openfe的使用

文章和代码:https://www.baidu.com/link?url=nz86RZmDqSVtQ85oRCyiQw1NLINAy4c65DoIC-uFJ7jwrM_H7qX9cQHZxltjl7RW&wd=&eqid=f8a55867001e79ac000000036394620b

直接把数据丢里提升效果不太大。。。。。。。他会生成很多特征导致计算机运算需要很长时间，因此不建议放入较大的特征维度，需要自己选择一下。

import pandas as pd

from sklearn.datasets import fetch_california_housing

from openfe import openfe, transform, tree_to_formula

from sklearn.model_selection import train_test_split

import lightgbm as lgb

from sklearn.metrics import mean_squared_error

from sklearn.model_selection import cross_val_score

from sklearn import tree

import warnings

warnings.filterwarnings("ignore")

def get_score(train_x, test_x, train_y, test_y):

    train_x, val_x, train_y, val_y = train_test_split(train_x, train_y, test_size=0.2, random_state=1)

    params = {'n_estimators': 1000, 'n_jobs': n_jobs, 'seed': 1}

    gbm = lgb.LGBMRegressor(**params)

    gbm.fit(train_x, train_y, eval_set=[(val_x, val_y)], callbacks=[lgb.early_stopping(50, verbose=False)])

    pred = pd.DataFrame(gbm.predict(test_x), index=test_x.index)

    score = mean_squared_error(test_y, pred)

    return score

import numpy as np

import warnings

if __name__ == '__main__':

    df = pd.read_csv('train.csv')

    df=df.drop(['ID'],axis=1)

    df=df.to_numpy()

    feature=np.abs(np.fft.fft(df[:,:-1]))

    feature=np.concatenate((feature,np.reshape(df[:,-1],(-1,1))),axis=1)

    train=pd.DataFrame(feature)

    heat=train.corr()

    fe=heat.index[abs(heat[240])>0.3]

    train=df[:,fe]

    label=train[:,-1]

    train=train[:,:-1]

    # feature generation

    ofe = openfe()

    w=[]

    for i in range(train.shape[1]):

        s='T'

        w.append(s+str(i))

    train=pd.DataFrame(train)  

    train.columns=w

   

    df1 = pd.read_csv('test.csv')

    df1=df1.drop(['ID'],axis=1)

    df1=df1.to_numpy()

    test_feature=np.abs(np.fft.fft(df1[:,:]))

    test_feature=test_feature[:,fe[:-1]]

    features = ofe.fit(data=pd.DataFrame(train),task='classification', label=pd.DataFrame(label), n_jobs=3, stage1_metric='corr',stage2_metric='permutation')

    test_fe=pd.DataFrame(test_feature)

    test_fe.columns=w

    train_x, test_x = transform(train, test_fe, ofe.new_features_list[:10], n_jobs=3)

    print(train_x.shape)

    print(train.shape)

    for i in range(10):

        clf=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=i+1)

        clf1=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=i+1)

        scores = cross_val_score(clf, train, label, cv=5)

        scores1 = cross_val_score(clf1, train_x, label, cv=5)

        print(i,scores.mean(),scores1.mean())    

    print("The top 10 generated features are")

    clf1=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=3+1)

    clf1.fit(train_x,label)

    out=clf1.predict(train_x)

    print(out)

    out=clf1.predict(test_x)

    print(out)

    out=pd.DataFrame(out)

    out.columns = ['CLASS']

    w=[]

    for k in range(out.shape[0]):

        w.append(k+210)

    out['ID']=np.reshape(w,(-1,1))

    out[['ID','CLASS']].to_csv('out3.csv',index=False)