| """Genetic operators: selection, crossover, mutation.""" |
|
|
| from __future__ import annotations |
|
|
| import random |
|
|
| from factor_engine.gp.operators import ( |
| AbsOp, |
| Neg, |
| RankCS, |
| TsDecayLinear, |
| TsEMA, |
| TsMean, |
| TsRank, |
| TsSlope, |
| TsStd, |
| TsZScore, |
| ZScoreCS, |
| choose_binary_operator, |
| generate_random_tree, |
| make_binary, |
| make_unary, |
| ) |
|
|
|
|
| def tree_too_large(tree, max_depth, max_nodes): |
| return tree.get_depth() > max_depth or tree.get_size() > max_nodes |
|
|
|
|
| def tournament_selection(population, fitnesses, k=5): |
| idxs = random.sample(range(len(population)), k) |
| best_idx = max(idxs, key=lambda i: fitnesses[i]) |
| return population[best_idx].clone() |
|
|
|
|
| def replace_random_subtree(tree, new_subtree): |
| tree = tree.clone() |
| nodes = tree.get_nodes() |
| if len(nodes) <= 1: |
| return new_subtree.clone() |
| target = random.choice(nodes) |
| if target is tree: |
| return new_subtree.clone() |
| for node in nodes: |
| if hasattr(node, "children"): |
| for i, child in enumerate(node.children): |
| if child is target: |
| node.children[i] = new_subtree.clone() |
| return tree |
| return tree |
|
|
|
|
| def mutate(tree, max_init_depth): |
| if random.random() < 0.55: |
| new_subtree = generate_random_tree(1, max_depth=random.randint(2, max(2, max_init_depth - 1))) |
| return replace_random_subtree(tree, new_subtree) |
| if random.random() < 0.85: |
| child = tree.clone() |
| op = random.choice([AbsOp, Neg, RankCS, ZScoreCS, TsMean, TsStd, TsZScore, TsRank, TsDecayLinear, TsEMA, TsSlope]) |
| return make_unary(op, child) |
| return make_binary(choose_binary_operator(), tree.clone(), generate_random_tree(1, max_depth=3)) |
|
|
|
|
| def crossover(parent1, parent2): |
| p1 = parent1.clone() |
| donor = random.choice(parent2.clone().get_nodes()).clone() |
| return replace_random_subtree(p1, donor) |
|
|
