特征选择python代码_features_top = intermediate_features[-1].detach().

 

为什么不做特征之间相关性的动作呢，我理解实际上决策树模型应该可以解决绝大部分此类的工作，但是做一下也不妨，很多人不做是因为代码较为麻烦，还要区分数值型还是字符型。简洁一点也不是什么坏事。

#coding=utf-8

import numpy as np
import pandas as pd

'''单变量特征选取'''
from sklearn.feature_selection import SelectKBest, chi2
'''去除方差小的特征'''
from sklearn.feature_selection import VarianceThreshold
'''循环特征选取'''
from sklearn.svm import SVC
from sklearn.feature_selection import RFE
'''RFE_CV'''
from sklearn.ensemble import ExtraTreesClassifier

class FeatureSelection(object):
    def __init__(self, feature_num):
        self.feature_num = feature_num
        self.train_test, self.label, self.test = self.read_data()    # features #
        self.feature_name = list(self.train_test.columns)     # feature name #

    def read_data(self):
        test = pd.read_csv(r'test_feature.csv', encoding='utf-8')
        train_test = pd.read_csv(r'train_test_feature.csv', encoding='utf-8')
        print('读取数据完毕。。。')
        label = train_test[['target']]
        test = test.iloc[:, 1:]
        train_test = train_test.iloc[:, 2:]
        return train_test, label, test

    def variance_threshold(self):
        sel = VarianceThreshold()
        sel.fit_transform(self.train_test)
        feature_var = list(sel.variances_)    # feature variance #
        features = dict(zip(self.feature_name, feature_var))
        features = list(dict(sorted(features.items(), key=lambda d: d[1])).keys())[-self.feature_num:]
        # print(features)   # 100 cols #
        return set(features)   # return set type #

    def select_k_best(self):
        ch2 = SelectKBest(chi2, k=self.feature_num)
        ch2.fit(self.train_test, self.label)
        feature_var = list(ch2.scores_)  # feature scores #
        features = dict(zip(self.feature_name, feature_var))
        features = list(dict(sorted(features.items(), key=lambda d: d[1])).keys())[-self.feature_num:]
        # print(features)     # 100 cols #
        return set(features)    # return set type #

    def svc_select(self):
        svc = SVC(kernel='rbf', C=1, random_state=2018)    # linear #
        rfe = RFE(estimator=svc, n_features_to_select=self.feature_num, step=1)
        rfe.fit(self.train_test, self.label.ravel())
        print(rfe.ranking_)
        return rfe.ranking_

    def tree_select(self):
        clf = ExtraTreesClassifier(n_estimators=300, max_depth=7, n_jobs=4)
        clf.fit(self.train_test, self.label)
        feature_var = list(clf.feature_importances_)  # feature scores #
        features = dict(zip(self.feature_name, feature_var))
        features = list(dict(sorted(features.items(), key=lambda d: d[1])).keys())[-self.feature_num:]
        # print(features)     # 100 cols #
        return set(features)  # return set type #

        def return_feature_set(self, variance_threshold=False, select_k_best=False, svc_select=False, tree_select=False):
        names = set([])
        if variance_threshold is True:
            name_one = self.variance_threshold()
            names = names.union(name_one)
        if select_k_best is True:
            name_two = self.select_k_best()
            names = names.intersection(name_two)
        if svc_select is True:
            name_three = self.svc_select()
            names = names.intersection(name_three)
        if tree_select is True:
            name_four = self.tree_select()
            names = names.intersection(name_four)

        print(names)
        return list(names)

selection = FeatureSelection(100)
selection.return_feature_set(variance_threshold=True, select_k_best=True, svc_select=False, tree_select=True)