Calling all mathematicians and statisticians

[jdavidbakr]

I'm working on an update to Conscriptor that will do a mock draft and allow for scout picks automatically based on the relative roster strengths. I'm using the best future value of each position (using the higher position types like G, T, DE, ILB, rather than LG, RG, etc) subtracted from 100 to gauge the need at each position. So if a team has a 55/85 qb, their need for QB would be 15, while if their best guard was 33/33, their need for guard would be 67. I am taking this value along with the adjusted grade for the players in the draft to create a priority list for each pick in the mock draft or the scout pick. Here is the formula I have come up with that seems to generate a good balance between team need and adjusted grade:

score=(adjusted_grade^2)*team_need

Nice and simple, and yet seems to give a nice weight to the adjusted grade without overwhelming it. For my test draft I'm using the vNFL draft data, the first pick goes to CIN. This is the result of the sorting:

Team Need	Adjusted Grade of Player	Score
61	7.70	3616.68
45	8.80	3484.80
48	7.40	2628.48

The 8.80 player was a very good-looking QB, who was actually taken by CIN the draft. But notice that his need for a QB was only 45 compared to his need for a RB at 61 (which is what the 7.70 player is), which is why this algorithm selected the RB over the QB in this situation.

So my question for you guys is, does this look like an acceptable formula, or should it be adjusted any? Anyone who works with numbers like these more than I do could probably tell me if I'm close at all to having a good algorithm.

FWIW, CIN didn't have a K on their roster, so the need came in at 100. The top adjusted score for a kicker is 3.7, which gave a score of 1369, placing it as #13 on the list, so it's unlikely that a team will draft a kicker in an early round with this formula. But it did take the kicker at #13 before a 5.00 FB (#14), their need was 49 for a FB (which means the future value of their FB is 51, so probably a kicker would have been a good choice before a FB anyway). #15 was a 5.9 DT with their need at 26, followed by the final spot at #16 as the punter (3.2, need = 52)

[Matthean]

Well, if you were Cincy, would you have taken the obviously better pick even though it was a #2 need versus the guy that fills the #1 need? You'll get a lot of split answers on that. It's going to be a huge debate myself when I do my draft and if two guys are sitting there. A general rule of thumb is you'll go for the need pick early on and fill in your roster later with that's the best player available.

[Celeval]

Given that you're doing a mock from this, I think it's a pretty reasonable starting space. Best bet is to run a number of drafts, look for the holes, and adjust if/as necessary. Good stuff.

[NawlinsFan]

Personally........ I like it JDB. I understand how some will align themselves with the prettiest pick but I am one of those that believes in trying to create strength all around with adequate depth to carry the team during injuries. I look to filling where the strongest need is, so this is perfect for someone like me. Then again, I still consider myself an amatuer with this game and have yet to learn the in's and outs like most here have.

[Sgt Tomorrow]

For a MOCK draft, it's purpose to GENERATE discussion in multi-player league forums, it's fantastic! Great job Jon

[jdavidbakr]

Thanks for the feedback. It does seem to be a good mock draft formula, I'm also using it for "scout pick" now in the utility. I agree that the pick would be kind of a tough call for Cincy in this case, which is what makes that sample such an interesting case and a good one to gauge the formula by.

Maybe I could even have a series of formulas that the user chooses from when they pick "scout pick" and even randomly assign the formulas to each team or even each pick for the mock draft.

[OldSchool]

Originally Posted by jdavidbakr

score=(adjusted_grade^2)*team_need

The half-assed algorithm I use to come up with mocks includes one additional component, a threshold for how much lower rated a player can be than the "best player available" and still be considered for the pick.

[WebEwbank]

This is great stuff. I think that you might want to consider a few tweaks later on. For one, when the numbers generated are as close as 34xx and 36xx you may want to underline that it is judgment call (although the mock does require a decision). Another thought is to weight the positions' importance so that you DON'T get the kicker or a FB early no matter what the need and the player quality.

[jdavidbakr]

Originally Posted by WebEwbank

This is great stuff. I think that you might want to consider a few tweaks later on. For one, when the numbers generated are as close as 34xx and 36xx you may want to underline that it is judgment call (although the mock does require a decision). Another thought is to weight the positions' importance so that you DON'T get the kicker or a FB early no matter what the need and the player quality.

That got me thinking ... why not add commentary to the mock draft? So, using a bunch of AI decisions, I added commentary. Here's the same sample (from the current vNFL draft data that I have been using to test):