DML下的不同因果推断方法

DML + 因果森林

DML + 广义随机森林 = 正交随机森林

代码：因果森林+ DML

from lightgbm import LGBMRegressor
from econml.dml import DML, CausalForestDML

######第一步，训练uplift模型########
dmlmodel = CausalForestDML(
    criterion='mse',
    n_estimators=240,
    max_depth=4,
    min_samples_leaf=2000,
    min_samples_split=2000,
    n_jobs=-1,
    model_y=LGBMRegressor(n_estimators=250, ##
                       max_depth=5,
                       num_leaves=31,
                       learning_rate=0.01,
                       subsample=0.7,
                       min_child_samples=2000,
                       reg_alpha=0.01,
                       reg_lambda=0.01,
                       importance_type='gain'),
     #model_t=RandomPropensityScoreModel(t)，此处可以自定义，但是如果不添加，那么用默认的数据
    , verbose=0
    , discrete_treatment=False
    , honest=True
    , min_var_fraction_leaf=0.1
    , min_var_leaf_on_val=True
)

dmlmodel.fit(Y=y, T=t, X=