使用xgboost進行特徵選擇

xgboost在各大資料探勘比賽中是一個大殺器，往往可以取得比其他各種機器學習演算法更好的效果。資料預處理，特徵工程，調參對Xgboost的效果有著非常重要的影響。這裡介紹一下運用xgboost的特徵選擇，運用xgboost的特徵選擇可以篩選出更加有效的特徵代入Xgboost模型。

這裡採用的資料集來自於Kaggle | Allstate Claims Severity比賽(https://www.kaggle.com/c/allstate-claims-severity/data)，這裡的訓練集如下所示，有116個離散特徵（cat1-cat116）,14個連續特徵（cont1 -cont14），離散特徵用字串表示，先要對其進行數值化：

   id cat1 cat2 cat3 cat4 cat5 cat6 cat7 cat8 cat9   ...        cont6  \
0   1    A    B    A    B    A    A    A    A    B   ...     0.718367   
1   2    A    B    A    A    A    A    A    A    B   ...     0.438917   
2   5    A    B    A    A    B    A    A    A    B   ...     0.289648   
3  10    B    B    A    B    A    A    A    A    B   ...     0.440945   
4  11    A    B    A    B    A    A    A    A    B   ...     0.178193   

      cont7    cont8    cont9   cont10    cont11    cont12    cont13  \
0  0.335060  0.30260  0.67135  0.83510  0.569745  0.594646  0.822493   
1  0.436585  0.60087  0.35127  0.43919  0.338312  0.366307  0.611431   
2  0.315545  0.27320  0.26076  0.32446  0.381398  0.373424  0.195709   
3  0.391128  0.31796  0.32128  0.44467  0.327915  0.321570  0.605077   
4  0.247408  0.24564  0.22089  0.21230  0.204687  0.202213  0.246011

xgboost的特徵選擇的程式碼如下：

import numpy as np
import pandas as pd
import xgboost as xgb
import operator
import matplotlib.pyplot as plt

def ceate_feature_map(features):
    outfile = open('xgb.fmap', 'w')
    i = 0
    for feat in features:
        outfile.write('{0}\t{1}\tq\n'.format(i, feat))
        i = i + 1
    outfile.close()


if __name__ == '__main__':
    train = pd.read_csv("../input/train.csv")
    cat_sel = [n for n in train.columns if n.startswith('cat')]  #類別特徵數值化
    for column in cat_sel:
        train[column] = pd.factorize(train[column].values , sort=True)[0] + 1

    params = {
        'min_child_weight': 100,
        'eta': 0.02,
        'colsample_bytree': 0.7,
        'max_depth': 12,
        'subsample': 0.7,
        'alpha': 1,
        'gamma': 1,
        'silent': 1,
        'verbose_eval': True,
        'seed': 12
    }
    rounds = 10
    y = train['loss']
    X = train.drop(['loss', 'id'], 1)

    xgtrain = xgb.DMatrix(X, label=y)
    bst = xgb.train(params, xgtrain, num_boost_round=rounds)

    features = [x for x in train.columns if x not in ['id','loss']]
    ceate_feature_map(features)

    importance = bst.get_fscore(fmap='xgb.fmap')
    importance = sorted(importance.items(), key=operator.itemgetter(1))

    df = pd.DataFrame(importance, columns=['feature', 'fscore'])
    df['fscore'] = df['fscore'] / df['fscore'].sum()
    df.to_csv("../input/feat_sel/feat_importance.csv", index=False)

    plt.figure()
    df.plot(kind='barh', x='feature', y='fscore', legend=False, figsize=(6, 10))
    plt.title('XGBoost Feature Importance')
    plt.xlabel('relative importance')
    plt.show()

參考:
http://blog.csdn.net/q383700092/article/details/53698760
http://blog.csdn.net/qq_34264472/article/details/53363384