmers would swing themselves right into that game, at the detriment of OOTP and other baseball sims. That's how important many feel a top-notch AI is. Baseball is a thinking man's game. How can a great AI not top the list as most important?

I know I'd be one. As much as I like the setup, options, looks, stats, customability, etc. of this fantastic game, if I find one with a great AI, I'm there. No doubts. And will gladly fork over well more than the industry standard $20-$50 range for it too. Because I've been looking for it my whole adult life. Not much time left either, so I'm thinking this one might be the best I'm gonna find.

It's AI is bad, and I've said it before. But it's about the best I've seen. And I've tried many, many, many other games.

Would still love to find someone put forth the effort though. I was wondering one time if it would be feasible if some really smart baseball guys got together with some really smart writers of algorithms, and the just wrote the basic algorithms to good, solid AI. Two sets of AI. On field, in-game management AI, and then organization wide, roster building, financially motivated planning AI.

And they sold the algorithms to those who wanted to make and sell the actual sims. In other words, if they created these wonderful algorithms and copyrighted them and sold them under the name "RealAI" or something, then individual sellers of the product could incorporate this AI into their particular game, whether it be arcade style or long term sim style, and market their particular baseball sim as "Includes RealAI."

Then, maybe the cost of an endeavor might more likely provide returns as its able to make back money on more than one line of baseball games.

Ahh, dreamt it a long time, and I'm sure its not gonna happen.

[SittingDuck]

Even more than wargamers, I think the market for BB text sims is mighty small. So I think you're correct there.

[satchel]

Originally Posted by RonCo

The "right call" is difficult to define, better yet make a piece of software that determines it. ...

Given the sheer number of OOTP games in a day, the repitition of a 100-1 shot had better happen quite a number of times or else the game's engine is not truly random. ...

baseball is a game far more complex than chess, and it took major think-tanks decades to write chess programs that could compete with humans.

Three great points in a thread full of them. My hat is off to you, RonCo, for your insight.

[conception]

Ronco, you astound me with your scientific analysis of many aspects of the game. I've learned enough not to mess with the league totals much at all in my league because its darn confusing!

[StyxNCa]

Originally Posted by Cryomaniac

That's because AI is an ass to code.

That's his job and he is getting paid to do it. Jobs aren't always easy.

[StyxNCa]

Originally Posted by OldFatGuy

......If you played them all out like I do, and saw the AI make stupid moves (which it does all the time) but then in this one particular stupid move have it succeed EVERY TIME you've seen it, wouldn't you at least question it too?

Maybe the reason there isn't any serious attempt to work on the AI is because, from what I have seen in the years I have been reading this forum, most people DON'T play out the games. Every time they come out with a new version these people just sim through 100 years or so just to look at the resulting stats. So, this is why Markus has concentrated a lot more on having stats that look good rather than dealing with the AI.

[oldtime]

As someone who travels a good deal, and gets to listen to sports talk radio all over the country, I would say that OOTP game owners opinion of the Game AI and fans opinion of their team's GM and Manager HI (human intelligence) are about equal. So maybe Markus has nailed it.

[SittingDuck]

I believe Markus also mainly sims his leagues. I could be wrong, but remember reading something to this affect from him in a post. In reality, he needs to play out all his games to really get a feel for what the AI is doing.

That's like developing a car and not actually putting it through an obstacle course yourself.

[deadringer]

I've often said and I often believe that OOTP does a good job of getting the big picture right at the cost of all the finer details.

When you look at the numbers for an entire league over the course of a season you see why OOTP is the best baseball sim on the market. The closer you look at AI decisions and the way player's careers play out the less impressive it is.

The AI's in game decisions are one of the areas I see that need the most work, but then again I play out each game.