sklearn pipeline 实现多个模型统一调参

实现多模型统一调参

解决问题：在复现GBDT+LR的经典结构的时候，发现需要对两个模型一起进行调参，网上找不到相关代码，研究之后实现LGB + LR的统一调参

需写3个自定义管道流的类来完成, 两个模型用于预测, 一个实现将GBDT的预测值作为下一步LR的特征的转换

from sklearn.base import BaseEstimator, TransformerMixin
from lightgbm import LGBMClassifier
from sklearn.linear_model import LogisticRegression

class LgbmPipeline(BaseEstimator, TransformerMixin):
	# 传入lgb的模型参数, 详细见下方调用
    def __init__(self, other_params):
        self.other_params = other_params

    def fit(self, X, y=None):
        lgb = LGBMClassifier(**self.other_params)
        lgb.fit(X, y)
        self.lgb = lgb
        return self

    def transform(self, X):
        lgb_feature = self.lgb.predict(X, pred_leaf=True)
        return lgb_feature

class GbdtMatrixPipeline(BaseEstimator, TransformerMixin):
    # 传入叶子数量
    def __init__(self, num_leaf):
        self.num_leaf = num_leaf

    def fit(self, X, y=None):
        return self

    def transform(self, X):
        transformed_matrix = np.zeros([len(X), len(X[0]) * self.num_leaf],
                            dtype=np.int8)  # N * num_tress * num_leafs

        for i in range(0, len(X)):
            temp = np.arange(len(X[0])) * self.num_leaf + np.array(X[i])#计算onehot在100 * 64列当中的位置 arange(100)*64 + y_pred[i]
            transformed_matrix[i][temp] += 1

        return transformed_matrix

class LrPipeline(BaseEstimator, TransformerMixin):
    def __init__(self, other_params):
        self.other_params = other_params

    def fit(self, X, y=None):
        lr = LogisticRegression(**self.other_params)
        lr.fit(X, y)
        self.lr = lr
        return self

    def transform(self, X):
        y_pred_scores = self.lr.predict_proba(X)
        return y_pred_scores

管道流实现fit与调用, X_train_tr为ndarry格式的数据

from sklearn.pipeline import Pipeline
other_params = {
    'task': 'train',
    'boosting_type': 'gbdt',  # GBDT算法为基础
    'objective': 'binary',
    'metric': 'auc',  # 评判指标
    'max_bin': 255,  # 大会有更准的效果,更慢的速度
    'learning_rate': 0.1,  # 学习率
    'num_leaves': 32,  # 大会更准,但可能过拟合
    'max_depth': -1,  # 小数据集下限制最大深度可防止过拟合,小于0表示无限制
    'feature_fraction': 0.8,  # 防止过拟合
    'bagging_freq': 5,  # 防止过拟合
    'bagging_fraction': 0.8,  # 防止过拟合
    'min_data_in_leaf': 21,  # 防止过拟合
    'min_sum_hessian_in_leaf': 3.0,  # 防止过拟合
    'min_child_weight': 0.1,
    'lambda_l1': 0.2,
    'lambda_l2': 20,
    'is_unbalance': True,
    'n_estimators': 100
}

Lr_params = {
    'class_weight':dict({0: 1, 1:8}),
    'penalty': 'l2'
}
model_pipeline = Pipeline([
    ("lgb", LgbmPipeline(other_params)),
    ("matrix", GbdtMatrixPipeline(other_params['num_leaves'])),
    ("lr", LrPipeline(Lr_params))
])
# X_train_tr为ndarry格式的数据
model_pipeline.fit(X_train_tr, y_train)
model_pipeline.transform(X_test_tr)

对pipeline进行网格调参步骤详见：https://sklearn.apachecn.org/docs/master/38.html