AI Manager Issues

Getch · 10-13-2011, 04:14 PM

So with the new online game viewer I wrote from OU, all of a sudden a lot of online GM's are watching their teams play out games. And across all the leagues I am in, the sentiment is that the AI Manager is quite terrible. I'm sure my fellow GM's will chime in, but here is a few examples of terrible managing:

1. Double switches - The biggest gripe comes from the NL managers with how the game does double switches. They are seeing countless times where a pitcher will bat, only to immediately be taken out the next inning. Here is one example:

Game Log - St. Louis Cardinals vs Los Angeles Dodgers, Thursday, June 14th, 2040

Cisneros batted as the last out in the bottom of the 7th in a tie game, only to be replaced in the top of the 8th.

There are other complaints and I'll get more concrete examples from them.

2. There's been a lot of gripes about bullpen usage in general. However, it has mostly been limited to "why did he pitch XXXX in a tie game in the 9th", or "Why did he use my best reliever for only one batter?" I will try to get more concrete details and examples

3. Usage of intentional walks. I've noticed the AI giving out free passes when I am UP in a game. For instance:

Game Log - Bryan Bombers vs Portland Beavers, Friday, May 4th, 2029

In the bottom of the 8th, I am up 5-1, and the opponent leads off with a double. The next batter, a weak 6/3/5 hitter (on 1-10 scale) is intentionally walked. Why would anyone do this in this situation? You have 4 runs... if the man on second scores you are still in good shape. Giving up 2 or more runs starts to get things close. The intentionally walked batter ends up scoring and they ultimately tie the game.

Intentional walks are only used if giving up one run pretty much closes out the game. In this case, if the intentionally walked batter also scores, well... it doesn't really matter at that point, the game is out of hand (or over). But up a few runs, the LAST thing you want to do is give them more baserunners; Concede the run in the hopes of preventing a big inning.

4. Using pinch runners. Example:

Game Log - Tampa Bay Devil Rays vs Detroit Tigers, Wednesday, June 6th, 2040

Bottom 9, down one run, my backup catcher leads off with a walk. He has a speed of 2. I have my entire bench, including a 10/10 speed/steal guy and a few other speed guys, but he's not pinch run for. With 2 outs, one of my guys hits a double to the gap but he can only get to third, and I lose on the next batter.

This happened again here:

Game Log - Portland Beavers vs Evansville Otters, Tuesday, May 15th, 2029

bottom 9, down 1 run, lead off hitter with a speed of 1 gets on, but is never pinch run for. This is slightly less egregious as my bench was getting thin by this point, but there were 3 players left, including a 7/9 player.

A third time:

Game Log - Portland Beavers vs Guelph Royals, Friday, May 18th, 2029

Again, down 1 in the 9th (seems to happen a lot to me). In this case, the batter on first with a speed of 2 WAS pinch run for, but only after he was advanced to second. Why was he not pinch run for as soon as he got on base?

I'll ask the GM's to keep track of the gripes and post them here as they see them.

geoff · 10-14-2011, 07:49 AM

I'll admit, I'm one of the people who has been griping since seeing games played out via Getch's amazing Gamecast utility. I requested an NL team when I re-joined an online league recently, and the Commissioner was surprised because of the game's issues with pinch hitting/relief/double switches. Apparently the league has been unsuccessfully seeking changing the NL to use the DH to counter these game issues for quite some time.

I've certainly noticed issues with pitchers hitting with men on base, only to be immediately taken out the next inning. I cant cite specific examples, as Getch has done, but I'll watch for them and post them here when I next see them. Likewise, the reliever selection has been a head-scratcher. The AI seems to fall in love with certain pitchers and use them almost exclusively, regardless of placement in the bullpen "hierarchy" or rest requirements. Little nagging issues that may have seemed purely anecdotal are suddenly a lot more visible and real when you watch a number of games with the Gamecast utility. I'll try to keep my eye out for specific examples, as I said, but wanted to reply in support of Getch's excellent post and examples. I hope these issues will be looked into.

Markus Heinsohn · 10-14-2011, 08:23 AM

Hmmm... I checked your examples in the code, and all these situations are covered by the proper behavior. AI is pretty complex though, and sometimes certain factors contribute to a 'strange' decision under normal circumstances.

The pitcher hitting and then being replaced at the start of the next inning is pretty mysterious though. I just ran a test season, and did not have this happen once.

hluraven · 10-14-2011, 10:25 AM

I've let pitchers bat only to sub them out before pitching again dozens of times! But I would agree that the AI never should be as careless as me on this one.

Getch · 10-14-2011, 10:26 AM

Yeah, I understand the AI is complex; I wouldn't want to program it. And I fully expect a lot of the bullpen issues people have will have a defensible argument behind why the AI did what it did. Personally, I haven't had much issues with the bullpen, but that may be because my pens usually consist of 5 terrible pitchers so it doesn't matter who the AI chooses

.

However, I would be curious as to the logic behind the intentional walks and lack of pitch running. I really tried hard to come up with a reasoning behind the AI's decision in each case. But I cannot come up with any logic that would dictate an intentional walk that would bring the tying run to the plate or still somewhere on-deck, unless you're facing a juiced up Bonds maybe.

For the pinch running, I tried to see if there was any conditions that might prevent using a pinch runner. In one case, the AI would have to do a little positional juggling to cover all the positions with someone who had a rating at each position, but it could have been done pretty easily. All of these cases were when the team was down in the 9th, not an earlier inning (personally, I hate pinch running unless its do-or-die situations), so it's not as if the AI was worrying about losing a future at-bat.

Another possibility is saving the pinch runner for the winning run instead of the tying run, maybe? But in the first case, the next three hitters were all speed guys, so there would be no need to pinch run them at all.

Anyway, I'd be content if there was defensible logic for why the AI did what it did (even though I'd disagree with it, I can at least respect it).

Next sim is today and the GM's are planning on watching closely. I'll post the results.

David Watts · 10-14-2011, 11:01 AM

Quote:

Originally Posted by Getch

Yeah, I understand the AI is complex; I wouldn't want to program it. And I fully expect a lot of the bullpen issues people have will have a defensible argument behind why the AI did what it did. Personally, I haven't had much issues with the bullpen, but that may be because my pens usually consist of 5 terrible pitchers so it doesn't matter who the AI chooses

.

However, I would be curious as to the logic behind the intentional walks and lack of pitch running. I really tried hard to come up with a reasoning behind the AI's decision in each case. But I cannot come up with any logic that would dictate an intentional walk that would bring the tying run to the plate or still somewhere on-deck, unless you're facing a juiced up Bonds maybe.

For the pinch running, I tried to see if there was any conditions that might prevent using a pinch runner. In one case, the AI would have to do a little positional juggling to cover all the positions with someone who had a rating at each position, but it could have been done pretty easily. All of these cases were when the team was down in the 9th, not an earlier inning (personally, I hate pinch running unless its do-or-die situations), so it's not as if the AI was worrying about losing a future at-bat.

Another possibility is saving the pinch runner for the winning run instead of the tying run, maybe? But in the first case, the next three hitters were all speed guys, so there would be no need to pinch run them at all.

Anyway, I'd be content if there was defensible logic for why the AI did what it did (even though I'd disagree with it, I can at least respect it).

Next sim is today and the GM's are planning on watching closely. I'll post the results.

Hmm, I think I read somewhere that Markus let Ron Washington program the pinch running logic for OOTP12, much to the chagrin of Tim McCarver

geoff · 10-14-2011, 12:08 PM

Found an example pretty quickly, actually. Pasted from a game log.

Top of the 8th - Florida Marlins batting - Pitching for Washington Nationals : LHP Tony Fernández
Now at 2B: Eric Foster
Now in RF: José Morales
Batting: LHB Raúl Hernández 0-0: Called Strike
0-1: Ball
1-1: Called Strike
1-2: Strikes out swinging
Batting: RHB Kazu Kurota 0-0: Called Strike
0-1: DOUBLE (Line Drive, 7LM)
Batting: LHB Ken Mahoney 0-0: Bunted foul
0-1: Ball
1-1: Sac Bunt - play at first, batter OUT! 1-4
Kazu Kurota to third
Batting: RHB Michael Barnes 0-0: Swinging Strike
0-1: Called Strike
0-2: Fly out, F7 (Flyball, 7D)
Top of the 8th over - 0 run(s), 1 hit(s), 0 error(s), 1 left on base; Florida 1 - Washington 0

Bottom of the 8th - Washington Nationals batting - Pitching for Florida Marlins : LHP Ken Mahoney
Batting: LHB Tony Fernández 0-0: Ground out 4-3 (Groundball, 4M)
Pitching: LHP Mauro Fernández

To summarize...Kurota hits a double with 1 out in the 8th inning of a game we're losing 1-0. Mahoney, my starting pitcher, bats next and bunts him to 3rd, getting the 2nd out in the process. Barnes comes up next and flies out to end the inning. Bottom of the 8th, Mahoney is promptly pulled for a reliever without throwing a single pitch.

I may be wrong though, I'm no Tony LaRussa, but he bunts the guy over when Kurota is effectively already in scoring position and has 8/8/6 baserunning. It seems to me that using a PH in that situation makes sense in virtually every instance.

bp_ · 10-14-2011, 05:55 PM

Quote:

Originally Posted by geoff

I may be wrong though, I'm no Tony LaRussa, but he bunts the guy over when Kurota is effectively already in scoring position and has 8/8/6 baserunning. It seems to me that using a PH in that situation makes sense in virtually every instance.

I would agree with this. Seems very odd to see this happen.

Getch · 10-17-2011, 11:54 AM

Two more issues, this time with Pinch Hitting:

In this game:

Game Log - Detroit Tigers vs Minnesota Twins, Friday, June 22nd, 2040

I am down 2 runs in the ninth. I get two batters on with one out, and my worst hitting SS McDonald, a righty who is batting less than .200, is left in to hit against their righty closer. I have two lefties with 10 power versus righties sitting on the bench. Neither one is called in to hit and he hits into a game-ending DP.

Sadly, the same thing happens next game:

Game Log - Detroit Tigers vs Minnesota Twins, Saturday, June 23rd, 2040

Down 2 in the ninth, he's left in to hit, gets the second out of the inning, and I end up leaving the bases loaded. In this game a few bench players were already used in the game, but there was enough to get this guy pinch hit for.

MLHFFYP · 10-18-2011, 05:53 PM

The manager AI is something I have had problems with as well.

When I commented on it someone suggested that I adjust manager settings to reflect more reality.

So I did a ton of research and created a few dozen manager profiles that would be based on historical managers. The manager uses the in game strategy settings so I adjusted the sliders accordingly and ended up with 35 historical managers. I also made 15 fictional managers as well for variety sake. I also adjusted the overal roster strategies as well to reflect the real life GM tendencies. So Joe McCarthy has in game strategies reflecting his years as the manager of the 1930's New York Yankees while the overall roster strategies reflect his general manager, Ed Barrow.

This was a project submitted for OOTP 12 but wasn't incorporated.

I am currently playing a random debut league and all of the managers for the teams are based on historical managers. The last two World Series were won by John McGraw's Texas Rangers while Rube Foster's Houston Astros won before that.

I found that using these kinds of settings has a huge impact on the game. It forces the AI to replicate what a manger might do in a given situation based on his real life tendencies. So a Connie Mack team will not be managed the same as a team skippered by Cap Anson. Likewise, having overall roster strategy settings that mirror the GM keeps the team looking for players (via trade or draft) that will fit into the managers system. The AI will still do some head scratching things on occasion which are, I think, due to a randomness factor but overall it stays true to the profile.

My overall point is that the manager AI, when modified, will present a better challenge than what comes with the game.

10-13-2011, 04:14 PM	#1
Getch All Star Reserve Join Date: Jan 2006 Posts: 868	AI Manager Issues So with the new online game viewer I wrote from OU, all of a sudden a lot of online GM's are watching their teams play out games. And across all the leagues I am in, the sentiment is that the AI Manager is quite terrible. I'm sure my fellow GM's will chime in, but here is a few examples of terrible managing: 1. Double switches - The biggest gripe comes from the NL managers with how the game does double switches. They are seeing countless times where a pitcher will bat, only to immediately be taken out the next inning. Here is one example: Game Log - St. Louis Cardinals vs Los Angeles Dodgers, Thursday, June 14th, 2040 Cisneros batted as the last out in the bottom of the 7th in a tie game, only to be replaced in the top of the 8th. There are other complaints and I'll get more concrete examples from them. 2. There's been a lot of gripes about bullpen usage in general. However, it has mostly been limited to "why did he pitch XXXX in a tie game in the 9th", or "Why did he use my best reliever for only one batter?" I will try to get more concrete details and examples 3. Usage of intentional walks. I've noticed the AI giving out free passes when I am UP in a game. For instance: Game Log - Bryan Bombers vs Portland Beavers, Friday, May 4th, 2029 In the bottom of the 8th, I am up 5-1, and the opponent leads off with a double. The next batter, a weak 6/3/5 hitter (on 1-10 scale) is intentionally walked. Why would anyone do this in this situation? You have 4 runs... if the man on second scores you are still in good shape. Giving up 2 or more runs starts to get things close. The intentionally walked batter ends up scoring and they ultimately tie the game. Intentional walks are only used if giving up one run pretty much closes out the game. In this case, if the intentionally walked batter also scores, well... it doesn't really matter at that point, the game is out of hand (or over). But up a few runs, the LAST thing you want to do is give them more baserunners; Concede the run in the hopes of preventing a big inning. 4. Using pinch runners. Example: Game Log - Tampa Bay Devil Rays vs Detroit Tigers, Wednesday, June 6th, 2040 Bottom 9, down one run, my backup catcher leads off with a walk. He has a speed of 2. I have my entire bench, including a 10/10 speed/steal guy and a few other speed guys, but he's not pinch run for. With 2 outs, one of my guys hits a double to the gap but he can only get to third, and I lose on the next batter. This happened again here: Game Log - Portland Beavers vs Evansville Otters, Tuesday, May 15th, 2029 bottom 9, down 1 run, lead off hitter with a speed of 1 gets on, but is never pinch run for. This is slightly less egregious as my bench was getting thin by this point, but there were 3 players left, including a 7/9 player. A third time: Game Log - Portland Beavers vs Guelph Royals, Friday, May 18th, 2029 Again, down 1 in the 9th (seems to happen a lot to me). In this case, the batter on first with a speed of 2 WAS pinch run for, but only after he was advanced to second. Why was he not pinch run for as soon as he got on base? I'll ask the GM's to keep track of the gripes and post them here as they see them. __________________ Get the OOTP Online Utilities for online leagues! Includes Gamecast, Development, Live Sims, Voting and more. Check here for more details

10-14-2011, 08:23 AM	#3
Markus Heinsohn Developer OOTP Join Date: Dec 2001 Location: Germany Posts: 24,805	Hmmm... I checked your examples in the code, and all these situations are covered by the proper behavior. AI is pretty complex though, and sometimes certain factors contribute to a 'strange' decision under normal circumstances. The pitcher hitting and then being replaced at the start of the next inning is pretty mysterious though. I just ran a test season, and did not have this happen once. __________________ Buy Out of the Park Baseball 25 now! Download OOTP Go 25 for iOS Download OOTP Go 25 for Android

10-14-2011, 10:26 AM	#5
Getch All Star Reserve Join Date: Jan 2006 Posts: 868	Yeah, I understand the AI is complex; I wouldn't want to program it. And I fully expect a lot of the bullpen issues people have will have a defensible argument behind why the AI did what it did. Personally, I haven't had much issues with the bullpen, but that may be because my pens usually consist of 5 terrible pitchers so it doesn't matter who the AI chooses . However, I would be curious as to the logic behind the intentional walks and lack of pitch running. I really tried hard to come up with a reasoning behind the AI's decision in each case. But I cannot come up with any logic that would dictate an intentional walk that would bring the tying run to the plate or still somewhere on-deck, unless you're facing a juiced up Bonds maybe. For the pinch running, I tried to see if there was any conditions that might prevent using a pinch runner. In one case, the AI would have to do a little positional juggling to cover all the positions with someone who had a rating at each position, but it could have been done pretty easily. All of these cases were when the team was down in the 9th, not an earlier inning (personally, I hate pinch running unless its do-or-die situations), so it's not as if the AI was worrying about losing a future at-bat. Another possibility is saving the pinch runner for the winning run instead of the tying run, maybe? But in the first case, the next three hitters were all speed guys, so there would be no need to pinch run them at all. Anyway, I'd be content if there was defensible logic for why the AI did what it did (even though I'd disagree with it, I can at least respect it). Next sim is today and the GM's are planning on watching closely. I'll post the results. __________________ Get the OOTP Online Utilities for online leagues! Includes Gamecast, Development, Live Sims, Voting and more. Check here for more details

10-14-2011, 12:08 PM	#7
geoff Minors (Single A) Join Date: Jun 2003 Location: Ft Hood, TX Posts: 85	Found an example pretty quickly, actually. Pasted from a game log. Top of the 8th - Florida Marlins batting - Pitching for Washington Nationals : LHP Tony Fernández Now at 2B: Eric Foster Now in RF: José Morales Batting: LHB Raúl Hernández 0-0: Called Strike 0-1: Ball 1-1: Called Strike 1-2: Strikes out swinging Batting: RHB Kazu Kurota 0-0: Called Strike 0-1: DOUBLE (Line Drive, 7LM) Batting: LHB Ken Mahoney 0-0: Bunted foul 0-1: Ball 1-1: Sac Bunt - play at first, batter OUT! 1-4 Kazu Kurota to third Batting: RHB Michael Barnes 0-0: Swinging Strike 0-1: Called Strike 0-2: Fly out, F7 (Flyball, 7D) Top of the 8th over - 0 run(s), 1 hit(s), 0 error(s), 1 left on base; Florida 1 - Washington 0 Bottom of the 8th - Washington Nationals batting - Pitching for Florida Marlins : LHP Ken Mahoney Batting: LHB Tony Fernández 0-0: Ground out 4-3 (Groundball, 4M) Pitching: LHP Mauro Fernández To summarize...Kurota hits a double with 1 out in the 8th inning of a game we're losing 1-0. Mahoney, my starting pitcher, bats next and bunts him to 3rd, getting the 2nd out in the process. Barnes comes up next and flies out to end the inning. Bottom of the 8th, Mahoney is promptly pulled for a reliever without throwing a single pitch. I may be wrong though, I'm no Tony LaRussa, but he bunts the guy over when Kurota is effectively already in scoring position and has 8/8/6 baserunning. It seems to me that using a PH in that situation makes sense in virtually every instance. Last edited by geoff; 10-14-2011 at 12:11 PM.

10-17-2011, 11:54 AM	#9
Getch All Star Reserve Join Date: Jan 2006 Posts: 868	Two more issues, this time with Pinch Hitting: In this game: Game Log - Detroit Tigers vs Minnesota Twins, Friday, June 22nd, 2040 I am down 2 runs in the ninth. I get two batters on with one out, and my worst hitting SS McDonald, a righty who is batting less than .200, is left in to hit against their righty closer. I have two lefties with 10 power versus righties sitting on the bench. Neither one is called in to hit and he hits into a game-ending DP. Sadly, the same thing happens next game: Game Log - Detroit Tigers vs Minnesota Twins, Saturday, June 23rd, 2040 Down 2 in the ninth, he's left in to hit, gets the second out of the inning, and I end up leaving the bases loaded. In this game a few bench players were already used in the game, but there was enough to get this guy pinch hit for. __________________ Get the OOTP Online Utilities for online leagues! Includes Gamecast, Development, Live Sims, Voting and more. Check here for more details

10-14-2011, 07:49 AM	#2
geoff Minors (Single A) Join Date: Jun 2003 Location: Ft Hood, TX Posts: 85	I'll admit, I'm one of the people who has been griping since seeing games played out via Getch's amazing Gamecast utility. I requested an NL team when I re-joined an online league recently, and the Commissioner was surprised because of the game's issues with pinch hitting/relief/double switches. Apparently the league has been unsuccessfully seeking changing the NL to use the DH to counter these game issues for quite some time. I've certainly noticed issues with pitchers hitting with men on base, only to be immediately taken out the next inning. I cant cite specific examples, as Getch has done, but I'll watch for them and post them here when I next see them. Likewise, the reliever selection has been a head-scratcher. The AI seems to fall in love with certain pitchers and use them almost exclusively, regardless of placement in the bullpen "hierarchy" or rest requirements. Little nagging issues that may have seemed purely anecdotal are suddenly a lot more visible and real when you watch a number of games with the Gamecast utility. I'll try to keep my eye out for specific examples, as I said, but wanted to reply in support of Getch's excellent post and examples. I hope these issues will be looked into.

10-14-2011, 10:25 AM	#4
hluraven Major Leagues Join Date: Jul 2010 Location: London, England Posts: 355	I've let pitchers bat only to sub them out before pitching again dozens of times! But I would agree that the AI never should be as careless as me on this one.

10-18-2011, 05:53 PM	#10
MLHFFYP Minors (Double A) Join Date: Feb 2010 Posts: 148	Manager AI The manager AI is something I have had problems with as well. When I commented on it someone suggested that I adjust manager settings to reflect more reality. So I did a ton of research and created a few dozen manager profiles that would be based on historical managers. The manager uses the in game strategy settings so I adjusted the sliders accordingly and ended up with 35 historical managers. I also made 15 fictional managers as well for variety sake. I also adjusted the overal roster strategies as well to reflect the real life GM tendencies. So Joe McCarthy has in game strategies reflecting his years as the manager of the 1930's New York Yankees while the overall roster strategies reflect his general manager, Ed Barrow. This was a project submitted for OOTP 12 but wasn't incorporated. I am currently playing a random debut league and all of the managers for the teams are based on historical managers. The last two World Series were won by John McGraw's Texas Rangers while Rube Foster's Houston Astros won before that. I found that using these kinds of settings has a huge impact on the game. It forces the AI to replicate what a manger might do in a given situation based on his real life tendencies. So a Connie Mack team will not be managed the same as a team skippered by Cap Anson. Likewise, having overall roster strategy settings that mirror the GM keeps the team looking for players (via trade or draft) that will fit into the managers system. The AI will still do some head scratching things on occasion which are, I think, due to a randomness factor but overall it stays true to the profile. My overall point is that the manager AI, when modified, will present a better challenge than what comes with the game.