An experiment on CPU offensive AI and sliders.....

[swaldo]

Re: An experiment on CPU offensive AI and sliders.....

Just played an interesting game and is a classic example why some people use the term "rubber band AI."

Prior to this game I played against the Dodgers which was my first test game. In order to keep my offense terrible I mainly bunted and didn't get any hits from those bunt attempts. In the 7th inning I started to bat normally just so I didn't have to sit and watch a Dodger no-hit celebration. I finally slipped one through the infield for my token hit so I went back to flailing away at bunt attempts.

So before my second game I browsed the teams and found the Oakland A's are rated the worst offensively so I chose them as my next test subject. It's Angels (me) at Oakland (CPU) with maxed offensive legend sliders. I pitch with Pineiro again because overall he is rated average: good control, with slightly less than average stuff.

This time I bat normally until I got my token hit from Matsui in the 2nd inning. Now I go back to bunting so I can quickly get through innings without any offense. In the 2nd inning the A's jumped out to a 1-0 lead. Then, over the next three innings they can only muster one hit. In the 6th inning I couldn't believe what happened...

Napoli starts out the inning with a well placed bunt single (his drag bunt rating is 3 with 29 speed.) I then purposely strike out with Wood by flailing a bunt attempt at a pitch outside. The A's pitcher then threw a wild pitch so Napoli took second base (up to this point I only had one hit along with a bunch of bunt strike-outs so the A's pitcher had max command and confidence so ??) Anyway, in steps Aybar who dribbles a perfectly placed ball down the third base line for another bunt single (he is rated 39 in drag bunts.) So now it's first and third with one out and I really don't want to score because it defeats the purpose of the test. In order to NOT score I suppose I could just swing early on every pitch and strike out. But no, at this point it's too tempting and I think "Heck, maybe I'll get a trophy for bunting over a run or something." In steps Abreu (rated 18 in bunt ability) who hits a nice little sac bunt to the side of the pitcher who barely has time to throw to first. Napoli scores and it's now a tied game 1-1.

In the two games I've played the only bunt hits I got were in the inning noted above and it was the only wild pitch thrown by the A's pitcher. I'm batting with legend difficulty, default sliders. Also, two of the 3 bunts that were laid down did not even touch the bat!

AYBAR BUNTS A SINGLE DOWN THE THIRD BASE LINE:
Aybar-Bunt.jpg

ABREU SAC BUNTS BETWEEN PITCHER AND 1ST BASEMAN:
Abreu-Bunt.jpg

As 1.jpg

So now it's bottom of the 6th inning and all of a sudden I can't locate anything in the strike zone so I walk Cust. I'm getting a bad feeling and sure enough Chavez homers to right. Dang, this is too much for my bunt offense to overcome.

Later, in the 7th inning Pineiro's energy was spent, but I left him in because I wanted to see if OAK would hit him. In two innings he gave up a walk, bunt single, single and finally a homer to Cust.

Just looking at stats only when I was losing in the game, I gave up 1 run (the Cust homer with Pineiro at zero energy) on 4 hits in 6 1/3 innings.

Thus far over two games Pineiro has a 3.85 ERA

As 2.jpg

[Dreifort]

Re: An experiment on CPU offensive AI and sliders.....

Just played an interesting game and is a classic example why some people use the term "rubber band AI."

Prior to this game I played against the Dodgers which was my first test game. In order to keep my offense terrible I mainly bunted and didn't get any hits from those bunt attempts. In the 7th inning I started to bat normally just so I didn't have to sit and watch a Dodger no-hit celebration. I finally slipped one through the infield for my token hit so I went back to flailing away at bunt attempts.

So before my second game I browsed the teams and found the Oakland A's are rated the worst offensively so I chose them as my next test subject. It's Angels (me) at Oakland (CPU) with maxed offensive legend sliders. I pitch with Pineiro again because overall he is rated average: good control, with slightly less than average stuff.

This time I bat normally until I got my token hit from Matsui in the 2nd inning. Now I go back to bunting so I can quickly get through innings without any offense. In the 2nd inning the A's jumped out to a 1-0 lead. Then, over the next three innings they can only muster one hit. In the 6th inning I couldn't believe what happened...

Napoli starts out the inning with a well placed bunt single (his drag bunt rating is 3 with 29 speed.) I then purposely strike out with Wood by flailing a bunt attempt at a pitch outside. The A's pitcher then threw a wild pitch so Napoli took second base (up to this point I only had one hit along with a bunch of bunt strike-outs so the A's pitcher had max command and confidence so ??) Anyway, in steps Aybar who dribbles a perfectly placed ball down the third base line for another bunt single (he is rated 39 in drag bunts.) So now it's first and third with one out and I really don't want to score because it defeats the purpose of the test. In order to NOT score I suppose I could just swing early on every pitch and strike out. But no, at this point it's too tempting and I think "Heck, maybe I'll get a trophy for bunting over a run or something." In steps Abreu (rated 18 in bunt ability) who hits a nice little sac bunt to the side of the pitcher who barely has time to throw to first. Napoli scores and it's now a tied game 1-1.

In the two games I've played the only bunt hits I got were in the inning noted above and it was the only wild pitch thrown by the A's pitcher. I'm batting with legend difficulty, default sliders. Also, two of the 3 bunts that were laid down did not even touch the bat!

AYBAR BUNTS A SINGLE DOWN THE THIRD BASE LINE:
[ATTACH]23705[/ATTACH]

ABREU SAC BUNTS BETWEEN PITCHER AND 1ST BASEMAN:
[ATTACH]23704[/ATTACH]

[ATTACH]23707[/ATTACH]

So now it's bottom of the 6th inning and all of a sudden I can't locate anything in the strike zone so I walk Cust. I'm getting a bad feeling and sure enough Chavez homers to right. Dang, this is too much for my bunt offense to overcome.

Later, in the 7th inning Pineiro's energy was spent, but I left him in because I wanted to see if OAK would hit him. In two innings he gave up a walk, bunt single, single and finally a homer to Cust.

Just looking at stats only when I was losing in the game, I gave up 1 run (the Cust homer with Pineiro at zero energy) on 4 hits in 6 1/3 innings.

Thus far over two games Pineiro has a 3.85 ERA

[ATTACH]23708[/ATTACH]

Call it what you will (some call it the c-word)... but this is a classic example of the computer "generating" the outcome it feels needs to happen.

In early stages of baseball sim games, if a career .350 hitter is batting .210 in July, the computer says..."wait! this guy needs to hit .350, he can't hit .210...that doesn't compute!" so the hitter goes on to hit .440 for the remaining 2 months.

now the guy didn't hit .350 every yr, but the computer had a goal that the hitter had a range (.320-.360) that the hitter had to reach. So if for some reason the hitter was below that goal, the computer has to play catch up.

baseball sims have gotten better to allow slumps, streaks and overall variables to make a more realistic outcome with stats.

BUT, it's still a computer and will ALWAYS be a computer. So 2+2 HAS TO EQUAL 4 no matter what. So if the computer comes up with the answer 4 and you provide a 2 --- the computer HAS to generate the other 2.

Thanks to great coders, this is becoming less and less influential in games, but it still exists. Somewhere in your game, it determined that the Angles are the better team and needs to win. Since there are human variables, you can have an effect on this. But for every effect you try to influence on the outcome, the CPU/AI is performing their own influence trying to get the outcome back to their desirable outcome.

In early baseball sims, the CPU would predetermine outcomes as a team (or batter) HAS to reach a certain outcome. Now its more of a needs to reach a certain outcome .... to maintain the stats and correlation of ratings against each other.

[phillyfan23]

Re: An experiment on CPU offensive AI and sliders.....

ok weird thing has been happening.....

i tested a number of legend maxed slider games where I didn't try to hit at all, and the highest offense the cpu created was 4 runs on 7 hits.

maybe 5-6 test games.

So I tried something today and played legend hitting with pitching on HOF maxed sliders:

very first game I tried this:

I had 0 hits

cpu had 10 runs on 16 hits

Next game, i really tried to hit and I mustered 5 hits 1 run

the cpu had 9 runs on 15 hits

The last game I tried I was shutout 5-0 , I had 2 hits and the cpu 11 hits.


maybe this phenomenon exists only on legend pitching? who knows? but I'm using HOF pitching from now on.

[Hypostatic]

Re: An experiment on CPU offensive AI and sliders.....

this game clearly has predetermined events - I have obsessively played the Twins opener at Anaheim - I have played it 13 times and Nathan has blown 4 out of 5 saves !!!! ridiculous - the angels ALWAYS make a big comeback in the 7,8, and 9th inning - I have been able to predict almost every at-bat after the 6th Inning?? This is may be the greatest sports game ever, but that is not saying as much as some people think. They need to focus on the on-field happens. All these little things can be toned down a bit. I think it is best to turn down the CPU contact and power to 2 clicks each after the 7th Inning, when playing on All Star

[Phoenixmgs]

Re: An experiment on CPU offensive AI and sliders.....

maybe this phenomenon exists only on legend pitching? who knows? but I'm using HOF pitching from now on.

Have fun with that, this is how the 1st 9 games of my season went on HoF pitching:


I played 2 more games since then and I've won 3-1 and 12-3; I think the CPU is throwing me some bones after those first 9 games.

this game clearly has predetermined events - I have obsessively played the Twins opener at Anaheim - I have played it 13 times and Nathan has blown 4 out of 5 saves !!!! ridiculous - the angels ALWAYS make a big comeback in the 7,8, and 9th inning - I have been able to predict almost every at-bat after the 6th Inning?? This is may be the greatest sports game ever, but that is not saying as much as some people think. They need to focus on the on-field happens. All these little things can be toned down a bit. I think it is best to turn down the CPU contact and power to 2 clicks each after the 7th Inning, when playing on All Star

Wow, another slider enthusiast is complaining about the CPU AI. I think it's clearly apparent there is something bad in the AI code whether it's phillyfan's theory or predetermined stuff or both. I know last year certain games would be predetermined to be low or high scoring. I never have auto-save on so when I got games that really felt scripted to being low scoring or high scoring, I replayed those games sometimes 2 or 3 times and each time, the game played the same. In my divisional series against the dodgers last year, every game was a pitching duel. Then, in the NLCS, every game was a slugfest where final scores were around 16-14, and whoever had the last AB won.



I think what we all want is just the pure player attributes to determine things. It's clear to that this is not what is going on.

[Brian SCEA]

Re: An experiment on CPU offensive AI and sliders.....

in real life it's natural because human psychology plays a part. You're up 8 you tend to relax in games.

but that psychological aspect can be eliminated in a video game.

and my lack of hitting skills SHOULD not affect how the cpu hits in this game, ESPECIALLy when the sliders on legend are MAXED out.

The cpu should hit very well on those sliders regardless how I do with my at bats. Right now, my games are not like that. The worse i hit, it seems that the cpu won't hit very well. and that's not right.

i will keep testing and hope the cpu kills me 15-0 some games, especially on MAXED OUT legend sliders.

The CPU's hitting has no relation to your hitting. It does have relation to your pitching, which if anything is what could be changing. Isn't it a more likely explanation that that the better you hit, the better you are allowing the CPU to hit? Some people don't play as hard when they're winning. In fact, that's more true in a video game than real MLB baseball, not less. An experiment needs to eliminate this bias.

What's more important is the control data. Every person running this experiment is going to consistently get a different result because of their skill level and the requirement to subjectively do "poorly". Whether someone gives up 2, 4, or 6 runs on Legend is meaningless unless we know how they would have done normally under the same conditions. I would expect some people to consistently average 2, and others to consistently average 6. There is no contradiction.

That said, in subjective experiments people get exactly the result they originally expected. Meaning people who believe there is a correlation find one, and people who don't don't find one.

[Brian SCEA]

Re: An experiment on CPU offensive AI and sliders.....

Here's some information that might be helpful as far as sliders go, since it sounds like there's some confusion here. The sliders exist to help people fine tune the game to their liking. A full five steps in one direction might make an average team into a top tier team, and vice versa.

Yet even top tier teams like the Yankees lose plenty of games, sometimes by large margins, to bottom tier teams. MLB teams even lose to AAA teams in practice games. If a game is lost by 10 runs, that really has nothing to do with the sliders at all. When the Yankees lose 10-0 to a weak team, can we automatically assume the other team cheated?

Look up how much better the Yankees are than average, factoring in their league and stadium. It's less than one run offensively, and less than one run defensively. Let's say 1 run total for simplicity. So if you're seeing huge effects, it has nothing to do with sliders. And likewise if you're expecting or assuming huge effects with sliders, you're not going to get them. That said, if you look at how a team like the Yankees does over 162 games, you'll see 1 run a game isn't as insignificant as it sounds. This is just the way baseball is.

This is why sliders are not used for difficulty so much as fine tuning, and more specifically allowing you to tune various aspects independantly. Few slider sets need them to be more drastic than they are, because over 162 games they'll add up. They exist for precision, not for show.

===NEW TOPIC===

As for experiments, statistically speaking most people get the conclusion they expect. I don't want to discourage anyone from having harmless fun, so what I'll say is a good experiment is blind and runs enough cases and uses a control experiment

Firstly, that means objective tests rather than subjective "try to do this" then "try to do that" goals. This is especially if the observer (i.e. experimenter) is the person generating all the data. You could teach a horse any language and to be psychic if you're allowed to do that: http://en.wikipedia.org/wiki/Clever_Hans. There is definitely a better way if you're interested.

Second, if you're drawing conclusions after 1-3 games, you need to ask yourself why. You can prove that Rookie is harder than Legend by playing 1-3 games! If someone insists on drawing a conclusion in a few games, they pretty much have no choice but to back up their conclusions from that point on. This is a subject of frequent exploit in sales. Needless to say, if an experiment continues after such a point has been crossed, it will get less reliable not more.

This is why, every scientist must ask themselves which is more important - their beliefs or the truth. Even if they are the same thing, which they prioritize completely changes the experiment and the data generated by it.

You also need control data of sufficient normal games.

Lastly, any experiment which consistently produces different results for different people is unfortunately fatally flawed, and will allow anyone to draw any conclusion. This is where we should revise the experiment with a fresh approach.

Enjoy!

[davehughes14]

Re: An experiment on CPU offensive AI and sliders.....

Here's some information that might be helpful as far as sliders go, since it sounds like there's some confusion here. The sliders exist to help people fine tune the game to their liking. A full five steps in one direction might make an average team into a top tier team, and vice versa.

Yet even top tier teams like the Yankees lose plenty of games, sometimes by large margins, to bottom tier teams. MLB teams even lose to AAA teams in practice games. If a game is lost by 10 runs, that really has nothing to do with the sliders at all. When the Yankees lose 10-0 to a weak team, can we automatically assume the other team cheated?

Look up how much better the Yankees are than average, factoring in their league and stadium. It's less than one run offensively, and less than one run defensively. Let's say 1 run total for simplicity. So if you're seeing huge effects, it has nothing to do with sliders. And likewise if you're expecting or assuming huge effects with sliders, you're not going to get them. That said, if you look at how a team like the Yankees does over 162 games, you'll see 1 run a game isn't as insignificant as it sounds. This is just the way baseball is.

This is why sliders are not used for difficulty so much as fine tuning, and more specifically allowing you to tune various aspects independantly. Few slider sets need them to be more drastic than they are, because over 162 games they'll add up. They exist for precision, not for show.

===NEW TOPIC===

As for experiments, statistically speaking most people get the conclusion they expect. I don't want to discourage anyone from having harmless fun, so what I'll say is a good experiment is blind and runs enough cases and uses a control experiment

Firstly, that means objective tests rather than subjective "try to do this" then "try to do that" goals. This is especially if the observer (i.e. experimenter) is the person generating all the data. You could teach a horse any language and to be psychic if you're allowed to do that: http://en.wikipedia.org/wiki/Clever_Hans. There is definitely a better way if you're interested.

Second, if you're drawing conclusions after 1-3 games, you need to ask yourself why. You can prove that Rookie is harder than Legend by playing 1-3 games! If someone insists on drawing a conclusion in a few games, they pretty much have no choice but to back up their conclusions from that point on. This is a subject of frequent exploit in sales. Needless to say, if an experiment continues after such a point has been crossed, it will get less reliable not more.

This is why, every scientist must ask themselves which is more important - their beliefs or the truth. Even if they are the same thing, which they prioritize completely changes the experiment and the data generated by it.

You also need control data of sufficient normal games.

Lastly, any experiment which consistently produces different results for different people is unfortunately fatally flawed, and will allow anyone to draw any conclusion. This is where we should revise the experiment with a fresh approach.

Enjoy!

[Phoenixmgs]

Re: An experiment on CPU offensive AI and sliders.....

If someone insists on drawing a conclusion in a few games, they pretty much have no choice but to back up their conclusions from that point on.

I have played probably around 200 or more games vs the CPU of The Show 09 and The Show 10. And, I have never seen so many well executed pitches hit hard in real life. The CPU can literally string 5+ hits in a row on well located, well executed pitches. In baseball, good pitches are hit hard from time to time; sometimes the pitcher just has to tip his cap to the batter, he threw the pitch where he wanted and the batter hit it. However, very rarely do you see big innings in real baseball when the pitcher is executing his pitches well. Big innings occasionally happen when a pitcher is pitching good, but most of those innings involve the batters getting quite a few bloopers and seeing-eye singles; pretty much the hitters getting breaks and the pitcher not getting breaks. It's never hard hit after hard hit against a pitcher executing well. The CPU hitters in this game do not hit like humans, they don't chase nearly enough. The amount of times the CPU PULLS a high and tight fastball foul is beyond ridiculous. And, really there isn't that much hitters can do even if they know the pitcher's pattern if the pitcher is throwing the ball where he wants to. If that were the case, Tom Glavine wouldn't have won 100 games, let alone 300, he did the same thing every game. He mixed it up a bit more when his velocity started to drop off, but he still went to his bread and butter most often, pounding the outside corner.

[phillyfan23]

Re: An experiment on CPU offensive AI and sliders.....

^^^

first of all....I went into this experiment with an open mind. I had my suspicions, but it wasn't final by any means. Also, these are things people have experienced from last year as well seeing how the cpu made comebacks too often.

for mlb 09,i saved a game after the 4th inning leading by a score of 4-0. using hamels against the pirates 5th starter. I replayed the game 30 times plus to see the final results.Logic and realism would indicate that the phillies would win most of those games and win a lot of those games by more than 5 or by 5 runs considering they have the more potent offense and the lead spotted. out of 30 plus games, I won by more than 4 runs ONCE. These games included 8th place hitters consistently getting xb hits and HRs from pitches off the plate and some other fishy stuff going on. anyways, that ship has sailed, and I don't see that problem this year.

the issue this year is legend difficulty level. I noticed right off the bat that the cpu would match how you were doing on offense to make things "exciting" I guess. Those were on legend sliders with sliders lower than default for the cpu. If I scored a lot of runs, i would win most of the times, but the cpu was also getting a lot of runs.

If i scored not too many runs, the cpu's bats would suddenly quiet down. So i increased the sliders to maximum hitting HOPING that the cpu would crush me regardless of how I was doing on offense. I tested against the best hitting teams and I intentionally struck out nearly every AB....the cpu would not explode and i could strike them out easily on MAX legend sliders. This should not happen.

Now I went back to HOF maxed sliders, and I noticed a difference right away. 5 games I've played on MAX HOF and the cpu has killed me regularly like I expected. The offense for the cpu did not care if i was hitting poorly or the oppposite. They would perform as their sliders would indicate, which leads me to believe that there is a mechanism going on in lgnd difficulty to allow such behavior.

You can say well we developped the code and so forth, but no dev team is perfect and an error could possibly be made. It just doesn't make any sense that on max legend I gave up a MAXIMUM of 4 runs and 7 hits with these experiments using the same techniques and concentration level that I used for HOF maxed sliders that I got killed on, absolutely killed.

I understand that the sample size isn't huge, but with similar experiennces from different users, i think something is going on in the legend difficulty. I could be wrong, but I don't wanna play on that level knowing that I also could be right.

anyways, this is no longer an issue for me on HOF level maxed sliders. ( I was just demolished by the arizona cardinals 12-2 after 6 innings using roy halladay). this after 5 or so games on maxed HOF sliders which I was demolished in 4 times.

On max legend sliders when I hit poorly I consistently threw 2 run 5 hit performances with Josh towers against the phillies offense.

anyways, thanks for the comments, and I highly appreciate your interaction with us.

[Brian SCEA]

Re: An experiment on CPU offensive AI and sliders.....

I have played probably around 200 or more games vs the CPU of The Show 09 and The Show 10. And, I have never seen so many well executed pitches hit hard in real life. The CPU can literally string 5+ hits in a row on well located, well executed pitches. In baseball, good pitches are hit hard from time to time; sometimes the pitcher just has to tip his cap to the batter, he threw the pitch where he wanted and the batter hit it. However, very rarely do you see big innings in real baseball when the pitcher is executing his pitches well. Big innings occasionally happen when a pitcher is pitching good, but most of those innings involve the batters getting quite a few bloopers and seeing-eye singles; pretty much the hitters getting breaks and the pitcher not getting breaks. It's never hard hit after hard hit against a pitcher executing well. The CPU hitters in this game do not hit like humans, they don't chase nearly enough. The amount of times the CPU pulls a high and tight fastball foul is beyond ridiculous. And, really there isn't that much hitters can do even if they know the pitcher's pattern if the pitcher is throwing the ball where he wants to. If that were the case, Tom Glavine wouldn't have won 100 games, let alone 300, he did the same thing every game. He mixed it up a bit more when his velocity started to drop off, but he still went to his bread and butter most often, pounding the outside corner.

So to summarize, you see the CPU is hitting too well against good pitches and not chasing enough pitches? That's good feedback, and I think in the latter case you can verify it by using the Pitcher-Batter Analysis. For example, are you getting less than 1 out of 5 balls out of the SZ chased?

I was only commenting on how the CPU performs relative to how the human is doing in a game, and how the CPU's performance doesn't change based on who's winning.

[Phoenixmgs]

Re: An experiment on CPU offensive AI and sliders.....

So to summarize, you see the CPU is hitting too well against good pitches and not chasing enough pitches? That's good feedback, and I think in the latter case you can verify it by using the Pitcher-Batter Analysis. For example, are you getting less than 1 out of 5 balls out of the SZ chased?

I was only commenting on how the CPU performs relative to how the human is doing in a game, and how the CPU's performance doesn't change based on who's winning.

I understand you were commenting mainly on the experiment and thanks for taking time to read and comment on my post. I think the CPU batter AI is perhaps the biggest issue with the game. I think some of the batting AI problem is due to meter pitching, it is far too accurate. Therefore, I guess, you have to program the CPU to hit good pitches better than they should be able to since even poor control pitchers can be pretty accurate with the meter, especially with fastballs. The CPU doesn't swing and miss enough against pitchers that just have great stuff. If you throw a great curveball or slider and it gets a good chuck of the plate, the CPU is very likely to get a good rip at it. It is almost impossible to pitch well with Carlos Marmol using classic pitching, the CPU tears him apart because his location is obviously not good, but he has great stuff and shouldn't have to locate well. Play on Legend or HoF and try pitching with Marmol using classic pitching. Marmol gives up runs in real life not because he is hit but because he is wild. If Marmol throws strikes, batters just aren't going to hit him. In this game, you don't feel dominant pitching with a strikeout pitcher because the CPU hitting is too location based. So pitching with someone who has overpowering stuff just doesn't feel right in the game because you really can't overpower the CPU.

[Phoenixmgs]

Re: An experiment on CPU offensive AI and sliders.....

Why was this topic moved? This thread isn't even about sliders. Phillyfan just said to max out the CPU sliders to give the CPU every once of an advantage possible. The thread is about batting AI on Legend.

[phillyfan23]

Re: An experiment on CPU offensive AI and sliders.....

Why was this topic moved? This thread isn't even about sliders. Phillyfan just said to max out the CPU sliders to give the CPU every once of an advantage possible. The thread is about batting AI on Legend.

Exactly, this thread has little to do with sliders and a whole lot of A.I. Could a mod move this back to the main page?

Thanks

[NAFBUC]

Re: An experiment on CPU offensive AI and sliders.....

The title has sliders and most of the posts mention sliders.