fix: count_bonus_coefficient + cql_alpha from_continuous defaults match Default

Automated mismatch scan confirmed all from_continuous fixed defaults now
match Default::default(). No more silent divergence between hyperopt
params and manual/test params.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-03-27 12:07:20 +01:00
parent 5e4fc2bcb1
commit 9a7db1bbff

View File

@@ -662,7 +662,7 @@ impl ParameterSpace for DQNParams {
norm_type: 1.0,
activation_type: 1.0,
noisy_epsilon_floor: 0.1,
count_bonus_coefficient: 0.2,
count_bonus_coefficient: 0.0, // match Default::default() — conflicts with NoisyNet
cql_alpha: 0.1, // match Default::default()
eval_softmax_temp: 1.0,
dsr_eta: 0.004,
@@ -4320,8 +4320,8 @@ mod tests {
// Verify fixed exploration params
assert!((params.curiosity_weight - 0.0).abs() < f64::EPSILON, "curiosity_weight should be 0.0");
assert!((params.noisy_epsilon_floor - 0.10).abs() < 1e-6, "noisy_epsilon_floor should be 0.10");
// count_bonus_coefficient is now fixed at 0.2
assert!((params.count_bonus_coefficient - 0.2).abs() < 1e-6, "count_bonus_coefficient should be fixed to 0.2, got {}", params.count_bonus_coefficient);
// count_bonus_coefficient fixed at 0.0 (conflicts with NoisyNet)
assert!((params.count_bonus_coefficient - 0.0).abs() < f64::EPSILON, "count_bonus_coefficient should be 0.0, got {}", params.count_bonus_coefficient);
}
#[test]