AI Bullpen Logic

[GAMEBREAKER85]

Re: AI Bullpen Logic

Sounding like nothing has changed for MLB24???

I am in my last week of September in my MLB23 Franchise - yes I played every inning of every game and I manage everything for all 30 teams.

I have a set of sliders for playing games, I have a set of sliders for simulating Minor League games and I have a set of sliders for simulating Major League games.

I've had to manually injure pitchers at times to keep their appearances down and move them into different bullpen slots. I can tell you for MLB22 & MLB23 the stamina for bullpen pitchers only does so much. There are multiple things you have to do to keep appearances and innings minimized. It was the exact same for MLB22 & MLB23.

[ltw0303cavs]

Re: AI Bullpen Logic

Hope people keep poking at this a bit more. Whether they find some workable sliders or not let us know. And if it's really not improved I'll do a story early in the year to re-highlight it.

And if anyone wants to make it clear at a high level what the biggest problems are once again, by all means do that as well to freshen it up for the new season.

Perfect, definitely going to try different things and post. To me the AI does not maximize minor league options to bring up and bring down Relievers which is a great work around to running your relievers into the ground. The other issue is when AI uses overall ratings to determine depth charts and pitchers stuff is not used or weighted properly in overall ? That's a problem. For example look at the relievers in baseball, a lot are low command. Which is why they are relievers. However command affects Overall but a 99 Break on pitches is not figured in or given enough weight in the overall?

[GAMEBREAKER85]

Re: AI Bullpen Logic

There are so many components to this issue one that I never see mentioned on here is how many pitchers anymore fluctuate from starters to relievers and vice versa. This wasn't the case 5-10 years ago I don't feel like. The Show's ratings are altered for relievers v. starters (kper9, Hitsper9 etc.). So when I moved Seth Lugo to my bullpen after acquiring a starter in my '23 Franchise he wasn't going to be the same pitcher he was as a starter b/c his ratings were set for him to be a starter. So many little nuances need tinkered with when trying to get accurate bullpen usage / stats.

If you change a closer to relief pitcher or vice versa you will see almost every pitcher's overall rating will change for better or for worse. Same thing with moving starters and relievers to the rotation/bullpen.

In turn this can lead to a guy being a "bad" pitcher and when the CPU uses him in simulation he is going to struggle possibly which will lead to the CPU having to over-use other relievers.

Etc. Etc. Etc.

[MiracleMet718]

Re: AI Bullpen Logic

So I finally got to do some simming analysis in 24 using my roster changes and three small tweaks to sliders:

Roster Changes
- RPs that should be SPs were switched to SPs (guys who have made mostly starts but were RPs to fit in the roster, like Mason Miller)
- All RPs with stamina under 50, most under 40
- Updated K/9 for half of the Starters to get accurate K stats (based on percentage rank in K/9 last year)

Slider Changes
- CPU Starter Stamina = 1
- CPU Reliever Stamina = 0
- CPU Manager Hook = 8
- NOTE: when playing a game, I used different sliders to adjust pitching innings last year, this is for simming only (happy to share those, although I haven’t tested in 24 yet)

Real MLB Stat Rules for RPs
- IPs based on pitchers who made >50% of their appearances as relievers in real life


I simulated only 5 seasons quickly and took averages, but here were my findings:

Pitching Stat Findings
- # of Relief Pitchers with Innings over 80 in MLBTS vs. Real MLB: 47 vs. 23
- # of Relief Pitchers with Innings over 90 in MLBTS vs Real MLB: 27 vs. 8
- # of Relief Pitchers with Innings over 100 in MLBTS vs Real MLB: 9 vs. 5
- # of Starters with Innings over 200 in MLBTS vs. Real MLB: 4 vs. 5
- # of Starters with 200+ Ks in MLBTS vs Real MLB: 12 vs. 17
- # of CGs in MLBTS vs Real MLB: 33 (22 SPs, highest with 4) vs. 35 (25 SPs, highest with 3)
# of SHO in MLBTS vs Real MLB: 28 (20 SPs, highest with 4) vs. 21 (19 SPs, highest with 2)

Next steps for me is to update all of the RPs to under 35 or 40 stamina. I also want to try messing with the injury slider since I think the major issue is not enough RPs get injured, optioned, or sent down, which is inflating the RPs at the high end of the spectrum. I’m also going to fix the K/9 for the rest of the SPs and some of the RPs as well to get the 200 K pitchers up a bit.

Also I haven’t tested it, but I don’t think I’ll have a solution for the high number of saves. Every year the top guys are at 45-50+ as it seems they get the majority of chances. I might fool around with making every CP an RP instead and see if that changes anything.

Let me know what you think or if you have any questions.

[Christian9]

Re: AI Bullpen Logic

Really appreciate the insight/analysis. I ran different scenarios repeating same season and came up with some of the same conclusions, but no one seemed interested.

One of the things was trying to see was whether moving players around in different slots would produce much different results. Have been busy today but I need to dig in and see if Clutch rating played a big difference in IP and Hld/Sv opps.

Your summary is very helpful. Thanks!


Sent from my iPhone using Operation Sports

[ripwalk]

Re: AI Bullpen Logic

With all of this talk, I was curious to run some numbers. I started a spreadsheet with a few scenarios mentioned here.

Turned off injuries, trades, and used the Phillies (out of familiarity). Definitely saw some interesting numbers, which I haven't fully gone through yet. Was hoping it would pique some interest and we can see some patterns, or at least hopefully understand better how the logic is supposed to work.

I agree with others that this should not be up to us finding a workaround, but figured I'd be proactive.

Doesn't look like I'm able to attach a spreadsheet, so had to zip it first.

Let me know what you think. And/or any scenarios worth testing.

Thanks for doing this, lots to digest. I’m looking through it now … but , have you extrapolated anything final from it yet in terms of take home message or settings best to use ?

[ripwalk]

Re: AI Bullpen Logic

I don’t particularly care if MLB stats are mirrored perfectly (I play a fictional/custom league anyway) …

So the priority of things I care about is…

1. Crappy poorly rated LR not having the highest IP by a long shot.

2. Best rated RPs getting most innings.

3. All good pitchers getting used (ie maybe not using a good pitcher in SU2 is important as I don’t think there’s every really a way to get them innings.

4. To a much lesser extent, SP total innings and CG numbers .. I’d prefer them not to be like 50 people throwing 200+ innings but I don’t care if CGs are a little high.

All that considered I feel like the slider sets that look the best to me are (listed from top to bottom in your spreadsheet)

5/5/5/5/10 and 4/2/4/2/9

Those to me have the right balance I’m looking for, or close to it.

I’d still love to see a slider set that has LR throw less than 2 IP/G.

You tested LR with most stamina in a couple cases. I’m curious to see how numbers would turn out with the above sliders but LR with least stamina put there.

Alternatively, one of the better relievers going in that spot so they get most innings could work.

I’m very curious to hear what conclusions you’ve drawn from your data.


And anyone who hasn’t checked it out yet, definitely download the spreadsheet. It’s really well done and informative.

[Christian9]

Re: AI Bullpen Logic

Thanks for doing this, lots to digest. I’m looking through it now … but , have you extrapolated anything final from it yet in terms of take home message or settings best to use ?

Not yet. I want to see how it plays out when I follow the LRP slots next. Busy weekend, so haven’t had a chance to do that or find patterns yet.

desouza7 has some great info as well (back a page).

More to come.


Sent from my iPhone using Operation Sports

[ripwalk]

Re: AI Bullpen Logic

Not yet. I want to see how it plays out when I follow the LRP slots next. Busy weekend, so haven’t had a chance to do that or find patterns yet.

desouza7 has some great info as well (back a page).

More to come.


Sent from my iPhone using Operation Sports

I had been using 4/2/6 (SP stamina/RP stamina/Manager Hook) but I think I’ll switch my testing to one of yours as I like those results better than what I was seeing.

Ultimately I’ll have to settle on a slider set that works best, then I do think I’ll bite the bullet and edit each bullpen at start of year to either remove anyone from SU2, or put worst reliever there . As I don’t think any setting will prevent that slot getting least inning …

And then need to determine if it’s best to put a low stamina as LRP, or maybe just go with one of the best relievers there, as it also doesn’t seem possible to not have this slot end up with most IP.

[ChaseB]

Re: AI Bullpen Logic

[MiracleMet718]

Re: AI Bullpen Logic

I’m still testing. I did get to test the RP vs CP for saves and it makes no difference. So I’m onto the injury slider next. Hope to have more findings this week.

[MiracleMet718]

Re: AI Bullpen Logic

Update - rather than testing anymore with sliders updates, I checked the stats on # of pitchers used in MLBTS vs. Real Life and as I suspected, this is the real issue with the reliever stats being off.

Just using the Mets and Phillies as an example, in real life last year they used 35 and 44 total pitchers respectively. In my tests of MLBTS, the average number of pitchers I was seeing used by teams was only 15-18 different pitchers (I checked minors pitchers to see who was previously called up so I didn’t just look at the MLB guys).

I then decided to up the simulated injury slider to 10 and see what would happen and the results were pretty much the same with maybe 1 or 2 more pitchers being used on average.

So I don’t think this is a slider or attribute issue as much as it’s a roster issue and not having enough RPs or managing the optionable pitchers like MLB teams do. I have gotten pretty close with sliders, but it’s either you have a few more 200 inning SPs and <90 inning RPs or you have less 200 inning SPs and a few 100 inning RPs.

[ML]

AI Bullpen Logic

Does anybody know if the devs have ever even acknowledged this? I feel like its a small minority who bring this up year in and year out and it continues to go untouched despite it being broken.


Sent from my iPhone using Operation Sports

[ripwalk]

Re: AI Bullpen Logic

Does anybody know if the devs have ever even acknowledged this? I feel like its a small minority who bring this up year in and year out and it continues to go untouched despite it being broken.


Sent from my iPhone using Operation Sports

I know I’ve read somewhere (heck it may even be first couple pages of this thread), that they were at least aware of it.

[ML]

Re: AI Bullpen Logic

I know I’ve read somewhere (heck it may even be first couple pages of this thread), that they were at least aware of it.

Thanks - I saw on the first page that it was allegedly improved this year but continuing to read the thread it appears to still be bad. I'm so on the fence because I really WANT to love the Show but I just can't justify myself supporting this game at full price at the moment even if I have the surplus to buy it.