AI Bullpen Management

[BrooksKieschnick]

There's one thing in OOTP that's consistently been an issue for me: the AI fails to adjust its rosters after games with heavy bullpen usage. I play standard MLB quickstarts, and it seems like there's always a stretch in July or August where I'm playing against a team that doesn't have any rested relievers. My opponent will use the starter for 150 pitches or use tired/exhausted relievers for one or two batters at a time, until eventually the last available bullpen guy just pitches forever. There have been so many times where this has allowed me to run up the score in the later innings, and it feels like a cheat.

Why doesn't the AI option tired relievers to the minors until they recover and bring up fresh arms to replace them? Is this something that could be implemented?

[DaBears]

The AI is a few years behind current MLB option strategies, no question.

[waittilnextyear]

Good thread. This is an area of AI that could be improved upon. I remember Justin Dunn of the Mets making 160+ pitches against me in a similar situation. This happens about 3-5 times per season give or take. All of my AI settings about SP/RP management are set to default modern day.

[Le Grande Orange]

I would add in the additional consideration of pitcher usage differences between the majors and the various levels of minor leagues (as well as spring training majors versus regular season versus post-season majors).

[Orcin]

Second the OP. This is the biggest AI weakness. I win 5-10 extra games per year due to tired AI bullpens.

[MizzouRah]

It's a must IMHO to go with 26 man rosters and enforce 13 pitchers.

[SirMichaelJordan]

Originally Posted by MizzouRah

It's a must IMHO to go with 26 man rosters and enforce 13 pitchers.

I force 13 pitchers with a 25 man roster. Teams are starting to do this anyways.

I also set hook reliever to very quick which helps some.

[NoOne]

i think a better depth chart would help.

get rid of "secondary role"

simply have lists for each role, and which names you put there are in the pecking order for that role -- use frequency would still be nice for further stratification within that list... or a value system to rank them further within each list, since often times #3,4,5, isn't good enough if oyu only want 3 maximized and 4/5 minimized etc.

i think the AI can handle that better and make better qualifications in addition to those lists as to when / where someone is used.

it definitely needs to know when to extend arms as opposed to 1PA outings that leads to 4 fielders pitching. (seen it happen recently).

math should make it fairly easy too, i'd think. time it takes to recover, # of pitchers total, # of pitchers available and various thresholds that work out. some sort of "oh %$#@" level of running out of bullpen "gas" that's calculated from know information in game.

[Dyzalot]

Originally Posted by SirMichaelJordan

I force 13 pitchers with a 25 man roster. Teams are starting to do this anyways.

I also set hook reliever to very quick which helps some.

I'm not convinced that using more relievers per game is conducive to keeping a bullpen relatively rested. I see too many times where the AI will "waste" relievers only to have no one left when the game goes extra innings. Throwing 20 pitches over two games in two days as a reliever has a much different effect on your fatigue and availability for the third game of a series than throwing 20 pitches in relief in the first game and not even being called to warmup in the game the following day.

[Cryomaniac]

Originally Posted by NoOne

i think a better depth chart would help.

get rid of "secondary role"

simply have lists for each role, and which names you put there are in the pecking order for that role -- use frequency would still be nice for further stratification within that list... or a value system to rank them further within each list, since often times #3,4,5, isn't good enough if oyu only want 3 maximized and 4/5 minimized etc.

That would be cool. You could maybe even combine it into one slot system (eg it uses them in the order you put them in, but if you leave the #1 slot blank then it doesn't use anyone more often, or something).

[Brad K]

I really dislike current baseball. I will occasionally play to the point 11 pitchers are required but not beyond that.

[gosensgo101]

I think this is a huge problem. The AI entering a game with 6 exhausted relievers and 1 rested one is basically just forfeiting. And it compounds, as if the games are somewhat close a couple, either everyone will pitch a bit (leaving them all exhausted again the next day), or a couple relievers will be completely hung out to dry and be unavailable for 2-3 days afterwards.

There's also the issue of what's happening to get the AI in these situations, as managing my bullpen myself I never get to the point of everyone being exhausted, but most importantly the AI needs to rotate relievers up and down from Triple-A when the run into this problem.

Another issue is that reliever stamina changed a while back so that if a SP who's normal turn would be tomorrow (ie. If they pitched today they'd be on 3 days rest, instead of a full 4) is set as a RP he will be listed as exhausted. This both prevents the next day's starter from soaking up innings if needed and, if the Triple-A bullpen has been heavily taxed too, leaves only 1 Triple-A SP rested to come up and help the bullpen on any given day.

[NoOne]

there's got to be some sort of downshift when bullpen fatigue reaches a certain level...

not just added up, but how it's distributed... you can calculate how many innings you can handle over next 2days and 3 days and use that as qualifications for extending use or allowing for optimal use (loogies, 1or 2 outs ok etc).

a longer term view may be needed for any stretch longer than ~12 games? just spitballing. but it needs to be able to dynamically shift before the need is too great. that would queue callups too for the bottom 2 spots in bullpen... unless incredibly weak, most teams won't be shuffling established RP up and down out of respect. i'd like the game to remain that way too.

just playing and writing down some systematic rules to follow and implementing them could figure out a better loop of logic than what is currently in use. either too complicated or too simple, currenlty. i don't know..

even quality of bullpen to start should shift strategy a bit. you typically know when you have a good/bad pen and when you need to lean on them or when you can minimize their use with a strong rotation etc etc etc.

[Dyzalot]

Originally Posted by NoOne

there's got to be some sort of downshift when bullpen fatigue reaches a certain level...

not just added up, but how it's distributed... you can calculate how many innings you can handle over next 2days and 3 days and use that as qualifications for extending use or allowing for optimal use (loogies, 1or 2 outs ok etc).

a longer term view may be needed for any stretch longer than ~12 games? just spitballing. but it needs to be able to dynamically shift before the need is too great. that would queue callups too for the bottom 2 spots in bullpen... unless incredibly weak, most teams won't be shuffling established RP up and down out of respect. i'd like the game to remain that way too.

just playing and writing down some systematic rules to follow and implementing them could figure out a better loop of logic than what is currently in use. either too complicated or too simple, currenlty. i don't know..

even quality of bullpen to start should shift strategy a bit. you typically know when you have a good/bad pen and when you need to lean on them or when you can minimize their use with a strong rotation etc etc etc.

Problem is, if the AI calculates the amount of innings you can get from the bullpen but then doesn't use them optimally, you'll just run into the same issue. To me one of the main issues is how many times the AI is "warming up" relief pitchers. I'd be interested to see how well games played by relief pitchers for a simmed season matches up with games played by relief pitchers in real life. My guess is that for the modern game, OOTP's simmed season will produce higher numbers than real life. This would affect fatigue in ways that wouldn't be so obvious if you were only looking at innings pitched or pitches thrown.

[Situational_Lefty]

Originally Posted by Dyzalot

Problem is, if the AI calculates the amount of innings you can get from the bullpen but then doesn't use them optimally, you'll just run into the same issue. To me one of the main issues is how many times the AI is "warming up" relief pitchers. I'd be interested to see how well games played by relief pitchers for a simmed season matches up with games played by relief pitchers in real life. My guess is that for the modern game, OOTP's simmed season will produce higher numbers than real life. This would affect fatigue in ways that wouldn't be so obvious if you were only looking at innings pitched or pitches thrown.

I have noticed that in the MLB games I have simmed, games pitched by relievers are noticibly higher than the real life totals. I generally see several RP's pitch over 80 games per year. That is a very unusual total to see in real life. In fact, last year only about 10 relievers even pitched in as many as 70 games.

[NoOne]

Originally Posted by Dyzalot

Problem is, if the AI calculates the amount of innings you can get from the bullpen but then doesn't use them optimally, you'll just run into the same issue. To me one of the main issues is how many times the AI is "warming up" relief pitchers. I'd be interested to see how well games played by relief pitchers for a simmed season matches up with games played by relief pitchers in real life. My guess is that for the modern game, OOTP's simmed season will produce higher numbers than real life. This would affect fatigue in ways that wouldn't be so obvious if you were only looking at innings pitched or pitches thrown.

oh, i don't mean under current logical use... something in addition to it... an extra qualification based on these things .. if it exists in a similar way, it needs a re-work. simpler likely being better.

maybe just tweaking the 'exceptions' that occur, and making them less prominent. loogies used more strictly. situational stuff toned down just a tad.

i think that would obviously reduce GP for RP too.

but, there's a threshold that will work. at some point Innings-potential of bullpen (based on existing fatigue and next day off) can easily dictate when you need to use that RP for 1.1ip or 2ip instead of 1ip or situational stuff. it can also dictate when things are dire and you need to keep that RP out there even if he gets shelled.

maybe it needs 3 gears to work well, maybe it needs 5. trial and error with a little math to limit flailing around like a chicken with its head cut off.

that's basically what a human manager does... when they know things are a problem and they need to limit shuffling up and down from mil, they do more unusual and heartless things based on absolute need. then, it is a continuous spectrum inbetween extremes, but 3-5 "gears" would be enough for a video game, i'd think. each has it's own loop of logic, slightly different from the others etc (for those that have never written a program).

'gear' = ? this comes from context of game variables and thresholds above
case select: a, b, c (abc et al. being 'gears'... goes to proper one, then runs that logic. unused logic is not relevant to speed or anything else of game)

it doesn't have to be one mess of nested logic for an all-in-one answer. it just has to recognize which logic to use. even if you've never programmed before, you/we do these things already, even if we don't recognize it as such. think about how you handle bullpen and if it's really solid logic, submit some suggestions in the other forum for sure.

i don't play game to game anymore... it could be hammered out, though. simply be systematic and take note of rules you adhere to. when you are more conservative and when you are more free-wheeling.

[Dyzalot]

I just don't agree. For example, I already see loogies being used too exclusively. Nothing like seeing the AI bring in a lefty to face a lefty, give up a walk or hit then replace the lefty with a righty because a righty is up next despite the fact that two lefties follow him in the order. I believe this kind of stuff is a contributing factor to over usage of bullpens and trying to be even stricter about having a loogy only face lefties would seem to only exacerbate the problem.

[NoOne]

that's exactly what i mean.

well the gears would fix half that.. the other half needs to look ahead in lineup more than just next batter... 2/3? 3/4? 2/4? (2 of next 3 or 4 etc) where is the line drawn and how does the manager slider affect that line? that's an easy fix, most likely.

that's a matter of robustness and adding an extra qualification to check for that before acting. if 2 of 3 batters next up, leave in through righty etc... (obviously simplified, but that step is there somewhere to use a loogie, this merely needs to be injected into the qualifications of that behaviour or adjusted etc.) the sliders make it a bit complicated to the eye if you saw the code i bet. gotta have a spectrum of behaviour there.

the gears would recognize whether to even think about doing so... it would stop it from reaching the 'use loogie' qualification and therefore won't be used at all in some situations.

completely recovered (90-100 overlal fatigue recovers?) -- has access to it
75% -- maybe?

50% -- probably not and lower even less so of course... it won't even look to use loogies, in fact at the lower "oh sh$%" gears it will use loogies as it uses Righties. and more so at lower gears due to increased fatigue.

you're basically taking the exisitng logic (assuming it's all "1") and copying it 3-5 times and adjusting it for that context. reduces complexity of qualifications becasue they are different foreach context of too many factors to try to cross reference at once. each loop of logic is accessed approriately, so it even reduces number of qualifications per decisions, almost inevitably. the 1 check for 'case' would be added (gear 1 2 or 3? etc.. simple and not resourece intensive, guaranteed efficiency increases)

i used percentages above to communicate in a simpler way, but i'd go with some algorythm that adds up potential innings of use left in bullpen (as well as max per various factors like stamina and quality of talent) those variables are calculated before "play ball" for any game and after any RP use -- even during.

when it recognizes low innings avialable for tomorow, next day, the next day (as far out as necessary) it acts more appropriately. it would assuredly end and overfrequent use of 1-out use and and 3 fielders pitching. also reduce using loogies in borderline situations and like you mentioned a bit more forethought about lineup as far as bridign a righty to get to 2 more lefties.

[Situational_Lefty]

I was looking through the Stats and AI in the league settings and noticed that there were settings for "hook for starting pitcher" and "hook for relief pitcher". I noticed my test leagues were set at minus 1, but there is actually a default setting listed as one of the options.

I don't know if anyone has tried to adjust those settings to help with the bullpen management issue, but I wanted to point that out. This is actually the first time I even noticed those settings.

[Drstrangelove]

Yup, that's been changed. I am in a DH league, and my global AI is set to default (presumably zero modification), and I've never used any other setting. (Starters are set to very low stamina as default; in fact, all those setting are set at default for me.)

I use 12 pitchers and set the starters and relievers to have an average or quicker than average hook, (although they are nowhere near the maximum hook rate), and I've had no issues. Two of my relievers are bona fide starters who I set as alternates for long relief and will sometimes throw 30-50 pitches. This season I am on pace to have 4 relievers with between 72-77 appearances, with three others at 59, 57 and 46. The only time a non-pitcher was in a game was in an 18-inning game. Around the league, the AI high count for each team tops at around 89 appearances on a couple teams, but most teams are down around 75-78 or so. I'd say, adjusting my global settings to -1 would create issues for a few teams.

Certainly, the AI will have the same problem a human will, if the bullpen has no long relievers and the manager has a quick hook on all the pitchers. This is probably worse in a non-DH league. Adding another reliever is a solution, but changing the hook rate or adding long relievers would help as well. Sometimes a team has to have a pitcher take a beating if they are down by 5 runs in the 7th, rather than burning through 3 relievers. How the AI does it, I don't have a clue, but it's certainly going to be harder with a quicker hook.