Stats-based sliders for CPU

[PsychoBulk]

Re: Stats-based sliders for CPU

I read similar discussions from the last year's threads. IIRC from those threads, the rationale to increase the consistency/control in an attempt to increase BB is that when pitcher AI has more confidence in their command, they keep painting corners (rather than pitching down the middle) for strikes in hitters count, and hence sometime these end up in walks, whereas pitches aimed at center of zone rarely end up in balls.

This is, give or take, true.

[nomo17k]

Re: Stats-based sliders for CPU

Yes, it does make sense if AI is programmed to be that way and I'm not surprised if upping control/consistency end up in more walks. But I couldn't find BB/K stats with the slider setting (with at least about > 30 games) posted last year.

I'm about 10 games in with pitcher control zero, and walks are still low (still small sample). So it might be good to start looking for yet another possible way to increase walks. Maybe someone with the increasing control/consistency post stats? If nobody does I'll rack up stats but I'd do at least 30 games with the current sliders before changing anything.

[Heroesandvillains]

Re: Stats-based sliders for CPU

In theory, raising control will cause pitchers to paint more. Thus, missing off of the plate...BUT...

This completely ignores the pitchers individual/pitch confidence. Once confidence rises, BB are nearly non-existent.

Averages will also go down, inflating confidence, and lowering walks.

This series is remarkably confidence driven. This is why game results are so variable from day to day. 10-1, 4-3, 13-4...etc, etc.

[Heroesandvillains]

Re: Stats-based sliders for CPU

Also keep in mind, lowering control will increase meatballs.

I'm itching to see these results.

[nomo17k]

Re: Stats-based sliders for CPU

After 15 games I think it's pretty likely BB isn't going to increase substantially enough with pitcher control set to 0. I get 2.6 BB per game per team so far, which is only slight increase from the default; not very likely 0.7 BB deficit will be made up from now on by playing more games. Interestingly strike % doesn't even seem to change much either.

To test other possibility, I'll set pitcher strike frequency to zero for now for the next couple dozen games. Rationale is simple: make pitchers pitch from behind early in the count to make 3-X counts happen more often. I'll set pitcher control to default (5).

That's the only change I'll be making for now.

[Heroesandvillains]

Re: Stats-based sliders for CPU

Nomo, that's a bummer.

Though it's what I'd suspected, it looks like ParisB is spot on.

Dropping consistency in tandem with with strike frequency to get walks...

Or...

You could try Snake's theory of raising control and consistency.

However, since Daiyoung's lowering consistency seems to have established just about enough walks, it may be better to go that route for now (consistency and strike frequency)...and living with the WP. Clearly, since HBP are still too low with lowered consistency, the problem has to be most likely address via (a.) patch, or (b.) catcher ratings...both in order to combat the baserunner aggressiveness of knocked down balls by the catcher.

[nomo17k]

Re: Stats-based sliders for CPU

I think lowering the pitcher control slider does help a bit (remember I have contact lower a bit which isn't favorable for walks; but it has a good effect on K; I see more K and whiff so far. Anyway will post stats later...), but the effect is not so large to fill the almost 1 BB gap.

Good thing is we will be testing control, consistency, and strike frequency (almost) independently, so we will know which slider has the most effect on BB if varies independently.

We'll see how DaiYoung's stats come out. The concern obviously is WP, HBP, things of that sort. I'm also curious if his offense numbers will stay low.

[Snake.Plissken]

Re: Stats-based sliders for CPU

The reason why I try to go by stats in this thread is numbers are basically the best we can do to be objective, and philosophies are often unfounded. So I'm not sure what to take away from the suggestions to increase control/consistency since nobody has posted stats with sufficient sample size. If all else fail, I might try upping those to see what's suggested really happen. But if you already have the stats to substantiate your claim, could you share some stats?

I don't play CPU vs CPU games and I've just nailed done my slider set recently so I don't have a big sample size. I play the games myself and I was able to draw 3 walks off Strasburg in 5 innings (a 4 pitch walk as well) and in another game, I didn't draw a walk until the 11th inning where I was down to my last out and I drew 2 consecutive walks. I wouldn't have been able to draw those walks if the pitches were constantly JUST missing the corners as I was down to my last batter and didn't have the luxury to take borderline strikes.

I'm not exactly sure what's going on in the programming but I'm seeing plenty of balls miss the strike zone by a good margin not just off the corners. At control 7 & consistency 7 even control pitchers are really wild and miss the zone by a lot quite often, too often.

Steal Ability: 3

Do you have any stats to determine if this slider is doing anything? I did a test with it at 0 and it at 10 (since testing stealing is easy to do) and I didn't notice any difference in the results. You would probably have to record a good sample size of data to see the difference. At 0, stealing should be noticeably harder while at 10 stealing should be noticeably easier especially when compared against the 0 setting.

[nomo17k]

Re: Stats-based sliders for CPU

Do you have any stats to determine if this slider is doing anything? I did a test with it at 0 and it at 10 (since testing stealing is easy to do) and I didn't notice any difference in the results. You would probably have to record a good sample size of data to see the difference. At 0, stealing should be noticeably harder while at 10 stealing should be noticeably easier especially when compared against the 0 setting.

How many steal attempts are you using to make your conclusions? I'll keep the steal related sliders as it to rack up more stats, but my tentative conclusion is that I'm seeing more CS (by decreasing steal ability) and a lot more steal attempts (by increasing steal frequency). The way I intended them to be! So I'm happy there so far.

[DaiYoung]

Re: Stats-based sliders for CPU

How many steal attempts are you using to make your conclusions? I'll keep the steal related sliders as it to rack up more stats, but my tentative conclusion is that I'm seeing more CS (by decreasing steal ability) and a lot more steal attempts (by increasing steal frequency). The way I intended them to be! So I'm happy there so far.

Good to hear that the steal success slider is the right way round this year! That slider completely baffled me for weeks last year.

[Snake.Plissken]

Re: Stats-based sliders for CPU

How many steal attempts are you using to make your conclusions? I'll keep the steal related sliders as it to rack up more stats, but my tentative conclusion is that I'm seeing more CS (by decreasing steal ability) and a lot more steal attempts (by increasing steal frequency). The way I intended them to be! So I'm happy there so far.

I used Seattle as my test with Ichiro, Figgens, and Guitierrez as my test runners. I would intentionally walk them then switch teams and have them steal with no extended leadoff, I did at least 10 steals at the lowest and highest slider setting. I really did not notice a difference between the slider at 0 and 10. I expected steals to be basically automatic at 10 with Ichiro but they weren't and he got thrown basically as much as at 0. Even steals of third didn't seem to alter much. Steals are too reliant on the jump. If every runner just got the same jump, then it would be so much easier to balance steals. A guy with say 75 speed should always get thrown out if he ran on a fastball and the catcher made a good throw. Even the best base stealers don't have a great success rate when they run on a pitcher that throws a fastball and a catcher makes a solid throw. They are depending on the catcher being a bit off on the throw or running on a changeup or the catcher receiving a pitch that's not optimal for a catcher to throw off of.

[Heroesandvillains]

Re: Stats-based sliders for CPU

I used Seattle as my test with Ichiro, Figgens, and Guitierrez as my test runners. I would intentionally walk them then switch teams and have them steal with no extended leadoff, I did at least 10 steals at the lowest and highest slider setting. I really did not notice a difference between the slider at 0 and 10. I expected steals to be basically automatic at 10 with Ichiro but they weren't and he got thrown basically as much as at 0. Even steals of third didn't seem to alter much. Steals are too reliant on the jump. If every runner just got the same jump, then it would be so much easier to balance steals. A guy with say 75 speed should always get thrown out if he ran on a fastball and the catcher made a good throw. Even the best base stealers don't have a great success rate when they run on a pitcher that throws a fastball and a catcher makes a solid throw. They are depending on the catcher being a bit off on the throw or running on a changeup or the catcher receiving a pitch that's not optimal for a catcher to throw off of.

This may be decent info for me...but sort of irrelevant for CPU and MOM guys, as there is no option to turn off one step leads.

Might I ask...or maybe post it in your thread, why you'd test steals with no leads? Last year, that worked. But that means your assumption was that SCEA did not work on their stealing logic this year. I can tell you for certain, stealing in general is much improved now. Not perfect, but improved nonetheless.

[nomo17k]

Re: Stats-based sliders for CPU

Even steals of third didn't seem to alter much. Steals are too reliant on the jump. If every runner just got the same jump, then it would be so much easier to balance steals.

This is exactly why you cannot make any conclusions *on the effect of slider* from your experiment. In order to isolate the effect of slider, you have to repeat the same experiment under exactly the same condition. Yes, that only happens in ideal world, so the next best thing is to control your environment.

In this case, I would just use one base stealer, steal off the same pitcher on the same pitch (maybe fastball). And you need to repeat this many more than just 10 attempts. Then compare the results from different slider setting.

Just to put into perspective: if you get two base hits in 10 attempts, you are a terrible hitter. Three, you are decent. But that difference of one can just be due to coincidence, pure luck. Just hitting a screaming line drive right at the center fielder, if caught, changes the results, heaven and hell.

That's why I don't weigh any conclusions made based on only a few games.

[Snake.Plissken]

Re: Stats-based sliders for CPU

This may be decent info for me...but sort of irrelevant for CPU and MOM guys, as there is no option to turn off one step leads.

Might I ask...or maybe post it in your thread, why you'd test steals with no leads? Last year, that worked. But that means your assumption was that SCEA did not work on their stealing logic this year. I can tell you for certain, stealing in general is much improved now. Not perfect, but improved nonetheless.

I did it with no leads just to see how much a 0 setting could impact steals. Also, I wouldn't have to worry about the pitcher throwing over. Then, once I got steals to work with no leads, I was ideally going to lower the slider a notch to then require that steals would need a lead of some kind to have a good success rate.

This is exactly why you cannot make any conclusions *on the effect of slider* from your experiment. In order to isolate the effect of slider, you have to repeat the same experiment under exactly the same condition. Yes, that only happens in ideal world, so the next best thing is to control your environment.

In this case, I would just use one base stealer, steal off the same pitcher on the same pitch (maybe fastball). And you need to repeat this many more than just 10 attempts. Then compare the results from different slider setting.

Just to put into perspective: if you get two base hits in 10 attempts, you are a terrible hitter. Three, you are decent. But that difference of one can just be due to coincidence, pure luck. Just hitting a screaming line drive right at the center fielder, if caught, changes the results, heaven and hell.

That's why I don't weigh any conclusions made based on only a few games.

I realize my experiment wasn't perfect but my goal was just to be able to a get feel for what the slider is doing and how sensitive it is. In past years, the steal ability slider has had a noticeable affect on stealing just moving it a notch or two but I didn't see any noticeable change just by minimizing and maximizing the slider, which I feel you should see a noticeable difference without requiring data to be recorded and analyzed. Kinda like how you could just see the difference in power if you were to put it at 0 or 10.

[nomo17k]

Re: Stats-based sliders for CPU

Guys, if you are losing sleep over how your CPU pitchers are not walking hitters, I think the best option is to lower the strike frequency slider! Upon reflection this is obvious, and this may be a better option so we don't necessarily have to mess with pitch consistency to avoid increasing WP.

Okay, so for completeness, I'll post updates after 15 and 5 games with the following slider adjustments. Small number of games for sure, but I'm fairly confident this is the right way to go given pitch strike % is significantly different between the two sets; the number of pitches are in thousands so these stats are fairly sampled well.

With pitcher control zero, strike frequency 5:




With pitcher control 5, strike frequency 0:




There are a couple extra innings games, but BB 2.6 to 4.7 per game is a huge difference, and strike % goes down from 64.3% to 60.6%. In early counts (started tracking these), it goes down from 61.9% to 55.7%. I obviously decreased the slider a bit much, but I like what I see.

I'll now make the following adjustment. On scales of 0 - 10 (5 default):

Contact: 4
Power: 4
Foul Frequency: 3
Strike Frequency: 2
Fielding Error: 9
Throwing Error: 4
BR Steal Ability: 3
BR Steal Frequency: 8

Everything else is at default (5).

Comments: With contact at 3, I got hitters to swing and miss more than the MLB ave. (which also shows up in higher K). It might've been too much, so I'm setting contact to 4. I lowered foul frequency because I've been consistently above the MLB average.

I set strike frequency to 2, so that the early-count strike % becomes slightly above the MLB ave. (58.7%, which is a rough guess of mine; it's the first pitch strike % in MLB). Hopefully this causes BB to be closer to the MLB ave.

I'm seeing way more throwing errors than fielding errors, so I upped fielding error slider but lowered throwing error slider. I lowered steal frequency to better match the MLB ave.