What do you set the AI Evaluation weights at?

CardinalFan · 04-23-2007, 12:48 PM

I guess the default setting is
30% Ratings
55% Current Stats
15% Last Season Stats

But what do you set the weights at? Is that the recommended setting?
What about when starting a fictional league from scratch where there are no past seasons and the current stats are very low?

biggamejrs · 04-23-2007, 12:50 PM

I think there is a 4th category. The last is for two years ago stats or similar.

CardinalFan · 04-23-2007, 12:58 PM

Quote:

Originally Posted by biggamejrs

I think there is a 4th category. The last is for two years ago stats or similar.

Yeah...probably is. I don't have the game or manual here in front of me and was posting from memory.

Nukester · 04-23-2007, 01:08 PM

I currently have mine set to:

40% - Ratings
30% - Current Stats
25% - Last Seasons Stats
5% - 2 Years Ago Stats

Im trying these out, as thats probably about the way I view players. Im not sure what happens when there are no past stats to look at

james17 · 04-23-2007, 01:41 PM

60% ratings
25% current year
10% last season
5% 2 years ago.

Ksyrup · 04-23-2007, 01:51 PM

10/50/30/10

Curtis · 04-23-2007, 01:58 PM

50/29/14/7

CardinalFan · 04-23-2007, 01:58 PM

I am surprised.
I thought there would be some kind of consensus, but it really appears that we each differ in where we set the weights.

1998 Yankees · 04-23-2007, 02:36 PM

I am a big fan of leaving settings at default values unless I have reason to change them.

The Wolf · 04-23-2007, 09:55 PM

Quote:

Originally Posted by Ksyrup

10/50/30/10

I'm trying this now and so far I like it better than using the default settings.

pstrickert · 04-23-2007, 10:13 PM

Quote:

Originally Posted by The Wolf

I'm trying this now and so far I like it better than using the default settings.

Because . . . ?

The Wolf · 04-23-2007, 10:25 PM

Because the AI seems to be making better decisions. It's picking people who may not have good ratings but who do perform well over those who have good ratings but who perform badly. There are a lot fewer low-BA, low-OBP players in lineups now.

ravinhood · 04-23-2007, 10:33 PM

33%- ratings
33%- current stats
33%- previous year stats
1- 2 year old stats

This gives the AI the best look at a player for the current year, the previous year and the ratings. Ratings in many cases don't mean squat, it's the actual stats that have value, but, even they can be deceiving since one year a player can have a great year and be crap the next year. Two years ago to me means squat and I don't even look at 2 year old stats vs ratings and current and previous year stats. This also keeps the AI from offering huge amounts of money to those aging players as well like in 6.5a and I see more reasonable free agency outcomes in price for players than the constant 20 mil plus as in 6.5a as well.

DougWyatt · 04-23-2007, 10:37 PM

40
30
20
10

Russ · 04-23-2007, 10:43 PM

Does this control star rating too or is it solely affecting AI roster decisions?

ravinhood · 04-23-2007, 10:53 PM

It effects a lot of things, one being trades another being getting free agency players and how much they will offer. Not sure it effects roster decisions since that should come from the strategy settings of your team AI manager. If you're just playing General Manager and letting the AI manage the team as I do save for initial lineups. Oh it also effects negotiation of salaries offered for extentions as well.

CatKnight · 04-24-2007, 03:22 AM

I read in another thread somewhere that the AI may be jumping the gun with players, sending someone with falling ratings but still solid performance to the minors or releasing them prematurely.

I tried a season at 20/50/20/10 and it seemed to work well.

The Wolf · 04-24-2007, 08:57 AM

Quote:

Originally Posted by CatKnight

I read in another thread somewhere that the AI may be jumping the gun with players, sending someone with falling ratings but still solid performance to the minors or releasing them prematurely.

I tried a season at 20/50/20/10 and it seemed to work well.

Yes. Going 10/50/30/10 put a quick stop to that, too.

I may try 20/50/20/10 or maybe 15/50/25/10 next. Does the AI keep poor performers with good ratings in the lineups at 20/50/20/10?

Ksyrup · 04-24-2007, 10:50 AM

Even with 10/50/30/10, I'm still seeing what I think is the AI not properly evaluating guys based on stats, but it's more of an anecdotal thing than anything scientific. I have a couple of pitchers with decent ratings but who have performed incredibly well, and just doing a test to see what the AI would offer me in a trade, they aren't fetching anything. For instance, I'm in 1910 and George Mullin of the Tigers is coming off of 3 straight 25-35 win, 1.50-1.90 ERA seasons, and I can't give him away. Given the evaluation criteria, his past couple of years should weigh heavily in the AI's "mind." He's still in his early 30s, so it's not like he's about to retire.

It's things like that that still make me think the AI is putting too much emphasis on ratings versus performance. I'd like to see a true 10/50/30/10 AI evaluation process, because I don't think we're quite there yet.

t-bone shuffle · 04-24-2007, 10:56 AM

I play around with it some, but almost always have the ratings based at either 40 or 50%. What I do that hasn't been mentioned here, but I first picked up from 6.5 discussions, is to adjust the % for current season/last season stats as the current season progresses. For example, early in the season, I'd have last seasons stats at close to 50% and current at about 5%. I'll then adjust these monthly until current season stats are closer to the 50% number somewhere around mid August.

I haven't progressed far enough into a season to report on how 2007 is reacting to this, but early season seems to be working just fine.

04-23-2007, 12:48 PM	#1
CardinalFan Bat Boy Join Date: Jul 2003 Posts: 14	What do you set the AI Evaluation weights at? I guess the default setting is 30% Ratings 55% Current Stats 15% Last Season Stats But what do you set the weights at? Is that the recommended setting? What about when starting a fictional league from scratch where there are no past seasons and the current stats are very low?

04-23-2007, 12:50 PM	#2
biggamejrs Minors (Triple A) Join Date: Dec 2001 Location: Greenville, USA Posts: 212	I think there is a 4th category. The last is for two years ago stats or similar. __________________ Go Braves!!

04-23-2007, 01:08 PM	#4
Nukester All Star Starter Join Date: Dec 2001 Location: Newburgh, NY Posts: 1,923	I currently have mine set to: 40% - Ratings 30% - Current Stats 25% - Last Seasons Stats 5% - 2 Years Ago Stats Im trying these out, as thats probably about the way I view players. Im not sure what happens when there are no past stats to look at Last edited by Nukester; 04-23-2007 at 01:11 PM.

04-23-2007, 10:33 PM	#13
ravinhood Major Leagues Join Date: Apr 2004 Posts: 374 Infractions: 0/1 (1)	33%- ratings 33%- current stats 33%- previous year stats 1- 2 year old stats This gives the AI the best look at a player for the current year, the previous year and the ratings. Ratings in many cases don't mean squat, it's the actual stats that have value, but, even they can be deceiving since one year a player can have a great year and be crap the next year. Two years ago to me means squat and I don't even look at 2 year old stats vs ratings and current and previous year stats. This also keeps the AI from offering huge amounts of money to those aging players as well like in 6.5a and I see more reasonable free agency outcomes in price for players than the constant 20 mil plus as in 6.5a as well. Last edited by ravinhood; 04-23-2007 at 10:34 PM.

04-23-2007, 10:43 PM	#15
Russ All Star Starter Join Date: Oct 2005 Location: Essex HON! Posts: 1,923 Infractions: 0/1 (1)	Does this control star rating too or is it solely affecting AI roster decisions? __________________ If you don't love Russ, you don't love America. This post brought to you by Carl's Jr.

04-23-2007, 01:41 PM	#5
james17 All Star Starter Join Date: Nov 2004 Posts: 1,305	60% ratings 25% current year 10% last season 5% 2 years ago.

04-23-2007, 01:51 PM	#6
Ksyrup Major Leagues Join Date: Feb 2002 Posts: 413	10/50/30/10

04-23-2007, 01:58 PM	#7
Curtis Hall Of Famer Join Date: Jul 2006 Location: Watertown, New York Posts: 4,567	50/29/14/7

04-23-2007, 02:36 PM	#9
1998 Yankees Hall Of Famer Join Date: Aug 2006 Location: Yankee Stadium, back in 1998. Posts: 8,645	I am a big fan of leaving settings at default values unless I have reason to change them.

04-23-2007, 10:37 PM	#14
DougWyatt All Star Starter Join Date: Dec 2003 Location: Michigan Posts: 1,117	40 30 20 10

04-23-2007, 10:53 PM	#16
ravinhood Major Leagues Join Date: Apr 2004 Posts: 374 Infractions: 0/1 (1)	It effects a lot of things, one being trades another being getting free agency players and how much they will offer. Not sure it effects roster decisions since that should come from the strategy settings of your team AI manager. If you're just playing General Manager and letting the AI manage the team as I do save for initial lineups. Oh it also effects negotiation of salaries offered for extentions as well. Last edited by ravinhood; 04-23-2007 at 10:56 PM.

04-24-2007, 03:22 AM	#17
CatKnight All Star Starter Join Date: Apr 2007 Location: Cleveland, OH Posts: 1,199	I read in another thread somewhere that the AI may be jumping the gun with players, sending someone with falling ratings but still solid performance to the minors or releasing them prematurely. I tried a season at 20/50/20/10 and it seemed to work well.

04-24-2007, 10:50 AM	#19
Ksyrup Major Leagues Join Date: Feb 2002 Posts: 413	Even with 10/50/30/10, I'm still seeing what I think is the AI not properly evaluating guys based on stats, but it's more of an anecdotal thing than anything scientific. I have a couple of pitchers with decent ratings but who have performed incredibly well, and just doing a test to see what the AI would offer me in a trade, they aren't fetching anything. For instance, I'm in 1910 and George Mullin of the Tigers is coming off of 3 straight 25-35 win, 1.50-1.90 ERA seasons, and I can't give him away. Given the evaluation criteria, his past couple of years should weigh heavily in the AI's "mind." He's still in his early 30s, so it's not like he's about to retire. It's things like that that still make me think the AI is putting too much emphasis on ratings versus performance. I'd like to see a true 10/50/30/10 AI evaluation process, because I don't think we're quite there yet.

04-24-2007, 10:56 AM	#20
t-bone shuffle All Star Reserve Join Date: Jan 2004 Location: The dull edge of the blade Posts: 867	I play around with it some, but almost always have the ratings based at either 40 or 50%. What I do that hasn't been mentioned here, but I first picked up from 6.5 discussions, is to adjust the % for current season/last season stats as the current season progresses. For example, early in the season, I'd have last seasons stats at close to 50% and current at about 5%. I'll then adjust these monthly until current season stats are closer to the 50% number somewhere around mid August. I haven't progressed far enough into a season to report on how 2007 is reacting to this, but early season seems to be working just fine.