Stats-Based CPU Sliders [MLB 17 Version]

[My993C2]

Re: Stats-Based CPU Sliders [MLB 17 Version]

If you are playing in Colorado, then there may also be some park factors in play.

This wasn't a Coors Field thing when I commented about the HRs the other day. I am 3 games into a 10 game road trip away from Coors Field. Those HRs were hit at Target Field. But there were only 2 HR hits (one by each team) in last night's series finale at Target Field and of course 2 HRs in a game is nothing to be concerned about.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

So a bit of summary from the 5/14v2 and 5/16 (only change is Reliever Stamina 10).

Since both sets are almost identical, I added results from both sets and that's a pretty good sample of 88 CPU vs CPU games, but really 176-game worth of stats since each game is played by two teams. Here's the result:




As have been mentioned, the batting average is ever so slightly lower than the target and HR% is slightly high, but overall the numbers look pretty good, I'd say even the batting average and HR% does not look as as off as I have been worried about.


I would actually consider this a good beta set, since the numbers look close enough to the target that I think trying to further improve on this likely become a game of making trade-offs... doing better in one area would compromise things that's right on at this moment.

I may still want to explore options to bring batting average firmly within .250 - .260, and more importantly slightly reduce HRs (I prefer slightly under producing HRs than overproducing them), but there are a few reasonable possibilities for testing that I'm wondering what should I try.

If anyone who has been following all these adjustments and have some opinions about what adjustment that I should have tried but I have not, please chime in with your comments. If reasonable I may want to incorporate them to my decision on what to test next.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

I would actually mention that the batting average and HR% are actually right on with the MLB averages for the 2017 season so far (BA .250 and HR/G 1.20). So if I were adjusting the game for this year, even these concerns of mine go away (but I'd have to bring SO/G up to 8.19... MLB guys are striking out like crazy these days...).

But the MLB batting average tends to starts out lower in spring and go up a bit toward the summer, so I think it's likely the MLB batting average would land somewhere in .250 - .255 range by the end of this season.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

So I'm wondering what this offensive surge in the 5/18 set means... the only real change from the 5/16 set is Pitcher Consistency from 4 to 5.

If you guess the answer right, you can get a fraction of Clayton Kershaw's $33 million salary (you do the negotiation with him)...

[tgreer]

Re: Stats-Based CPU Sliders [MLB 17 Version]

So I'm wondering what this offensive surge in the 5/18 set means... the only real change from the 5/16 set is Pitcher Consistency from 4 to 5.

If you guess the answer right, you can get a fraction of Clayton Kershaw's $33 million salary (you do the negotiation with him)...

Was wondering the same thing...and the big uptick in K's? Is this where the "that's baseball" term comes in? Just the ebbs and flows of the game?

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

After my last post I've added a few more games in for the 5/18 set, and the offense doesn't look as strangely boosted as it was initially felt. But I think for a relatively small change (Pitcher Consistency 4 -> 5), I am seeing quite a drastic change in how the game plays.

It's sill just about 30 games so far (it was probably around 20 games when I posted last time), so some stats have not fully stabilized yet, but I think the following observations on standard stats would likely remain true:

• Batting average got higher (.249 -> .265)
• HR% got lower (3.2% -> 2.6%)
• Doubles per BIP got higher (6.9 -> 7.7)
• BB% got lower (8.9% -> 7.2%)
• SO% got higher (21.1% -> 23.1%)

This is a set of changes that I would not have expected from such a relatively minor change of increasing Pitcher Consistency by one (looking at the 5/14v2 and 5/16 set, I was only trying to decrease BB% a little bit and also HR% by decreasing poorly-located meat pitches by one stone).

Some clue I can as to what's happening is per pitch stats:

• Foul per swing increased (39.8% -> 41.7%)
• Swing % slightly increased (46.2% -> 47.4%)
• O-Swing % slightly increased (25.5% -> 27.4%)

It seems like what's happening is that by having better command, the CPU pitchers are able to nibble more effectively, which CPU hitters are chasing more, leading to more strikeouts. Since pitches are better commanded, CPU hitters are also not able to square up and hit HRs.

But one puzzling thing is that CPU hitters are not having trouble hitting for the average, and it looks more like they are overall hitting better, only that they now hit more doubles instead of HRs.

Last couple of tabs, I've added "Deep Strikes" to one of the stats that I track, to use it as a proxy for how well CPU pitchers are avoiding throwing meat pitches by mistake... and oddly they are actually pretty similar (about 10.4 - 10.8 per game)... so it's probably not entirely true that all this is happening because the CPU pitchers are effective in avoiding meat pitches.


I think this is one of those moments where the game doesn't appear to transition smoothly over changes in the parameters (which are the things adjusted via sliders) used to tune certain aspect of the game.

Initially things felt so sudden and somewhat glitch-y to me that I thought about filing a bug report (or shooting a PM to Brian the AI programmer)... but I think it's "a feature not a bug," and it is probably exposing some quirkiness of how the game works. It's a good piece of anecdotal evidence that some slider adjustments do not introduce smooth, relatively linear change(s) in the gameplay.

After all these years of working with slider, there are still a lot of game internals not thoroughly clear to me.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

Was wondering the same thing...and the big uptick in K's? Is this where the "that's baseball" term comes in? Just the ebbs and flows of the game?

I just wrote out what I think is happening, though still quite a bit puzzled by the HR-double conundrum.

I would understand better command leading to some conversion of HRs to doubles, but if this is totally by design, I don't think the transition should be so abrupt and the effect rather big.

[rjackson]

Re: Stats-Based CPU Sliders [MLB 17 Version]

I was just thinking about how great it would be if you started tracking deep strikes and now I find out that you have been! Using the 5x5 grid on fangraphs, 9% is the magic number but I haven't been able to get near that. Depending on how the pitcher is doing, it is in the 4% range when confidence is up and 8% when pitching poorly and giving up a lot of runs. I'm beginning to think that it is one of those things that you cannot influence much like fb/gb ratio. It looks like you are getting 7%.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

I was just thinking about how great it would be if you started tracking deep strikes and now I find out that you have been! Using the 5x5 grid on fangraphs, 9% is the magic number but I haven't been able to get near that. Depending on how the pitcher is doing, it is in the 4% range when confidence is up and 8% when pitching poorly and giving up a lot of runs. I'm beginning to think that it is one of those things that you cannot influence much like fb/gb ratio. It looks like you are getting 7%.

I actually didn't realize it's possible to get pitch location for the entire MLB via FanGraphs. I see that they do have heatmaps for individual players, but I don't think I can filter for the entire league/MLB.

Can you actually do that at FanGraphs?

I think you can influence the "Deep Strike" %, if needed, by Pitcher Control slider. I have looked at the batting average as a function of "deep strike" % before, and usually the higher the fraction of deep strikes you/CPU see, the higher the batting average becomes, for obvious reason.

I think I looked at that correlation in the context of some people complaining about "unstoppable" CPU hitting in one of those fruitless comeback code/CPU boost discussions, and some people who make such complaints often play using undesirable settings (e.g., Legend difficulty with Pitcher Control/Consistency lowered significantly), and at the time it was pretty clear that low Pitcher Control leading to more meatballs is at least one of the factors why CPU hit like madmen.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

For now, I have promoted a slightly modified version of the 5/16 & 5/14v2 sets as the beta, and have updated the OP accordingly. I think that's the version that plays most stably with right amount of variation, to the extent I have been able to test things.

For realistic CPU vs. CPU gameplay, using that set is recommended, as opposed to using more recently dated sets, which are rather experimental at the moment.

[rjackson]

Re: Stats-Based CPU Sliders [MLB 17 Version]

I actually didn't realize it's possible to get pitch location for the entire MLB via FanGraphs. I see that they do have heatmaps for individual players, but I don't think I can filter for the entire league/MLB.

Can you actually do that at FanGraphs?

I think you can influence the "Deep Strike" %, if needed, by Pitcher Control slider. I have looked at the batting average as a function of "deep strike" % before, and usually the higher the fraction of deep strikes you/CPU see, the higher the batting average becomes, for obvious reason.

I think I looked at that correlation in the context of some people complaining about "unstoppable" CPU hitting in one of those fruitless comeback code/CPU boost discussions, and some people who make such complaints often play using undesirable settings (e.g., Legend difficulty with Pitcher Control/Consistency lowered significantly), and at the time it was pretty clear that low Pitcher Control leading to more meatballs is at least one of the factors why CPU hit like madmen.

No, you're right. You have to look at a large sample of players and the game does better than I thought initially.

New patch messing with the manager hook slider making it more sensitive and improving some of the logic prior to 6th inning. Good long term, bad for short term slider settings.

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

No, you're right. You have to look at a large sample of players and the game does better than I thought initially.

New patch messing with the manager hook slider making it more sensitive and improving some of the logic prior to 6th inning. Good long term, bad for short term slider settings.

I would rather see patches improving the game than not... so I wouldn't consider the changes on CPU manager hook bad at all. It does take more time for cat-and-mouse game of tuning but I'd rather have a better game for a year than being stuck with some fixable issues, if the trade-off is just a couple extra days of tuning. Will see what really changed (so I'm bringing Manager Hook and Stamina sliders to default a bit).

On the "deep strike" fraction... one reason why I didn't track it till now is that I thought it's one of those stats that would not necessarily come close to the real-life number. (It looks like I track huge amount of stats, but I'm still only tracking those that (1) have some effect and tune-able with sliders and (2) the stats that are easily obtainable in the game and in real life.

I actually expect something like deep strike % not being exactly aligned with the real-life number, because I think the pitchers are generally more accurate in the game than in real life, and CPU hitters do not chase pitches outside strike zone as often as real-life hitters do... so in order for the game to produce realistic outcomes (as it does very well), a little more "variety" of possible outcomes need to be squeezed within the strike zone (e.g., the amount of strikes that happen by hitters chasing in real-life, it has to happen with the pitches in the strike zone instead).

So even if the pitchers in the game are more accurate and therefore throw less meat pitches as deep strikes compared to real-life, the game would still produce realistic outcomes... I'm more concerned with the overall baseball experience being correct than fighting with some compromises the game might be making in detail (they always exist), so I think it's reasonable to not try too hard to match in-game and real-life stats in those occasions (this is also the case for O-Swing/chase %).

[orye74]

Re: Stats-Based CPU Sliders [MLB 17 Version]

By the time we get these sliders right, my seasons gonna be over. 😜

[nomo17k]

Re: Stats-Based CPU Sliders [MLB 17 Version]

By the time we get these sliders right, my seasons gonna be over. ��

That's actually the kind of mindset we should definitely try hard to avoid while adjusting the game to our liking with sliders and whatnot... we can always find whatever flaws by looking harder into details if you go for perfection.

Given its elusive nature, I'd prefer to have a mindset of incremental improvements/adjustments, where I try my best to find the best compromise quickly, and make a series of smaller adjustments, if desired, as I move forward

This year is an exception in that the initially released game was quite terribly tuned which the first patch fixed rather significantly... so that can throw things off, but I personally would move forward with the very initial set that gives me "good enough" stats within ballpark.

I'm slow in releasing "beta" or "final" and such release of slider sets, but that's only because I don't see the point of updating a recommendation frequently given that I make the master data public anyways, and I only want to finalize a set after the final patch for the year drops.

[rjackson]

Re: Stats-Based CPU Sliders [MLB 17 Version]

With pitcher control at 2, it was looking like walks were down for a bit. It is normalizing a bit now. Wondering if you ever considered ticking up strike frequency to 6 - FP STK% seems consistently low although overall strick% is pretty good. The 1-0 count is good for hitters as well. Though with 2-5-5 cont-cons-freq settings, it is interesting that walks and HR's are down while AVG is up. If you did tick up freq, it probably affects other stuff as well but I'm not sure what. Contact I'm guessing.