People Generated Rosters

[miahwitt]

I am not sure if anyone has noticed this, but there is something people generated rosters that change some of the dynamics we see on the field.

My friends and I a few months back started an online dynasty with BossHawg rosters. They were great to have and the realist rosters were fun. But something I noticed being way off was speed. I couldn’t get to the edge I could break big ones. And if I tapped R2 for a speed boost I swear the defense was coming faster.

After a few season it dawned on me how manipulated the rosters are. And I thought maybe that affected the play we were seeing on the field.

Sure enough we hit year 6. Every roster would be filled with CPU generated players. Created in a way that would match the way the original game developers put the game together for us to experience a certain “type” of gameplay.

I would say aside from sliders the game is smoother with CPU rosters. I plan on any future endeavors to sim 5 years to achieve this.


Sent from my iPhone using Operation Sports

[FarFromEer]

Last year, I simmed out a 60 year dynasty, retaining save files from the 2018 season to the end to look at how the system created players and how training results progressed. With regards to roster files, I used the preseason save for each year and accumulated the data to see what the max/min and average for each skill was by position, tendency, and years in the game.

Since you mentioned speed, I'll break it down a bit compared to the 2018/19 roster project. Pocket QBs are rated 9-14 points faster in the roster file than the average of the CPU created. Balanced QBs rate about 5 points faster. Scrambling QBs rated 1-4 points slower. Of all the categories of players in my data set, the average speed ratings are anywhere from -13 to 17 points from the CPU generated average.

I've played with sliders quite a bit, following along with many active threads as they struggled to do the same. So many players are trying to find a way to fix the inherent issues left in the game. There are many different styles of players, from just wanting to play for fun, to wanting realism in the stats, to wanting to see realism in the animations, and varying combinations thereof. Within these players, there are those that don't use the editor to those that love the editor as much as the game itself. The issue with using one scale system (player generated) vs another (CPU generated) is the level of continued involvement you want to put into the game as it progresses. Having player skills that are drastically different from the CPU requires living with the change over the first 5 seasons (and probably slider tweaks to offset) or rescaling the recruiting class before each season to the player generated scale. There is also a third option, which is rescaling the player generated roster to try to match the CPU models.

[big09]

Quote: Originally Posted by mcfeenathan

Last year, I simmed out a 60 year dynasty, retaining save files from the 2018 season to the end to look at how the system created players and how training results progressed. With regards to roster files, I used the preseason save for each year and accumulated the data to see what the max/min and average for each skill was by position, tendency, and years in the game. Since you mentioned speed, I'll break it down a bit compared to the 2018/19 roster project. Pocket QBs are rated 9-14 points faster in the roster file than the average of the CPU created. Balanced QBs rate about 5 points faster. Scrambling QBs rated 1-4 points slower. Of all the categories of players in my data set, the average speed ratings are anywhere from -13 to 17 points from the CPU generated average. I've played with sliders quite a bit, following along with many active threads as they struggled to do the same. So many players are trying to find a way to fix the inherent issues left in the game. There are many different styles of players, from just wanting to play for fun, to wanting realism in the stats, to wanting to see realism in the animations, and varying combinations thereof. Within these players, there are those that don't use the editor to those that love the editor as much as the game itself. The issue with using one scale system (player generated) vs another (CPU generated) is the level of continued involvement you want to put into the game as it progresses. Having player skills that are drastically different from the CPU requires living with the change over the first 5 seasons (and probably slider tweaks to offset) or rescaling the recruiting class before each season to the player generated scale. There is also a third option, which is rescaling the player generated roster to try to match the CPU models.

Do you have any ideas how to "scale to player generated rosters?" Cleary you have put in a great amount of time and research and I would love to hear your thoughts on this. I have started on a 2020 real recruiting class and would hope that I can come up with something that plays good with all rosters. Any input from your research would be welcomed.

[FarFromEer]

Quote: Originally Posted by big09

Do you have any ideas how to "scale to player generated rosters?" Cleary you have put in a great amount of time and research and I would love to hear your thoughts on this. I have started on a 2020 real recruiting class and would hope that I can come up with something that plays good with all rosters. Any input from your research would be welcomed.

I haven't analysed recruiting classes, but here would be my method:

- Generate and save a lot of recruiting classes in the game (at least 20-25, but the more you get, the better your data will be)
- Export and compile the data into a spreadsheet
- Find a way to group players (I would probably use position, tendency, and star rating)
- Find the max/min and average for each grouping combination (the biggest issue will be the athletes. You probably have to figure out what positions they will be best at, then apply multiple criteria based on that. )

The issue with scaling from player generated to the CPU models is realism. The method BossHawg developed creates a very consistent skills from player to player, regardless of position. The CPU averages, however do not align with realistic straight line measurables like speed would. The position/tendency averages EA is using are not realistic. I would create a realistic method, then apply the max/min and average accordingly. What I mean is, if 3 star power HB average out with your method to an 85 speed, but the CPU average is 78, subtract 7 point from every 3 star power HB in your list, but make sure none of your 3 star power HB recruits speed is over or under the max/min of the CPU players (this is where the more CPU data you collect, the better).