For each legal move, evaluate the resulting position from the mover's
perspective, attach a win_prob, derive a chess-engine-style label
("best" / "strong" / "ok" / "weak" / "blunder") from the
probability drop vs the best move, and rank the rows.
With method = "search", ranks use the negamax score; win_prob
remains the one-ply heuristic estimate and label is NA, because that
probability model has not been calibrated for search scores.
Arguments
- game
An
mxo_gameobject.- method
One of
"search"(negamax depth lookup),"mcts", or"heuristic". Default"heuristic"for speed;"search"and"mcts"cost more computation. MCTS ratings use random rollouts by default; passrollout = "heuristic"to select the more expensive policy.- depth
Search depth when
method = "search". Default2L.- mcts_iter
Iterations when
method = "mcts". Default200L.- branch_policy
Branch-policy passed through to search/MCTS. Default
"promising".- ...
Additional arguments forwarded to the chosen method.
Value
A type-stable tibble with columns kind, L_src, t_src,
idx, player, score, win_prob, rank, label.
Examples
g <- mxo_new_game(n = 3L, k = 3L)
head(mxo_rate_moves(g))
#> # A tibble: 6 × 9
#> kind L_src t_src idx player score win_prob rank label
#> <chr> <int> <int> <int> <int> <dbl> <dbl> <int> <chr>
#> 1 present 0 0 0 1 14 0.543 2 best
#> 2 present 0 0 1 1 8 0.543 16 best
#> 3 present 0 0 2 1 14 0.543 2 best
#> 4 present 0 0 3 1 8 0.543 16 best
#> 5 present 0 0 4 1 10 0.543 10 best
#> 6 present 0 0 5 1 8 0.543 16 best