Estimate Player Stat Ratings Using Bayesian Conjugate Priors

For each player, estimates "true stat rating" at a reference date using all prior matches. Uses conjugate Bayesian updating with exponential time decay.

Usage

estimate_player_stat_ratings(
  stat_data,
  ref_date = NULL,
  params = NULL,
  stat_defs = NULL,
  compute_ci = TRUE
)

Arguments

stat_data

data.table from prepare_stat_rating_data.

ref_date

Date to estimate ratings as of. Only matches before this date are used. If NULL, uses one day after the latest match.

params

Named list from default_stat_rating_params.

stat_defs

data.frame from stat_rating_definitions.

compute_ci

Logical. If TRUE (default), compute credible intervals.

Value

data.table with one row per player containing: player_id, role_group, n_matches, wt_matches, ref_date, and for each stat: {stat}_rating, {stat}_rating_lower, {stat}_rating_upper.

Details

Rate stats (per-ball): Gamma-Poisson model. Raw event counts are decay-weighted; the Gamma prior is centered on the role mean with strength controlled by prior_strength. Posterior mean gives the estimated per-ball rate.

Efficiency stats (proportions): Beta-Binomial model. Successes and attempts are decay-weighted; the Beta prior is centered on the role mean with strength controlled by prior_strength.