Records per-ply diagnostics (when record_eval is TRUE) so the result
feeds the calibration model and D's win-prob curve directly.
Usage
mxo_self_play(
policy_x,
policy_o,
config = mxo_config_default(),
seed = NULL,
record_eval = TRUE
)Arguments
- policy_x, policy_o
Two
mxo_policyobjects (usemxo_policy()).- config
A config list (see
mxo_config_default()).- seed
Optional integer seed for deterministic policy stochasticity.
- record_eval
Logical. If
TRUE(default), recordmxo_evaluateandmxo_win_probfor each ply (X's perspective).
Examples
set.seed(1)
rec <- mxo_self_play(mxo_policy("random"), mxo_policy("random"),
config = mxo_config_default(n = 3L, ply_cap = 8L),
seed = 1L, record_eval = FALSE)
rec$outcome
#> [1] "draw"