AI Lineups & Ordering: Anecdotes, Analysis, Suggestions

Cruzatte · 06-21-2011, 06:02 PM

Overview
Basically, there are two keys to this whole thing: (1) how the AI picks a lineup; and (2) how the AI sets a batting order within that lineup.

The problem with "Traditional Lineup Selection" isn't the order, but the lineup selection itself; it uses the most nontraditional aspects of "sabermetrics" (no Versus consistency in the lineup; borderline-illogical sit/start decisions based on badly-prioritized split ratings).

The problem with "Sabermetric Lineup Selection" isn't the lineup itself, but the batting order; it has some of the worst aspects of "tradition" (basically, it doesn't do a good job following widely-accepted statistical principles of batting order construction).

TRADITIONAL.
The "traditional" AI selection isn't traditional. The problem (in terms of "tradition") is the lack of unity between the Versus lineups. And the problem in terms of "this making any sense at all" is that it sits better players for mediocre players based on bizarre distinctions.

E.g. a star RF (74/80 overall) sits against every RHP, in favor of a 58-overall RF who has +5/20 contact vs. RHP, but -4 power and -4 RF fielding. (Against LHPs, the star is substantially better in all respects; both are 33 years old and playing to their max potential.)
(Also note: I'm trying not to be myopic about the "overall" metric... I think I'm fairly judging these guys by their stats & split ratings, not just "HOW COULD YOU START MY 4.5 STAR GUY OVER MY 5 STAR GUY?!?!")

So, "Traditional" lineup A.I. needs:
(1) Realistic sit/start decisions based on Ratings. (No MLB manager would make the above-described lineup decision... and a good number of those managers are far more SABR than old-school. Plus, the 74-overall guy makes way more money.)
(2) Realistic consistency between the Versus lineups. (I.e., prioritize consistency between lineups, and allow platoons only for major Ratings differences.)

I realize, Mr. The Wolf, that this might require re-imagining the Versus lineups in such a way that the AI could recognize that it's making a "platoon," as opposed to making discrete & unrelated decisions. But this is absolutely necessary to make the lineups "traditional." And to make them reflect, you know, real baseball.

WORKAROUNDS:
(1) Set your own lineups--a potentially monumental task, whether you're a simmer or pitch-by-pitch'er.
(2) Force-start a player in Player Strategy to avoid the platoon.
(3) Crank up "Pinch-Hit for Position Players." At least then, your 74-overall guy will play as soon as a lefty comes in.

SABERMETRIC.
The "sabermetric" AI selection just reshuffles the batting order of the "traditional" lineup (same 9 guys, including bizarre platoons, but with a different batting order). Maybe there are instances when it would actually make a distinct lineup, but I've never seen it happen.

If we assume that Traditional AI lineups are basically Sabermetric, since they're still splits-obsessed, then the Sabermetric AI's reshuffling should be based on a substantially different ordering algorithm.

But it's not. The reshuffling does prioritize splits, but not pursuant to any truly "Sabermetric" logic. The overwhelming lineup scheme still appears to be traditional (e.g., the low-OBP speedster is still in the leadoff spot). In other words, it doesn't really comport with The Book.*

When I use Sabermetric Lineups, I'm much more willing to allow the AI to make a facially bizarre sit/start decision (like 74 vs. 58, above). So if we're confident that OOTP makes a rational decision here, then maybe it doesn't need tweaking for Sabermetric mode.

Anyway, point is, the stats-based batting order construction leaves a good deal to be desired.

FINAL SUGGESTION:
In the perfect world (the kind of world that maybe somebody could create using a computer), we'd have two distinct systems...
One that makes lineup selections: do I want to start my best 9 players without respect to their splits ratings, unless there's a substantial platoon advantage (Traditional), or do I want to start the best 9 depending on the opposing pitcher's handedness (Saber).
And one that makes order selections: do I want to use an old-school, speed-at-the-top order (Traditional) or go by The Book (Saber).

The bigger problem from a gameplay standpoint, in my view, is the first one: by the time the AI is making "traditional" ORDER decisions, it's forced to work with a nontraditional (i.e. heavily splits-based) LINEUP.

* The Book's construction, just for the record:

(1) top-3 hitter w/ high OBP
(2) top-3 hitter
(3) 5th-best (or 4th-best, if high HR power)
(4) top-3 hitter w/ high SLG
(5) 4th-best (or 5th, if 4th-best has high HR power)
(6) 6th-best (or lower, if 7/8/9 has SB speed)
(7) 7th-best
(8) 8th-best (or 9th-best, if 9th is seriously bad, i.e. pitcher)
(9) 9th-best (again, unless he's seriously bad, then put #8 here)

Simmo13 · 06-22-2011, 06:16 AM

Quote:

Originally Posted by Cruzatte

Overview
Basically, there are two keys to this whole thing: (1) how the AI picks a lineup; and (2) how the AI sets a batting order within that lineup.

The problem with "Traditional Lineup Selection" isn't the order, but the lineup selection itself; it uses the most nontraditional aspects of "sabermetrics" (no Versus consistency in the lineup; borderline-illogical sit/start decisions based on badly-prioritized split ratings).

The problem with "Sabermetric Lineup Selection" isn't the lineup itself, but the batting order; it has some of the worst aspects of "tradition" (basically, it doesn't do a good job following widely-accepted statistical principles of batting order construction).

TRADITIONAL.
The "traditional" AI selection isn't traditional. The problem (in terms of "tradition") is the lack of unity between the Versus lineups. And the problem in terms of "this making any sense at all" is that it sits better players for mediocre players based on bizarre distinctions.

E.g. a star RF (74/80 overall) sits against every RHP, in favor of a 58-overall RF who has +5/20 contact vs. RHP, but -4 power and -4 RF fielding. (Against LHPs, the star is substantially better in all respects; both are 33 years old and playing to their max potential.)
(Also note: I'm trying not to be myopic about the "overall" metric... I think I'm fairly judging these guys by their stats & split ratings, not just "HOW COULD YOU START MY 4.5 STAR GUY OVER MY 5 STAR GUY?!?!")

So, "Traditional" lineup A.I. needs:
(1) Realistic sit/start decisions based on Ratings. (No MLB manager would make the above-described lineup decision... and a good number of those managers are far more SABR than old-school. Plus, the 74-overall guy makes way more money.)
(2) Realistic consistency between the Versus lineups. (I.e., prioritize consistency between lineups, and allow platoons only for major Ratings differences.)

I realize, Mr. The Wolf, that this might require re-imagining the Versus lineups in such a way that the AI could recognize that it's making a "platoon," as opposed to making discrete & unrelated decisions. But this is absolutely necessary to make the lineups "traditional." And to make them reflect, you know, real baseball.

WORKAROUNDS:
(1) Set your own lineups--a potentially monumental task, whether you're a simmer or pitch-by-pitch'er.
(2) Force-start a player in Player Strategy to avoid the platoon.
(3) Crank up "Pinch-Hit for Position Players." At least then, your 74-overall guy will play as soon as a lefty comes in.

SABERMETRIC.
The "sabermetric" AI selection just reshuffles the batting order of the "traditional" lineup (same 9 guys, including bizarre platoons, but with a different batting order). Maybe there are instances when it would actually make a distinct lineup, but I've never seen it happen.

If we assume that Traditional AI lineups are basically Sabermetric, since they're still splits-obsessed, then the Sabermetric AI's reshuffling should be based on a substantially different ordering algorithm.

But it's not. The reshuffling does prioritize splits, but not pursuant to any truly "Sabermetric" logic. The overwhelming lineup scheme still appears to be traditional (e.g., the low-OBP speedster is still in the leadoff spot). In other words, it doesn't really comport with The Book.*

When I use Sabermetric Lineups, I'm much more willing to allow the AI to make a facially bizarre sit/start decision (like 74 vs. 58, above). So if we're confident that OOTP makes a rational decision here, then maybe it doesn't need tweaking for Sabermetric mode.

Anyway, point is, the stats-based batting order construction leaves a good deal to be desired.

FINAL SUGGESTION:
In the perfect world (the kind of world that maybe somebody could create using a computer), we'd have two distinct systems...
One that makes lineup selections: do I want to start my best 9 players without respect to their splits ratings, unless there's a substantial platoon advantage (Traditional), or do I want to start the best 9 depending on the opposing pitcher's handedness (Saber).
And one that makes order selections: do I want to use an old-school, speed-at-the-top order (Traditional) or go by The Book (Saber).

The bigger problem from a gameplay standpoint, in my view, is the first one: by the time the AI is making "traditional" ORDER decisions, it's forced to work with a nontraditional (i.e. heavily splits-based) LINEUP.

* The Book's construction, just for the record:

(1) top-3 hitter w/ high OBP
(2) top-3 hitter
(3) 5th-best (or 4th-best, if high HR power)
(4) top-3 hitter w/ high SLG
(5) 4th-best (or 5th, if 4th-best has high HR power)
(6) 6th-best (or lower, if 7/8/9 has SB speed)
(7) 7th-best
(8) 8th-best (or 9th-best, if 9th is seriously bad, i.e. pitcher)
(9) 9th-best (again, unless he's seriously bad, then put #8 here)

Would be interesting to get Markus' views on this. I use "traditional" line-up setting and tend to agree with this post.

actionjackson · 06-22-2011, 02:49 PM

bump. This interests me too. I don't understand calling it Sabermetric when there isn't much use of the latest sabermetric research in the algorithm.

Cruzatte · 06-22-2011, 08:17 PM

I realized after I posted that, with respect to the second part of the system I suggested, you kinda CAN differentiate AI order selection from lineup selection, by asking the manager to generate a lineup based on a depth chart that you set yourself.

But that doesn't actually change the analysis and problems, since what we're after is having the AI set up a depth chart (that is, select a lineup) that makes sense.

Bears repeating: the AI could set the batting order in a perfectly traditional way (I think it does a decent job at this), or in a perfectly sabermetric way (this it does not so well, as discussed above). But those functions are secondary to the AI making good lineup decisions.

06-21-2011, 06:02 PM	#1
Cruzatte Minors (Rookie Ball) Join Date: Jun 2011 Location: North Coast Posts: 31	AI Lineups & Ordering: Anecdotes, Analysis, Suggestions Overview Basically, there are two keys to this whole thing: (1) how the AI picks a lineup; and (2) how the AI sets a batting order within that lineup. The problem with "Traditional Lineup Selection" isn't the order, but the lineup selection itself; it uses the most nontraditional aspects of "sabermetrics" (no Versus consistency in the lineup; borderline-illogical sit/start decisions based on badly-prioritized split ratings). The problem with "Sabermetric Lineup Selection" isn't the lineup itself, but the batting order; it has some of the worst aspects of "tradition" (basically, it doesn't do a good job following widely-accepted statistical principles of batting order construction). TRADITIONAL. The "traditional" AI selection isn't traditional. The problem (in terms of "tradition") is the lack of unity between the Versus lineups. And the problem in terms of "this making any sense at all" is that it sits better players for mediocre players based on bizarre distinctions. E.g. a star RF (74/80 overall) sits against every RHP, in favor of a 58-overall RF who has +5/20 contact vs. RHP, but -4 power and -4 RF fielding. (Against LHPs, the star is substantially better in all respects; both are 33 years old and playing to their max potential.) (Also note: I'm trying not to be myopic about the "overall" metric... I think I'm fairly judging these guys by their stats & split ratings, not just "HOW COULD YOU START MY 4.5 STAR GUY OVER MY 5 STAR GUY?!?!") So, "Traditional" lineup A.I. needs: (1) Realistic sit/start decisions based on Ratings. (No MLB manager would make the above-described lineup decision... and a good number of those managers are far more SABR than old-school. Plus, the 74-overall guy makes way more money.) (2) Realistic consistency between the Versus lineups. (I.e., prioritize consistency between lineups, and allow platoons only for major Ratings differences.) I realize, Mr. The Wolf, that this might require re-imagining the Versus lineups in such a way that the AI could recognize that it's making a "platoon," as opposed to making discrete & unrelated decisions. But this is absolutely necessary to make the lineups "traditional." And to make them reflect, you know, real baseball. WORKAROUNDS: (1) Set your own lineups--a potentially monumental task, whether you're a simmer or pitch-by-pitch'er. (2) Force-start a player in Player Strategy to avoid the platoon. (3) Crank up "Pinch-Hit for Position Players." At least then, your 74-overall guy will play as soon as a lefty comes in. SABERMETRIC. The "sabermetric" AI selection just reshuffles the batting order of the "traditional" lineup (same 9 guys, including bizarre platoons, but with a different batting order). Maybe there are instances when it would actually make a distinct lineup, but I've never seen it happen. If we assume that Traditional AI lineups are basically Sabermetric, since they're still splits-obsessed, then the Sabermetric AI's reshuffling should be based on a substantially different ordering algorithm. But it's not. The reshuffling does prioritize splits, but not pursuant to any truly "Sabermetric" logic. The overwhelming lineup scheme still appears to be traditional (e.g., the low-OBP speedster is still in the leadoff spot). In other words, it doesn't really comport with The Book.* When I use Sabermetric Lineups, I'm much more willing to allow the AI to make a facially bizarre sit/start decision (like 74 vs. 58, above). So if we're confident that OOTP makes a rational decision here, then maybe it doesn't need tweaking for Sabermetric mode. Anyway, point is, the stats-based batting order construction leaves a good deal to be desired. FINAL SUGGESTION: In the perfect world (the kind of world that maybe somebody could create using a computer), we'd have two distinct systems... One that makes lineup selections: do I want to start my best 9 players without respect to their splits ratings, unless there's a substantial platoon advantage (Traditional), or do I want to start the best 9 depending on the opposing pitcher's handedness (Saber). And one that makes order selections: do I want to use an old-school, speed-at-the-top order (Traditional) or go by The Book (Saber). The bigger problem from a gameplay standpoint, in my view, is the first one: by the time the AI is making "traditional" ORDER decisions, it's forced to work with a nontraditional (i.e. heavily splits-based) LINEUP. * The Book's construction, just for the record: (1) top-3 hitter w/ high OBP (2) top-3 hitter (3) 5th-best (or 4th-best, if high HR power) (4) top-3 hitter w/ high SLG (5) 4th-best (or 5th, if 4th-best has high HR power) (6) 6th-best (or lower, if 7/8/9 has SB speed) (7) 7th-best (8) 8th-best (or 9th-best, if 9th is seriously bad, i.e. pitcher) (9) 9th-best (again, unless he's seriously bad, then put #8 here)

06-22-2011, 02:49 PM	#3
actionjackson Hall Of Famer Join Date: Apr 2009 Location: Toronto, ON Posts: 6,179	bump. This interests me too. I don't understand calling it Sabermetric when there isn't much use of the latest sabermetric research in the algorithm.

06-22-2011, 08:17 PM	#4
Cruzatte Minors (Rookie Ball) Join Date: Jun 2011 Location: North Coast Posts: 31	I realized after I posted that, with respect to the second part of the system I suggested, you kinda CAN differentiate AI order selection from lineup selection, by asking the manager to generate a lineup based on a depth chart that you set yourself. But that doesn't actually change the analysis and problems, since what we're after is having the AI set up a depth chart (that is, select a lineup) that makes sense. Bears repeating: the AI could set the batting order in a perfectly traditional way (I think it does a decent job at this), or in a perfectly sabermetric way (this it does not so well, as discussed above). But those functions are secondary to the AI making good lineup decisions.