AI Bullpen usage logic is the worst

[Jhack99]

Re: AI Bullpen usage logic is the worst

Decided to buy MLB 20 on sale for $25. This bullpen logic is still of course a huge problem after only playing one game. So dug up this thread to refresh my memory. In the process of adjusting AI bullpens and will quick manage literally every game to get the best out of it.

Sent from my LG-G710 using Operation Sports mobile app

True, but it will still play havoc on any games you sim (cpu teams). Unless you are gonna quick manage every single game. This year in particular the CPU (sim or actual gameplay against me) LOVES to throw the long guy all the time.

Its gotten to the point where I had to adjust some players stamina for the cpu, plus rearrange the bullpen setup. I'm using 30 team control but trying to let the cpu handle everything, but its irking me how so many "Long guys" have (110+ innings pitched for some guys).

Side note I do use adjusted sliders, and its still a issue. Plus they LOVE to have as many long guys as possible for some reason. Arizona in my franchise keeps trying to have 3 of them. One of them has 14 wins already and got there before I even finished August.

[TheWarmWind]

Re: AI Bullpen usage logic is the worst

Decided to buy MLB 20 on sale for $25. This bullpen logic is still of course a huge problem after only playing one game. So dug up this thread to refresh my memory. In the process of adjusting AI bullpens and will quick manage literally every game to get the best out of it.

Sent from my LG-G710 using Operation Sports mobile app

I thought I had posted my solution in this thread but I guess it was another thread.

I've discovered something odd about CPU bullpen management. It seems one of their biggest concerns is maintaining pitch effectiveness of their bullpen, or to put in another way, keeping their bullpen's stamina's above the point where they start losing effectiveness.

The issue arises because the pitching stamina sliders are too low (likely balanced for online play) causing that point to come way too early on most relievers. This triggers the logic of leaving one guy in to get overly tired in order to protect the rest of the bullpen far, far too often. The computer cares more about protecting their bullpen then winning, and you'll get cases where the CPU might trot out the same guy a few nights in a row to get slaughtered in order to protect a bullpen they never seem to want to use.

The key is to raise the pitching stamina sliders. This pushes the effective pitch count number much higher (while having a much smaller impact on total pitches before tired). This triggers the "protect the rest of the bullpen" logic FAR less often and sees bullpen's used far more logically more often as the CPU is more willing to use the bullpen thanks to the increase in effective pitches.

It's not perfect, I still see the odd head scratcher, but I've made no CPU rotation edits or opposing pitcher attribute edits and more often than not I agree with the CPU decision making.

I forget the exact numbers, but you can check out my sliders in the sliders sub-forum. I will say this: I currently like CPU manager hook at 6 slightly better than at 5 for bullpen management, but I found starters were pulled to early to often on that setting, and 5 was still acceptable for the bullpen. I have it set to 5 because I'm a set it and forget it kind of guy. If you're willing, you'll probably see even better results of you set CPU manager hook to 6 once the starter leaves.

Sent from my Pixel 3 using Operation Sports mobile app

[Jhack99]

Re: AI Bullpen usage logic is the worst

I thought I had posted my solution in this thread but I guess it was another thread.

I've discovered something odd about CPU bullpen management. It seems one of their biggest concerns is maintaining pitch effectiveness of their bullpen, or to put in another way, keeping their bullpen's stamina's above the point where they start losing effectiveness.

The issue arises because the pitching stamina sliders are too low (likely balanced for online play) causing that point to come way too early on most relievers. This triggers the logic of leaving one guy in to get overly tired in order to protect the rest of the bullpen far, far too often. The computer cares more about protecting their bullpen then winning, and you'll get cases where the CPU might trot out the same guy a few nights in a row to get slaughtered in order to protect a bullpen they never seem to want to use.

The key is to raise the pitching stamina sliders. This pushes the effective pitch count number much higher (while having a much smaller impact on total pitches before tired). This triggers the "protect the rest of the bullpen" logic FAR less often and sees bullpen's used far more logically more often as the CPU is more willing to use the bullpen thanks to the increase in effective pitches.

It's not perfect, I still see the odd head scratcher, but I've made no CPU rotation edits or opposing pitcher attribute edits and more often than not I agree with the CPU decision making.

I forget the exact numbers, but you can check out my sliders in the sliders sub-forum. I will say this: I currently like CPU manager hook at 6 slightly better than at 5 for bullpen management, but I found starters were pulled to early to often on that setting, and 5 was still acceptable for the bullpen. I have it set to 5 because I'm a set it and forget it kind of guy. If you're willing, you'll probably see even better results of you set CPU manager hook to 6 once the starter leaves.

Sent from my Pixel 3 using Operation Sports mobile app

Thanks for this. I'll try out your sliders for the cpu pitchers and relievers, and see what kind of results I get.

[DarthRambo]

Re: AI Bullpen usage logic is the worst

I plan to quick manage every game I don't play, cpu vs cpu as well. Use two controllers and it works pretty well. I wanted to comment on this thread again so I can look through the workaround from last year that worked pretty well.

Sent from my LG-G710 using Operation Sports mobile app

[DarthRambo]

Re: AI Bullpen usage logic is the worst

True, but it will still play havoc on any games you sim (cpu teams). Unless you are gonna quick manage every single game. This year in particular the CPU (sim or actual gameplay against me) LOVES to throw the long guy all the time.

Its gotten to the point where I had to adjust some players stamina for the cpu, plus rearrange the bullpen setup. I'm using 30 team control but trying to let the cpu handle everything, but its irking me how so many "Long guys" have (110+ innings pitched for some guys).

Side note I do use adjusted sliders, and its still a issue. Plus they LOVE to have as many long guys as possible for some reason. Arizona in my franchise keeps trying to have 3 of them. One of them has 14 wins already and got there before I even finished August.

I will quick manage every game using two controllers and it works great. Only takes about 10min per game so not crazy long.

And yeah going through AI bullpens and the Pirates have THREE starting pitchers in their bullpen! What in the hell is that?

Sent from my LG-G710 using Operation Sports mobile app

[garry1221]

Re: AI Bullpen usage logic is the worst

I've discovered something odd about CPU bullpen management. It seems one of their biggest concerns is maintaining pitch effectiveness of their bullpen, or to put in another way, keeping their bullpen's stamina's above the point where they start losing effectiveness.

The issue arises because the pitching stamina sliders are too low (likely balanced for online play) causing that point to come way too early on most relievers. This triggers the logic of leaving one guy in to get overly tired in order to protect the rest of the bullpen far, far too often. The computer cares more about protecting their bullpen then winning, and you'll get cases where the CPU might trot out the same guy a few nights in a row to get slaughtered in order to protect a bullpen they never seem to want to use.

The key is to raise the pitching stamina sliders. This pushes the effective pitch count number much higher (while having a much smaller impact on total pitches before tired). This triggers the "pct the rest of the bullpen" logic FAR less often and sees bullpen's used far more logically more often as the CPU is more willing to use the bullpen thanks to the increase in effective pitches.

It's not perfect, I still see the odd head scratcher, but I've made no CPU rotation edits or opposing pitcher attribute edits and more often than not I agree with the CPU decision making.

Your theory becomes more sound everytime i notice quirks arise. As a strict cpu-cpu sim player I'm noticing things this year that differ from last. Maybe I'm rethinking my way wanting to use my pen as well. I want my SP tossing 80 pitches per start on average, hopefully getting through 6 or 7 innings.

My past bullpen consisting of 1 LR, 4 MR, 2 SU and a closer wasn't giving me the results I was looking for this year. My single LR was getting overworked, and my SU guys hardly saw the mound. I've had good tests with 2 LR, 4 MR, 1 SU, 1 Closer. I've also seen good results with 3 LR, 4 MR and a closer. Tested across multiple teams to assure consistency. I've found the CPU will use LR1 more than 2, who will be used more than 3. (I might rotate these guys every few games to see if its position, OVR or slot based) It might even out in time, but in my sample size, i thought it significant. In mosts tests, MR's were evenly used. I'm still seeing little head scratchers, but so long as a LR doesnt throw 2+ IP in 3 consecutive games, I feel like it's as close as we may get.

I modified Josh's sliders a bit, because ratings SHOULD matter. For CPU-CPU I took SP Stamina down to 2. Any higher and the number of CG's goes through the roof over the course of a season. RP stamina stays at 5. Any lower and relievers fall apart wzy too quick. Manager hook set to 0 IMO works best. It feels very in the situation, though in some boxscores you occasionally wonder why someone didn't get pulled quicker. Any higher than 0, i recall seeing too many "why'd he get pulled so soon" or "why hasn't he been pulled" situations. With these settings, my LR are the only bullpen arms to toss 2+ IP per game. My MR haven't gone more than 1.2 per game.

If I'm actually playing, SP stamina stays at Josh's recommended 6.

I typically keep 6 SP on my roster, so my LR are 1 sp and 1 or 2 28+ STA RP. I've noticed position trumps slot placement. My LR SP will rarely get put in past the 6th, unless we're in extra innings, but my LR RP's will be used anytime.

[garry1221]

Re: AI Bullpen usage logic is the worst

Thinking about past franchises, i recall the cpu heavily favoring SP as opposed to RP. I started wondering if there was a better reason than just SP's being a dime a dozen, as far as quantity. My aim is to have fairly even MR usage.

With 2 SP's in long relief, 4 MR and a closer, i found my most consistent results. My trials are 30 games long. My initial test with the pirates, went 10-20. LR1 threw in 16 games, LR2 threw in 14. MR were between 9ish and 13ish IP. My LR's didn't toss more than 6 IP in back to back games and only my highest STA MR threw more than 3 games in a row (he had 4 games/4 IP, mid 50's STA).

I also tried 8 man pen tests. 1 with a high STA RP in LR3 and one with a high PCL in SU1. In both of these tests usage and consistency got thrown out the window. The 1st, the MR in LR3 only saw 5 IP and my MR3 got left on the bench for 2+ weeks between use. The latter saw SU1 used as sparsely as the closer (this result was expected, but wanted to get data).

Still want to run a few more trials with various teams, but it looks like best pen use comes with a 7 man cpu pen. I know many, including myself, like an 8 man pen, but if it means smarter cpu use, I'll gladly swap.

[garry1221]

Re: AI Bullpen usage logic is the worst

After more tests with similar results, I can say a 7 man pen is the closest we're going to see to smart pen usage. 2 SP in long relief, 4 RP and a CP. I found my most consistent results with NL teams, but steady results across the board still. AL teams leaned harder on MR1 & 2, than 3 & 4. While that was irritating, it was at least consistent.

Warmwind's 'save effective pitches/pitchers' theory still holds strong, and unfortunately that's the biggest stumbling block to seeing anything smarter. The only true fix is adding/writing new code. A MR should see the mound at least once every 7 to 10 days, max. LR should only enter after the 6th if the MR are all fatigued (if they have 2+ IP in the 2 previous days or 3+IP/3days (ideas)) and/or the game is tied in late innings (8th/9th). Hell, if SU1+2 weren't strictly used in 'set up a save' situations, it would solve alot as well while adding another arm to the pen.