A short word on Roster sets versus Game created players

Henry · 09-02-2004, 02:43 PM

This one seems to come up a lot in many different ways. There are some basic points that must be understood when analyzing player development, ratings etc... depending on whether the players are part of a roster set or game created.

Game created players are built directly from the engine, meaning that if 100 players are created, their talents, ratings, etc. will be distributed across a bell curve (or whatever distribution Markus has decided on). Player development with these players is predictable in the sense that the player creation, ability distribution, and future development are all part of the games "development plan" (internal formulas).

When a 3rd party constructs a "roster set", it is very likely the talent distribution is not equal, or "levelized". Ratings may now be out of sync with League totals, etc. As a result, the game may "adjust" these and other players in unexpected ways as the season(s) progress... designed to do one thing based on where it thinks the players have started in their values.

Were I to create a roster set, one of the things I would attempt to do is to first add up all the talents and ratings of the entire player population, then be sure as I adjusted or created players, that the totals would always remain the same. that way, the game should be able to develop the total talent pool in the same way it would were the players created in the game.

jbmagic · 09-02-2004, 03:02 PM

but in ootp 6 if u create a standard league its really cbl roster 2.0, just the names as been change

Steve Kuffrey · 09-02-2004, 03:25 PM

Which is why to get a more true respresentation of game results, player development, stats, ratings, etc it is best ot use a fictional league AND/OR sim 20-30 seasons before drawing conclusions about many matters. Many conclusions drawn part way into one season or after a season or two using a roster set can have skewed results due to the reasons Henry brings up.

Cowboys3356 · 09-02-2004, 03:39 PM

So you are suggesting that when I create a roster set if I can have the totals (player ratings, potentials, etc) equal to those of a computer generated fictional league then I will see a more stable long term development of my league?

I do something like that in the sense that I group players into different levels. I will have 5% hall of famers, 15% great, and so on down the line but I have never actually added up rating totals as you suggest.

Very interesting suggestion and I will apply it to the new 1994 League File I am working on.

Henry · 09-02-2004, 03:46 PM

Quote:

Originally Posted by Cowboys3356

So you are suggesting that when I create a roster set if I can have the totals (player ratings, potentials, etc) equal to those of a computer generated fictional league then I will see a more stable long term development of my league?

I do something like that in the sense that I group players into different levels. I will have 5% hall of famers, 15% great, and so on down the line but I have never actually added up rating totals as you suggest.

Very interesting suggestion and I will apply it to the new 1994 League File I am working on.

Yes. At one point in the past, I exported (CSVs) the talent and rating values of 10 different fictional leagues and got VERY similar totals. That's when I started wondering what would happen if a roster set was created that significantly changed those totals, or "levels" of the various players from the assumed bell curve of a generated player set. It's very likely that the player development formulas could get extreme, one way or another, resulting in unpredictable development curves.

I do want to stress what Steve said as well... ANY simulation designed to analyze the game MUST use game generated fictional players. Use of a roster set, in effect, changes the starting point rendering the study results questionable at best.

Jayzone · 09-02-2004, 09:31 PM

I concur with what Steve and Henry said in the thread.

Fictional league player composition will be pretty constant from league to league. Since the third-party roster sets are not based on the internal programming routines in OOTP but rather based on human-made judgements, the rating distribution of ficitonal league and roster league will often not match up well at all.

For fictional league league-wide rating distribution over time, I have already touched upon the subject in my league-wide rating level study for v6.01. (Look for section (4) and (5) for "Year by year ability/talent rating level summary") For what league-wide rating level distribution looks like, you can also try to read my two rating level studies for v6.01 and v6.03. In short, it has been shown that the rating level distribution of ficitonal league do not change a lot over a long period of time. The rating level is pretty consistent from year to year.

Links:

League-wide rating level study for v6.03
http://www.ootpdevelopments.com/boar...ad.php?t=71157

League-wide rating level study for v6.01
http://www.ootpdevelopments.com/boar...ad.php?t=64539

For long-term rating level stability of the custom-made roster set, the roster maker can try to match up the rating level distribution in the roster set with the ficitonal league rating level distribution. This way, the new players (draft class) generated each year can easily integrate into the existing league perfectly - in terms of raitng level. This is another reason why I would recommend roster makers to read my studies on rating level distribution. I personally has made a roster set (CPBL 2004 roster set) and I have tried to 'normalize' the rating distribution so that it will resemble the fictional league rating distribution. I will have to confess that I have not done the long-term test of my roster set, so I have no idea if this will help the long-term stability rating. In theory, however, it should help.

I also strongly echo Henry on necessity/requirement to have any study/analysis to be done with ficitonal league simulation setup. This way, we won't be seeing the effect of the human interference. To put it another way, the experiement is not done with contaminated samples.

adamj · 09-02-2004, 09:43 PM

I am playing a solo league, and started the league as a historical league, having the game import players from the Lahman Database.

I'm wondering if this would fall into the "good" category of game created leagues, or the "bad" category of roster-set leagues?

I started in 1983, but I don't import the real rookies when going to the next season. I have the amateur draft in June with fictional players. I chose to play this way because I like the theory of using fictional players, to avoid having the name recognition advantage over the AI, but I wanted to start from a world I was familiar with. This lets me move gradually into a league of fictional players. I did try starting with all fictional players, but that seemed like waking up from a 20 year coma, and taking over a team, knowing nothing about the league's history.

In any case, I guess the many, many options are a big part of this game's success, allowing each player to play in whatever way is fun for them.

Now that I've rambled enough, I guess my question is: How does the Lahman import work? Does it fit the players to an approximate bell curve based on their relative abilities or does it just assign ratings to each player individually based on his stats alone?

Steve Kuffrey · 09-02-2004, 10:04 PM

It's not that roster sets are bad, I usually start leagues with them....BUT many folks spend alot of time picking this or that area of the game apart, looking at this or that trying to judge it against RL. For this type of use roster sets are not a true measure. Using roster sets gives pretty good results(atleast I'm usually pleased with them) but to get a more accurate measure of the game using game generated ficitional players is the way to go.

jbmagic · 09-02-2004, 10:12 PM

but you guys know that the league roster file it creates when u create a standard league in ootp 6 is really a user made roster (cbl roster with the names have been changed)

Steve Kuffrey · 09-02-2004, 10:20 PM

You already posted this in this thread which is why I made the comment about fictional players right after it.

Jayzone · 09-02-2004, 10:27 PM

I will have to chip in a bit here.

No roster sets are bad. Every roster maker spends a lot of their own time to make an enjoyable league. Some roster sets are unique and they try to bring a different flavor of baseball. It's just some roster may be better at creating a replica of real-life baseball and some are better at giving the user a different 'feel' of baseball. We should take time to thanks every roster maker/mod maker for bringing us a more enriched OOTP experience.

And I agree with what Steve said about the difference between roster sets and ficitonal leagues.

dougaiton · 09-03-2004, 05:52 AM

I'm not quite sure what roster sets have to do with testing. I think everyone was pretty clear on where to go next (I have instructions in the joint '69 file as to how to create realistic-ish players for the 1970s).

What is relevant to this discussion is to what extent fictional leagues simulate real life, which is what most of the threads have been about (as far as I can see).

pat · 09-03-2004, 06:27 AM

Quote:

Originally Posted by Steve Kuffrey

Which is why to get a more true respresentation of game results, player development, stats, ratings, etc it is best ot use a fictional league AND/OR sim 20-30 seasons before drawing conclusions about many matters. Many conclusions drawn part way into one season or after a season or two using a roster set can have skewed results due to the reasons Henry brings up.

Can i use this Steve to plug into the league i am in?

WWBL is a fictional league and the commish simmed 20 seasons to generate history and i think that's real good because of what you said..

draven085 · 09-03-2004, 09:01 AM

Quote:

Originally Posted by dougaiton

What is relevant to this discussion is to what extent fictional leagues simulate real life, which is what most of the threads have been about (as far as I can see).

Right on. This entire thread is based on the assumption that talent distribution in ficitional leagues simulates that of rea-life professional baseball. That's not an assumption I am willing to make.

Henry · 09-03-2004, 09:19 AM

Quote:

Originally Posted by draven085

Right on. This entire thread is based on the assumption that talent distribution in ficitional leagues simulates that of rea-life professional baseball. That's not an assumption I am willing to make.

Sorry, but no. That's not what we are saying.

What we ARE saying is that to analyze the game correctly, you HAVE to use fictional players. That will tell you if the game engine is replicating the kind of world you want. Once you've establish that point, THEN you can begin studies on roster sets and see what is required to make those realistic.

In the perfect world, fictional player creation would simulate the MLB perfectly, and a set of available guidelines would contain instructions on how to build roster sets so that perfection is duplicated.

sporr · 09-03-2004, 11:24 AM

To use an extreme example, let's say that you create a roster and give every player 100s across the board regardless of age. If you notice that the game engine is giving all of these players severe talent hits day after day, you may draw the incorrect conclusion that the game player development is 'off' by hitting players too hard. In fact, this is just how the game is trying to adjust for the unnaturally high starting values.

Steve Kuffrey · 09-03-2004, 11:25 AM

Quote:

Originally Posted by sporr

To use an extreme example, let's say that you create a roster and give every player 100s across the board regardless of age. If you notice that the game engine is giving all of these players severe talent hits day after day, you may draw the incorrect conclusion that the game player development is 'off' by hitting players too hard. In fact, this is just how the game is trying to adjust for the unnaturally high starting values.

Good point

ctorg · 09-03-2004, 11:27 AM

I'd never even imagined that people would do this. Very good point to bring up, though. It completely undermines all testing if you start with a roster set.

Henry · 09-03-2004, 11:27 AM

FYI : I have a study underway that will detail out as much as possible what a typical MLB 30 team league looks like in terms of average talent, ratings, etc. I've got it about half done but am leaving for Baltimore today. I'll probably have it done by the end of next weekend (12th). This report might provide the "average" that all created leagues should levelize to.

Dan Theman · 09-03-2004, 12:06 PM

Thanks for the info, Henry - that sounds wonderful.

As an aside, after I had to go a few days without internet access (see my OT rant on cable companies for more info), I have to say I've got this game to thank for allowing me to keep my sanity through my internet withdrawal period. They should introduce it to the Betty Ford clinic

09-02-2004, 02:43 PM	#1
Henry Hall of Fame Join Date: Dec 2001 Posts: 6,498	A short word on Roster sets versus Game created players This one seems to come up a lot in many different ways. There are some basic points that must be understood when analyzing player development, ratings etc... depending on whether the players are part of a roster set or game created. Game created players are built directly from the engine, meaning that if 100 players are created, their talents, ratings, etc. will be distributed across a bell curve (or whatever distribution Markus has decided on). Player development with these players is predictable in the sense that the player creation, ability distribution, and future development are all part of the games "development plan" (internal formulas). When a 3rd party constructs a "roster set", it is very likely the talent distribution is not equal, or "levelized". Ratings may now be out of sync with League totals, etc. As a result, the game may "adjust" these and other players in unexpected ways as the season(s) progress... designed to do one thing based on where it thinks the players have started in their values. Were I to create a roster set, one of the things I would attempt to do is to first add up all the talents and ratings of the entire player population, then be sure as I adjusted or created players, that the totals would always remain the same. that way, the game should be able to develop the total talent pool in the same way it would were the players created in the game.

09-02-2004, 03:25 PM	#3
Steve Kuffrey Administrator Join Date: Jul 1999 Location: S.E. TN - Georgia born and raised Posts: 17,035	Which is why to get a more true respresentation of game results, player development, stats, ratings, etc it is best ot use a fictional league AND/OR sim 20-30 seasons before drawing conclusions about many matters. Many conclusions drawn part way into one season or after a season or two using a roster set can have skewed results due to the reasons Henry brings up. __________________ Steve Kuffrey DABS Atlanta Braves - 2008 Eastern Division Champ *DBLC Atlanta Braves - 2011, 2014 East Division Champ, 2012, 2013 NL Wildcard Baseball Maelstrom-Montreal Expos-2013 Tourney winner, 2014 WC Team Sparky's League - Tampa Bay D'Rays Epicenter Baseball League - Astros 2014 The CBL Rewind - Phillies '95

09-02-2004, 09:31 PM	#6
Jayzone All Star Reserve Join Date: Apr 2004 Posts: 706	I agree. The researches should be done ONLY with fictional league setup. I concur with what Steve and Henry said in the thread. Fictional league player composition will be pretty constant from league to league. Since the third-party roster sets are not based on the internal programming routines in OOTP but rather based on human-made judgements, the rating distribution of ficitonal league and roster league will often not match up well at all. For fictional league league-wide rating distribution over time, I have already touched upon the subject in my league-wide rating level study for v6.01. (Look for section (4) and (5) for "Year by year ability/talent rating level summary") For what league-wide rating level distribution looks like, you can also try to read my two rating level studies for v6.01 and v6.03. In short, it has been shown that the rating level distribution of ficitonal league do not change a lot over a long period of time. The rating level is pretty consistent from year to year. Links: League-wide rating level study for v6.03 http://www.ootpdevelopments.com/boar...ad.php?t=71157 League-wide rating level study for v6.01 http://www.ootpdevelopments.com/boar...ad.php?t=64539 For long-term rating level stability of the custom-made roster set, the roster maker can try to match up the rating level distribution in the roster set with the ficitonal league rating level distribution. This way, the new players (draft class) generated each year can easily integrate into the existing league perfectly - in terms of raitng level. This is another reason why I would recommend roster makers to read my studies on rating level distribution. I personally has made a roster set (CPBL 2004 roster set) and I have tried to 'normalize' the rating distribution so that it will resemble the fictional league rating distribution. I will have to confess that I have not done the long-term test of my roster set, so I have no idea if this will help the long-term stability rating. In theory, however, it should help. I also strongly echo Henry on necessity/requirement to have any study/analysis to be done with ficitonal league simulation setup. This way, we won't be seeing the effect of the human interference. To put it another way, the experiement is not done with contaminated samples.

09-02-2004, 10:04 PM	#8
Steve Kuffrey Administrator Join Date: Jul 1999 Location: S.E. TN - Georgia born and raised Posts: 17,035	It's not that roster sets are bad, I usually start leagues with them....BUT many folks spend alot of time picking this or that area of the game apart, looking at this or that trying to judge it against RL. For this type of use roster sets are not a true measure. Using roster sets gives pretty good results(atleast I'm usually pleased with them) but to get a more accurate measure of the game using game generated ficitional players is the way to go. __________________ Steve Kuffrey DABS Atlanta Braves - 2008 Eastern Division Champ *DBLC Atlanta Braves - 2011, 2014 East Division Champ, 2012, 2013 NL Wildcard Baseball Maelstrom-Montreal Expos-2013 Tourney winner, 2014 WC Team Sparky's League - Tampa Bay D'Rays Epicenter Baseball League - Astros 2014 The CBL Rewind - Phillies '95

09-02-2004, 10:20 PM	#10
Steve Kuffrey Administrator Join Date: Jul 1999 Location: S.E. TN - Georgia born and raised Posts: 17,035	You already posted this in this thread which is why I made the comment about fictional players right after it. __________________ Steve Kuffrey DABS Atlanta Braves - 2008 Eastern Division Champ *DBLC Atlanta Braves - 2011, 2014 East Division Champ, 2012, 2013 NL Wildcard Baseball Maelstrom-Montreal Expos-2013 Tourney winner, 2014 WC Team Sparky's League - Tampa Bay D'Rays Epicenter Baseball League - Astros 2014 The CBL Rewind - Phillies '95

09-02-2004, 03:02 PM	#2
jbmagic Hall Of Famer Join Date: Sep 2003 Location: Los Angeles Posts: 3,417	but in ootp 6 if u create a standard league its really cbl roster 2.0, just the names as been change

09-02-2004, 03:39 PM	#4
Cowboys3356 All Star Starter Join Date: Sep 2003 Posts: 1,571	So you are suggesting that when I create a roster set if I can have the totals (player ratings, potentials, etc) equal to those of a computer generated fictional league then I will see a more stable long term development of my league? I do something like that in the sense that I group players into different levels. I will have 5% hall of famers, 15% great, and so on down the line but I have never actually added up rating totals as you suggest. Very interesting suggestion and I will apply it to the new 1994 League File I am working on.

09-02-2004, 09:43 PM	#7
adamj Minors (Double A) Join Date: Jul 2004 Posts: 101	I am playing a solo league, and started the league as a historical league, having the game import players from the Lahman Database. I'm wondering if this would fall into the "good" category of game created leagues, or the "bad" category of roster-set leagues? I started in 1983, but I don't import the real rookies when going to the next season. I have the amateur draft in June with fictional players. I chose to play this way because I like the theory of using fictional players, to avoid having the name recognition advantage over the AI, but I wanted to start from a world I was familiar with. This lets me move gradually into a league of fictional players. I did try starting with all fictional players, but that seemed like waking up from a 20 year coma, and taking over a team, knowing nothing about the league's history. In any case, I guess the many, many options are a big part of this game's success, allowing each player to play in whatever way is fun for them. Now that I've rambled enough, I guess my question is: How does the Lahman import work? Does it fit the players to an approximate bell curve based on their relative abilities or does it just assign ratings to each player individually based on his stats alone?

09-02-2004, 10:12 PM	#9
jbmagic Hall Of Famer Join Date: Sep 2003 Location: Los Angeles Posts: 3,417	but you guys know that the league roster file it creates when u create a standard league in ootp 6 is really a user made roster (cbl roster with the names have been changed)

09-02-2004, 10:27 PM	#11
Jayzone All Star Reserve Join Date: Apr 2004 Posts: 706	I will have to chip in a bit here. No roster sets are bad. Every roster maker spends a lot of their own time to make an enjoyable league. Some roster sets are unique and they try to bring a different flavor of baseball. It's just some roster may be better at creating a replica of real-life baseball and some are better at giving the user a different 'feel' of baseball. We should take time to thanks every roster maker/mod maker for bringing us a more enriched OOTP experience. And I agree with what Steve said about the difference between roster sets and ficitonal leagues.

09-03-2004, 05:52 AM	#12
dougaiton Hall Of Famer Join Date: Mar 2003 Location: Location: Posts: 3,414	I'm not quite sure what roster sets have to do with testing. I think everyone was pretty clear on where to go next (I have instructions in the joint '69 file as to how to create realistic-ish players for the 1970s). What is relevant to this discussion is to what extent fictional leagues simulate real life, which is what most of the threads have been about (as far as I can see). __________________ tThis shower of clowns My park-adjusted, historically split Born in '69 league file! (also at Joe's wonderful site +BBSC)

09-03-2004, 11:24 AM	#16
sporr Global Moderator Join Date: Dec 2001 Location: Muscatine, IA Posts: 8,277	To use an extreme example, let's say that you create a roster and give every player 100s across the board regardless of age. If you notice that the game engine is giving all of these players severe talent hits day after day, you may draw the incorrect conclusion that the game player development is 'off' by hitting players too hard. In fact, this is just how the game is trying to adjust for the unnaturally high starting values.

09-03-2004, 11:27 AM	#18
ctorg Global Moderator Join Date: Nov 2002 Location: Queens, NY Posts: 9,848	I'd never even imagined that people would do this. Very good point to bring up, though. It completely undermines all testing if you start with a roster set. __________________ My music "When the trees blow back and forth, that's what makes the wind." - Steven Wright Fjord emena pancreas thorax fornicate marmalade morpheme proteolysis smaxa cabana offal srue vitriol grope hallelujah lentils