Stats-Based CPU Sliders [MLB 14 Version]

[Heroesandvillains]

Re: Stats-Based CPU Sliders [MLB 14 Version]

I've run a lot of cpu vs.cpu games so far and I agree about the strangeness of the stealing this year.

I bumped steal ability up to 7. With that setting through over 50 games I was getting 0.4 steals per team per game and also 0.4 caught stealings per game.

What I'm seeing when I have watched the games is more aggressive baserunning even by slower players. The cpu puts on the hit and run a lot more this year too. Many times with runners on first and second base. That could produce more caught stealings.

There are some new tag animations as well this year. On catcher throws into 2nd base, the 2B or SS is sometimes able to get the tag down quickly on a short hopped throw. In past years, that was almost always a stolen base.

I know it's early but I gotta ask...

Is the steal success slider reversed yet again?

[ralphieboy11]

Re: Stats-Based CPU Sliders [MLB 14 Version]

I know it's early but I gotta ask...

Is the steal success slider reversed yet again?

It's tough for me to tell so far. I don't think I've adjusted the slider far enough in either direction yet. I see nomo is trying at steal ability 10 for awhile, so we should find something out soon.

Plus I'm wondering on these hit and runs is that even tied to steal ability and frequency sliders? Maybe the cpu is programmed to do the hit and run sometimes no matter what the speed of the runners on base.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 14 Version]

It's tough for me to tell so far. I don't think I've adjusted the slider far enough in either direction yet. I see nomo is trying at steal ability 10 for awhile, so we should find something out soon.

Plus I'm wondering on these hit and runs is that even tied to steal ability and frequency sliders? Maybe the cpu is programmed to do the hit and run sometimes no matter what the speed of the runners on base.

Yes, setting the Steal Ability slider at 10 is mostly to see if it is still reversed or not, so will see.

I am actually not sure about hit & run, because we don't really know if those slow runners are taking off because the CPU is instructing them as such... I would expect more hit and runs to "fail" because the hitter fouls the pitch off, which I don't remember seeing often. But I don't really know for sure because I don't watch the game all the time.

I think there is some glitch in stealing this year. Probably not surprising given that the baserunning code has been completely redone.

I have filed a bug report on the stealing issues:

http://www.theshownation.com/bug_reports/10680

Sorry for off-topic post guys...

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

There appears to be something funny going on with both steal frequency and steal success rate. Steal success rate in game is only 36% (vs. 73% in MLB). It looks like there are quite a bit of steals attempted by runners who aren't supposed to steal.

Try running this data again and you'll get different results. ~500 AB for 16 steal attempts is a tiny amount of data and not statistically meaningful for steals (or any event of steal frequency, for that matter). 3000-6000 AB will get you just a ballpark estimate, let alone something you could tune based on.

Let me give some examples. I have three 1500 AB runs from one day. First one showed 19 stolen bases versus 26 caught stealing. Way too many steals! Way too many caught stealing! The very next run of 1500 AB, exact same teams and stadiums, 13 stolen bases versus 6 caugh stealing. Not enough steals! Third one was 22 steals and 12 caught stealing. The fact is, even 1500 AB is a tiny amount of data - I don't even normally look at 1500 AB runs for steals (or wild pitches, bean batter, etc..) for this very reason. 500 AB tells you even less. You'll have to take my word that these were the first three sets I looked at - I didn't cherry pick anything here.

Now maybe it turns out the 4500 AB of data I cited is totally wrong and your 500 AB of data is correct. There's no way for me to prove that. I'm merely saying you shouldn't draw any conclusions on steals with just 500 AB of data either. I have a thousand sets of 500 AB of data and can easily use them to prove anything I wanted.

When I want to test steals, I do longer 6000 AB runs, and the thing is that's still not nearly enough data. All it can do is raise red flags, and then I'll do 3-4 more 6000 AB runs. Steal attempts happen on the order of one a game per team. In other words, even when you combine both teams, one whole game is just two data points. You wouldn't calculate someone's batting average based on 16 At-bats.

The easiest way to prove me wrong is to do this again and get the same results. I know that's a tall order, but reproducibility is the single most important criteria for judging both scientific and statistical findings.

[Heroesandvillains]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Brian his 75 game test included approximately 4,980 at-bats.

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Brian his 75 game test included approximately 4,980 at-bats.

How do you see all 75 games? I see 14.

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

I have just finished running the initial 75 game set with default sliders (except Pitch Speed which will be always at 10).

Did I misread this or did you really do all the tests with pitch speed at 10? Doesn't that affect steal success?

[Heroesandvillains]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Did I misread this or did you really do all the tests with pitch speed at 10? Doesn't that affect steal success?

On the Google DOC, click the "Default" tab.

According to the slider description, pitch speed is a visual cue only and doesn't impact the AI.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Try running this data again and you'll get different results. ~500 AB for 16 steal attempts is a tiny amount of data and not statistically meaningful for steals (or any event of steal frequency, for that matter). 3000-6000 AB will get you just a ballpark estimate, let alone something you could tune based on.

Let me give some examples. I have three 1500 AB runs from one day. First one showed 19 stolen bases versus 26 caught stealing. Way too many steals! Way too many caught stealing! The very next run of 1500 AB, exact same teams and stadiums, 13 stolen bases versus 6 caugh stealing. Not enough steals! Third one was 22 steals and 12 caught stealing. The fact is, even 1500 AB is a tiny amount of data - I don't even normally look at 1500 AB runs for steals (or wild pitches, bean batter, etc..) for this very reason. 500 AB tells you even less. You'll have to take my word that these were the first three sets I looked at - I didn't cherry pick anything here.

Now maybe it turns out the 4500 AB of data I cited is totally wrong and your 500 AB of data is correct. There's no way for me to prove that. I'm merely saying you shouldn't draw any conclusions on steals with just 500 AB of data either. I have a thousand sets of 500 AB of data and can easily use them to prove anything I wanted.

When I want to test steals, I do longer 6000 AB runs, and the thing is that's still not nearly enough data. All it can do is raise red flags, and then I'll do 3-4 more 6000 AB runs. Steal attempts happen on the order of one a game per team. In other words, even when you combine both teams, one whole game is just two data points. You wouldn't calculate someone's batting average based on 16 At-bats.

The easiest way to prove me wrong is to do this again and get the same results. I know that's a tall order, but reproducibility is the single most important criteria for judging both scientific and statistical findings.

Thank you Brian for chiming in.

I think I do have a fairly comparable amount of data suggested. My initial set was from 75 CPU vs. CPU games, and given there are 33.2 ABs on average, with two teams I have 33.2 x 75 x 2 = 4980 AB worth of data now.

Even spot checking, I think there is something funny with the way steals are attempted this year. I don't know if the issue is more with steal success rate, or perhaps the type of players attempting is not right.

My observation so far is that there are a lot more steals are attempted by runners with rather low speed/steal attributes, so there might be something wrong with how the game is triggering steals.

If that helps with debugging, I can try to compile a list of players with SBs and CSs.

[bcruise]

Re: Stats-Based CPU Sliders [MLB 14 Version]

How do you see all 75 games? I see 14.

Clicking on the default tab shows you the entire list.

Also, yeah, I'm pretty sure Nomo runs all his games at pitch speed 10 for aesthetic reasons. I thought the catcher throw speed was adjusted (internally) to compensate for a pitch speed slider change though?

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

According to the slider description, pitch speed is a visual cue only and doesn't impact the AI.

Well if it really says that then it's 100% wrong.

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Clicking on the default tab shows you the entire list.

Also, yeah, I'm pretty sure Nomo runs all his games at pitch speed 10 for aesthetic reasons. I thought the catcher throw speed was adjusted (internally) to compensate for a pitch speed slider change though?

We tried that one year, not since then. It breaks too much.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Clicking on the default tab shows you the entire list.

Also, yeah, I'm pretty sure Nomo runs all his games at pitch speed 10 for aesthetic reasons. I thought the catcher throw speed was adjusted (internally) to compensate for a pitch speed slider change though?

Of course I'd stand corrected if Brian knew the Pitch Speed actually affect the steal aspect of the game, but in the past we are told by a dev (Eddie Cram I believe) that in-game Pitch Speed does not really affect the aspects of the game not related to pitching/hitting.

[Brian SCEA]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Thank you Brian for chiming in.

If that helps with debugging, I can try to compile a list of players with SBs and CSs.

Can you try some future games with sliders reset to default, particularly pitch speed? I'm just curious if this is affecting your results, because it should.

Also, I guess it's fair to say you did all these tests before patch 1? There was a related bug fixed in patch 1 for hit-and-runs, that would make it look like bad runners were stealing when in fact the batter should have swung.

[Heroesandvillains]

Re: Stats-Based CPU Sliders [MLB 14 Version]

Clicking on the default tab shows you the entire list.

Also, yeah, I'm pretty sure Nomo runs all his games at pitch speed 10 for aesthetic reasons. I thought the catcher throw speed was adjusted (internally) to compensate for a pitch speed slider change though?

Yes this has been explicitly stated on OS before. Man oh man I can't even imagine how to go about searching for that quote. LOL!

EDIT: I see Brian already addressed this.