A few AI ingame issues hopefully addressed by 1.03

markprior22 · 08-28-2006, 09:10 AM

1) Too many intentional walks by the AI, usually too early in the game.

2) Way too many steal attempts of 3B. These often happen in bad situations too such as tying run on 2B with 2 outs or no outs, etc.

3) Too much pinch hitting for position players even with the setup changed to very rarely.

These issues have all been mentioned before but I just want to make sure they aren't forgotten.

Markus Heinsohn · 08-28-2006, 09:15 AM

1) Not true, in my tests the league totals for IBB were usually in line with real life numbers.
2) Not true either, the frequencies match the ones in my retrosheet play-by-play analysis.
3) That just got tweaked.

kq76 · 08-28-2006, 09:31 AM

I wonder, are the results different depending on whether you play or sim the games out? I'm guessing Markus compares sim results to real life data whereas people are noticing play results.

andymac · 08-28-2006, 09:44 AM

Quote:

Originally Posted by kq76

I wonder, are the results different depending on whether you play or sim the games out? I'm guessing Markus compares sim results to real life data whereas people are noticing play results.

I'm pretty sure the sim and play engines are the same.

Markus Heinsohn · 08-28-2006, 09:48 AM

Quote:

Originally Posted by andymac

I'm pretty sure the sim and play engines are the same.

They are.

These observations are really meaningless, all that counts are test results in long term simulations, not impressions from playing out a few games.

kq76 · 08-28-2006, 09:58 AM

Well, if these things really bother someone then they should probably run their own tests.

Another possibility is maybe Markus is comparing results to a different time frame than others are. For example, maybe people are more comparing to what they've seen in real life over the last 10 years and maybe Markus is more comparing to what has happened over the last 40 years. Small things like this might change over time.

RonCo · 08-28-2006, 10:00 AM

Sample size is always important. Statistically, OOTP 2006 has been a soild improvement in most areas I've spent time to study with a prudent sample size...thoough I haven't studied steal of third opportunities. Hmmm...I'll have to think about how to even get the data to say for sure if this is out of whack or not. I can't think of a quick way to check it, anyway.

Questdog · 08-28-2006, 10:18 AM

Quote:

Originally Posted by Markus Heinsohn

1) Not true, in my tests the league totals for IBB were usually in line with real life numbers.
2) Not true either, the frequencies match the ones in my retrosheet play-by-play analysis.
3) That just got tweaked.

I believe the overall frequency of Intentional Walks is in line with norms, but the situations in which they are issued are not. Rarely do you see the batter walked with 2 outs to get a favorable match up as happens in real life; i.e. 8th place hitter walked to get to the pitcher or a righty walked to get to a lefty. The AI seems to ALWAYS walk a batter with 1st open and less than two out to set up a double play, regardless of whether this puts the winning run on base or at the plate.

I believe the same is true of the stolen base totals. Overall, they can be normalized, but runners are much more apt to try for third than they are for second, which is the exact opposite of real life. (A steal of third, while statistically easier to accomplish, holds a much higher failure cost). Especially, with two outs, runners should be much less ready to take off for third.

markprior22 · 08-28-2006, 11:14 AM

I guess I should have been clearer...I don't have a problem with the amount of intentional walks...it is the situations in which they occur. Same with steal attempts of third. I'm not debating the overall frequencies but some of the situations are totally absurd.

In general, overall statistical totals should be tested by long term stats. But, the only way to see the issues I'm referring to is by playing out games (because the quantity isn't being questioned as much as the situation). I've played out probably 90+ games since 1.02 has been out. I think that's plenty to take note of situations where intentional walks are being issued and attempts at stealing 3B are occurring. Again, I don't have a problem with the overall totals...it is the situations which are troubling.

Charlie Hough · 08-28-2006, 11:32 AM

Quote:

Originally Posted by Markus Heinsohn

1) Not true, in my tests the league totals for IBB were usually in line with real life numbers.

Yes, but it's been established that league totals are artificially maintained. For example, if most players in the database don't have a HBP rating, the actual hit by pitch results will all go to the few players who have this rating. So your totals might come out okay in the end, but they clearly aren't happening realistically in the games themselves.

The same would go for IBB. Your totals might be correct, but intentional walks are being given during early innings when they should be concentrated in later situations. So this means that IBBs are happening too often in the early innings and probably not enough in the later innings. The human gamer can try to address this by adjusting the strategy sliders, but doing this for every team in your baseball universe would be far too time-consuming. So this is something that may need to be tweaked as a default.

Quote:

2) Not true either, the frequencies match the ones in my retrosheet play-by-play analysis.

Again, it's not about frequencies. Your totals and frequencies might come out okay, but how did you get there? It's about the realism of stealing in certain situations. There is not enough realism in strategic decisions, and this takes some of the 'baseball' element out of the game. Getting the league totals and the frequencies correct is only half the battle. Then you have to do the detail work to provide a realistic rendering of a baseball game and all the strategy involved.

That is why there have been several topics lately that have addressed these issues. It's not about playing out your games versus simulating them either. I see the same strategic problems in the game logs for simmed games.

As I mentioned in another thread, a good example is that, when you tell teams and players to steal bases more often, you can't make a distinction between stealing second and stealing third. You also can't set your frequencies bases on the number of outs. So they will often steal second and/or try to steal third in the same inning, when it does not make strategic sense to do so.

markprior22 · 08-29-2006, 09:19 AM

Bump...hoping Markus sees these clarifications.

Markus Heinsohn · 08-29-2006, 09:44 AM

Yep, I saw these clarifications

But my real data also has it split up by inning/outs, and the simulated stolen bases reflect that.

But I will look into the intentional walks again...

markprior22 · 08-29-2006, 09:49 AM

Thanks Markus...I appreciate the time you take to look into all the suggestions/issues posted.

Pirate Fan · 08-29-2006, 12:15 PM

Quote:

Originally Posted by Markus Heinsohn

They are.

These observations are really meaningless, all that counts are test results in long term simulations, not impressions from playing out a few games.

These observations are not meaningless unless you care only about getting total IW in the game to agree more or less to actual totals.

I've seen something like the following happen numerous times in OOTP. Ninth inning AI team in the field and up by two, man on second no outs. IW to my leadoff hitter making it very likely that the AI team will have to face my 3 and 4 hitters with the tying run on base. If you watched an entire major league season, I'd be surprised if you saw a single instance of an IW in that spot.

You would not go far wrong IMO if the AI team stopped issuing IWs to the tying or winning run. Also bad is issuing IWs when the AI team is several runs ahead.

billethius · 08-29-2006, 12:17 PM

Quote:

Originally Posted by Pirate Fan

These observations are not meaningless unless you care only about getting total IW in the game to agree more or less to actual totals.

I've seen something like the following happen numerous times in OOTP. Ninth inning AI team in the field and up by two, man on second no outs. IW to my leadoff hitter making it very likely that the AI team will have to face my 3 and 4 hitters with the tying run on base. If you could watch an entire major league season, I'd be surprised if you saw a single instance of an IW in that spot.

You would not go far wrong IMO if the AI team stopped issuing IWs to the tying or winning run. Also bad is issuing IWs when the AI team is several runs ahead.

I've seen it. Ichiro gets walked all the time with first base open late in the game - especially when he's going through one of his hot streaks.

Markus Heinsohn · 08-29-2006, 12:38 PM

Quote:

Originally Posted by Pirate Fan

I've seen something like the following happen numerous times in OOTP. Ninth inning AI team in the field and up by two, man on second no outs. IW to my leadoff hitter making it very likely that the AI team will have to face my 3 and 4 hitters with the tying run on base. If you watched an entire major league season, I'd be surprised if you saw a single instance of an IW in that spot.

True. I just improved the AI in such situations

lynchjm24 · 08-29-2006, 10:00 PM

Quote:

Originally Posted by billethius

I've seen it. Ichiro gets walked all the time with first base open late in the game - especially when he's going through one of his hot streaks.

Find a game on retrosheet where up 2 runs in the 9th and a runner on second someone walked Ichiro. Even if he was hitting .500, you've still got half a chance of getting him out and the most damage he is probably going to do is equal to a walk. 'All the time? Find once.

TheBishop · 08-29-2006, 10:18 PM

Here's a problem I'm hoping will be fixed for this patch that has never been addressed once in tech support:

Player_list_panel::create_player_list - invalid column id

My home system gets this anytime I import players from a database, be it Arod/Garland or Lahman, though it happens with fictional players as well.

I've re-installed once, no change.

08-28-2006, 09:10 AM	#1
markprior22 All Star Starter Join Date: Apr 2002 Location: springfield, illinois Posts: 1,243	A few AI ingame issues hopefully addressed by 1.03 1) Too many intentional walks by the AI, usually too early in the game. 2) Way too many steal attempts of 3B. These often happen in bad situations too such as tying run on 2B with 2 outs or no outs, etc. 3) Too much pinch hitting for position players even with the setup changed to very rarely. These issues have all been mentioned before but I just want to make sure they aren't forgotten.

08-28-2006, 09:31 AM	#3
kq76 Global Moderator Join Date: Nov 2002 Posts: 12,035	I wonder, are the results different depending on whether you play or sim the games out? I'm guessing Markus compares sim results to real life data whereas people are noticing play results. __________________ My OOTP Wishlist \| My FAQ List OOTP Wiki \| Your Recommended Team Nicknames, By City (A Crowdsourced Project) For Beta/Devs: Full screen (1920x1080)

08-28-2006, 09:58 AM	#6
kq76 Global Moderator Join Date: Nov 2002 Posts: 12,035	Well, if these things really bother someone then they should probably run their own tests. Another possibility is maybe Markus is comparing results to a different time frame than others are. For example, maybe people are more comparing to what they've seen in real life over the last 10 years and maybe Markus is more comparing to what has happened over the last 40 years. Small things like this might change over time. __________________ My OOTP Wishlist \| My FAQ List OOTP Wiki \| Your Recommended Team Nicknames, By City (A Crowdsourced Project) For Beta/Devs: Full screen (1920x1080)

08-29-2006, 10:18 PM	#18
TheBishop Minors (Double A) Join Date: Jan 2004 Posts: 165	Here's a problem I'm hoping will be fixed for this patch that has never been addressed once in tech support: Player_list_panel::create_player_list - invalid column id My home system gets this anytime I import players from a database, be it Arod/Garland or Lahman, though it happens with fictional players as well. I've re-installed once, no change. __________________ Pappy's Baseball Union thread at BBSC Pappy's Baseball Union home page

08-28-2006, 09:15 AM	#2
Markus Heinsohn Developer OOTP Join Date: Dec 2001 Location: Germany Posts: 24,803	1) Not true, in my tests the league totals for IBB were usually in line with real life numbers. 2) Not true either, the frequencies match the ones in my retrosheet play-by-play analysis. 3) That just got tweaked.

08-28-2006, 10:00 AM	#7
RonCo Hall Of Famer Join Date: Aug 2003 Posts: 10,508	Sample size is always important. Statistically, OOTP 2006 has been a soild improvement in most areas I've spent time to study with a prudent sample size...thoough I haven't studied steal of third opportunities. Hmmm...I'll have to think about how to even get the data to say for sure if this is out of whack or not. I can't think of a quick way to check it, anyway.

08-28-2006, 11:14 AM	#9
markprior22 All Star Starter Join Date: Apr 2002 Location: springfield, illinois Posts: 1,243	I guess I should have been clearer...I don't have a problem with the amount of intentional walks...it is the situations in which they occur. Same with steal attempts of third. I'm not debating the overall frequencies but some of the situations are totally absurd. In general, overall statistical totals should be tested by long term stats. But, the only way to see the issues I'm referring to is by playing out games (because the quantity isn't being questioned as much as the situation). I've played out probably 90+ games since 1.02 has been out. I think that's plenty to take note of situations where intentional walks are being issued and attempts at stealing 3B are occurring. Again, I don't have a problem with the overall totals...it is the situations which are troubling.

08-29-2006, 09:19 AM	#11
markprior22 All Star Starter Join Date: Apr 2002 Location: springfield, illinois Posts: 1,243	Bump...hoping Markus sees these clarifications.

08-29-2006, 09:44 AM	#12
Markus Heinsohn Developer OOTP Join Date: Dec 2001 Location: Germany Posts: 24,803	Yep, I saw these clarifications But my real data also has it split up by inning/outs, and the simulated stolen bases reflect that. But I will look into the intentional walks again...

08-29-2006, 09:49 AM	#13
markprior22 All Star Starter Join Date: Apr 2002 Location: springfield, illinois Posts: 1,243	Thanks Markus...I appreciate the time you take to look into all the suggestions/issues posted.