search

AustinMayerhofer / NBA-metrics

Collection of different NBA metrics to assess player/team performance.
MIT License
0 stars 0 forks source link

All time rankings #16

Open AustinMayerhofer opened 2 years ago

AustinMayerhofer commented 2 years ago

Figure out an algorithm to rank players all time

Given a players' ratings array, of their individual season scores, how would you rank them?

e. g. Steph Curry [5, 3.3, 2, -1, 10]

James Harden [6, 3, 4, 4]

Brandon Roy [7]

Perhaps there's a statistical algorithm to rank different length arrays, giving bonuses to players who have exceptional years, not punishing players as much for bad years, and rewarding more years played but not too much past a certain threshold

AustinMayerhofer commented 2 years ago

Add extra points to a player's season if they make the conference finals, finals, or win a title.

How much points a player earns should be based on how much they contribute in the playoffs. If they are the #1 contributor, they earn the max points for the playoff finish. #2 contributor earns, say, 90%. And so on.

AustinMayerhofer commented 2 years ago

One idea for an algorithm is taking the array of player seasons. Sort the array, and sum the top N seasons (where N could be 8, 9, 10, whatever the best fit is) and any other seasons are extra, and added with a multiplier. What that multiplier is, is open for discussion.

AustinMayerhofer commented 2 years ago

Maybe add 5 to PlayerScore to get more positive z scores.

Edit: 6 probably better, don't want too many negative PlayerScores

AustinMayerhofer commented 2 years ago

Maybe add 5 to PlayerScore to get more positive z scores.

Piggybacking on this, summing PlayerScores for all time stats with +5 may take away from "great" seasons. A 12 vs 2 season becomes a 17 vs 7. 6:1 vs 2.43:1 ratio.

To counteract this, consider raising the new PlayerScore to an exponent (if PlayerScore is positive). Say, to the 1.5 power.

17^{1.5) = 70, 7^(1.5) = 18.5.

This is a 3.78:1 ratio, not as bad as the 2.43:1 ratio.