pub struct TrainingConfig {
pub use_reward_modulation: bool,
}Expand description
Configuration knobs for crate::PlasticityTrainer.
Every field here drives an explicit runtime code path in
crate::trainer::PlasticityTrainer::train_step — see that method’s rustdoc
for exactly how each field is consumed. TrainingConfig intentionally does
not expose knobs without a consumer (learning rate, homeostasis setpoints,
and batch size were removed for this reason; see CHANGELOG.md). Low-level
STDP / homeostasis tuning lives in neuromod::SpikingNetwork, which derives
its own learning rate and thresholds from the neuromodulator state passed
into step.
Values are serializable (serde) so callers can persist them as part of their
own experiment configs or checkpointing setup — this crate does not implement
checkpointing itself. Missing fields deserialize via Default
(#[serde(default)] on the struct), and unknown fields (for example from an
older config that still carries a since-removed knob) are ignored rather
than rejected, since the struct does not use deny_unknown_fields.
Fields§
§use_reward_modulation: boolWhen true (default), train_step adjusts neuromodulators from the reward.
When false, the network steps with its current modulators unchanged.
Trait Implementations§
Source§impl Clone for TrainingConfig
impl Clone for TrainingConfig
Source§fn clone(&self) -> TrainingConfig
fn clone(&self) -> TrainingConfig
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more