遗传算法优化lightgbm的代码

遗传算法是一种优化算法，可以用来优化机器学习模型的超参数，如lightgbm的参数。下面是一个使用遗传算法优化lightgbm参数的示例代码：

import lightgbm as lgb
import numpy as np
import random

# 定义参数空间
param_space = {
    'boosting_type': ['gbdt', 'dart', 'goss'],
    'num_leaves': list(range(20, 150)),
    'max_depth': list(range(3, 15)),
    'learning_rate': [0.01, 0.05, 0.1, 0.15, 0.2],
    'n_estimators': [50, 100, 200, 300, 400, 500],
    'min_child_samples': list(range(5, 50)),
    'subsample': [0.6, 0.7, 0.8, 0.9, 1],
    'colsample_bytree': [0.6, 0.7, 0.8, 0.9, 1]
}

# 定义遗传算法参数
POP_SIZE = 20  # 种群大小
CROSS_RATE = 0.8  # 交叉概率
MUTATION_RATE = 0.2  # 变异概率
N_GENERATIONS = 10  # 迭代次数

# 定义适应度函数
def fitness(estimator, X_train, y_train, X_valid, y_valid):
    estimator.fit(X_train, y_train, eval_set=[(X_valid, y_valid)], early_stopping_rounds=50, verbose=False)
    score = estimator.best_score_['valid_0']['rmse']
    return score

# 初始化种群
pop = []
for i in range(POP_SIZE):
    param = {}
    for key in param_space:
        param[key] = random.choice(param_space[key])
    pop.append(param)

# 进化
for i in range(N_GENERATIONS):
    # 计算适应度
    scores = []
    for param in pop:
        estimator = lgb.LGBMRegressor(**param)
        score = fitness(estimator, X_train, y_train, X_valid, y_valid)
        scores.append(score)
    # 选择
    indices = np.argsort(scores)
    pop = [pop[i] for i in indices[:int(POP_SIZE/2)]]
    # 交叉
    for i in range(int(POP_SIZE/2)):
        if np.random.rand() < CROSS_RATE:
            parent1 = pop[i]
            parent2 = pop[np.random.randint(int(POP_SIZE/2), POP_SIZE)]
            child1 = {}
            child2 = {}
            for key in param_space:
                if np.random.rand() < 0.5:
                    child1[key] = parent1[key]
                    child2[key] = parent2[key]
                else:
                    child1[key] = parent2[key]
                    child2[key] = parent1[key]
            pop.append(child1)
            pop.append(child2)
    # 变异
    for i in range(POP_SIZE):
        if np.random.rand() < MUTATION_RATE:
            param = pop[i]
            key = random.choice(list(param_space.keys()))
            param[key] = random.choice(param_space[key])

# 计算最优参数
scores = []
params = []
for param in pop:
    estimator = lgb.LGBMRegressor(**param)
    score = fitness(estimator, X_train, y_train, X_valid, y_valid)
    scores.append(score)
    params.append(param)
best_param = params[np.argmin(scores)]
print('Best params:', best_param)

在这个示例中，我们使用lightgbm建立回归模型，并使用RMSE作为适应度函数。我们定义了一个参数空间，其中包含了所有可能的参数取值。然后，我们使用遗传算法来搜索最优参数。我们初始化一个种群，然后迭代若干次，每次选择适应度最高的一半个体进行交叉和变异，得到新一代种群，最后计算最优参数。