parametric_mutations (macula_tweann v0.18.1)
View SourceParametric mutation operators for neural network evolution.
This module provides mutations that modify network parameters without changing structure: - mutate_weights: Perturb synaptic weights - mutate_af: Change activation function - mutate_aggr_f: Change aggregation function
Also includes evolutionary strategy mutations: - mutate_tuning_selection: Change weight selection strategy - mutate_annealing: Modify simulated annealing schedule - mutate_heredity_type: Switch between darwinian/lamarckian
Summary
Functions
Mutate activation function of a random neuron.
Generic function to mutate an agent parameter.
Mutate aggregation function of a random neuron.
Mutate heredity type (darwinian/lamarckian).
Mutate total topological mutations function.
Mutate annealing parameter.
Mutate tuning selection function.
Mutate weights of a random neuron.
Functions
Mutate activation function of a random neuron.
Selects a random neuron and changes its activation function to another available function from the constraint.
Generic function to mutate an agent parameter.
Reads the current value of a field, gets alternatives from constraint, and selects a new random value.
Mutate aggregation function of a random neuron.
-spec mutate_heredity_type(term()) -> ok | {error, no_alternatives}.
Mutate heredity type (darwinian/lamarckian).
-spec mutate_tot_topological_mutations(term()) -> ok | {error, no_alternatives}.
Mutate total topological mutations function.
-spec mutate_tuning_annealing(term()) -> ok | {error, no_alternatives}.
Mutate annealing parameter.
-spec mutate_tuning_selection(term()) -> ok | {error, no_alternatives}.
Mutate tuning selection function.
-spec mutate_weights(term()) -> ok.
Mutate weights of a random neuron.
Selects a random neuron and perturbs its input weights using the agent's perturbation range.