GBDT的使用例子

1.目前GBDT的算法比较好的库是xgboost。scikit-learn也可以。我使用的是sklearn需要安装numpy 和 sklearn，安装的时候会有各种包的依赖。建议安装 anaconda3，这样就全部安装好了。安装地址：https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
2.python3的代码如下（和python2差不多）
3.训练速度特别快，100W的训练数据不到5分钟。
# -*- coding:utf-8 -*-

import numpy as np
import codecs
import pickle

from sklearn import ensemble
from sklearn import datasets


# 加载训练数据
def load_train_data():
    
    X = []
    y = []

    with codecs.open("all_train.txt.108W.final.random", 'r', 'utf8') as reader:
        for line in reader:
            tokens = line.strip().split("\t")
            y.append(tokens[3])
            X.append(tokens[4:])
    X = np.array(X)
    y = np.array(y)

    return X, y


# 加载预测数据
def load_pred_data():
    
    ids = []
    inputs = []

    with codecs.open("data/pred.dat", 'r', 'utf8') as reader:
        for line in reader:
            tokens = line.strip().split("\t")
            #print len(tokens)
            if len(tokens) != 27:
                continue
            ids.append(tokens[3])
            inputs.append(tokens[6:])
    ids = np.array(ids)
    inputs = np.array(inputs)

    return ids, inputs


# 保存模型
def save_model(clf):

    s = pickle.dumps(clf)
    f = open("gbdt_classifier.model", "wb+")
    f.write(s)
    f.close()


# 训练GBDT模型
def train_gbdt():

    X, y = load_train_data()
    
    X = X.astype(np.float32)
    labels, y = np.unique(y, return_inverse=True)

    X_train, X_test = X[:90000], X[90000:]
    y_train, y_test = y[:90000], y[90000:]

    print (X[0:1])
    print (y[0:1])
    print ('+++++++++')

    original_params = {'n_estimators': 1000, 'max_leaf_nodes': 4, 'max_depth': None, 'random_state': 2, 'min_samples_split': 5}
    setting = {'learning_rate': 0.1, 'max_features': 2}
    params = dict(original_params)
    params.update(setting)

    clf = ensemble.GradientBoostingClassifier(**params)
    clf.fit(X_train, y_train)

    y_pred = clf.predict(X_test)
    y_predprob = clf.predict_proba(X_test)[:,1]

    print ("Accuracy : %.4g" % clf.score(X_test, y_test))

    save_model(clf)


# 基于GBDT模型预测
def predict_gbdt():

    ids, inputs = load_pred_data()
    
    f = open('model/gbdt_classifier.model','r')
    s = f.read()
    clf = pickle.loads(s)

    f = open('result/pred_classifier.dat','w')

    preds = clf.predict(inputs)
    for i in range(0, len(preds)):
        str_id = str(ids[i])
        str_pred = str(preds[i])
        str_rec = "%s\t%s\n" % (str_id, str_pred)
        f.write(str_rec)
    f.close()


# Start here.
train_gbdt()
#predict_gbdt()