sklearn-GDBT

GDBT 可以解决分类和回归问题

 

回归问题

def __init__(self, loss='ls', learning_rate=0.1, n_estimators=100,
                 subsample=1.0, criterion='friedman_mse', min_samples_split=2,
                 min_samples_leaf=1, min_weight_fraction_leaf=0.,
                 max_depth=3, min_impurity_decrease=0.,
                 min_impurity_split=None, init=None, random_state=None,
                 max_features=None, alpha=0.9, verbose=0, max_leaf_nodes=None,
                 warm_start=False, presort='auto')

 

示例

import numpy as np
from sklearn.metrics import mean_squared_error
from sklearn.datasets import make_friedman1
from sklearn.ensemble import GradientBoostingRegressor

X, y = make_friedman1(n_samples=1200, random_state=0, noise=1.0)
X_train, X_test = X[:200], X[200:]
y_train, y_test = y[:200], y[200:]

### 损失函数
# 如果损失函数为 误差绝对值，L=|y-f(x)|，负梯度为 sign(y-f(x)),即要么1，要么-1，sklearn 中对应为 loss='lad'
# 如果损失函数为 huber，sklearn 中对应为 loss='huber'
# 如果损失函数为 均方误差，sklearn 中对应为 loss='ls'
est = GradientBoostingRegressor(n_estimators=100, learning_rate=0.1, max_depth=1, random_state=0, loss='huber').fit(X_train, y_train)

pred = est.predict(X_test)
error = mean_squared_error(pred, y_test)

print(max(y_test), min(y_test)) # (27.214332670044374, 0.8719243023544349)
print(error)
# loss='ls' 5.009154859960321
# loss='lad' 5.817510629608294
# loss='huber' 4.690823542377095

 

分类问题

def __init__(self, loss='deviance', learning_rate=0.1, n_estimators=100,
                 subsample=1.0, criterion='friedman_mse', min_samples_split=2,
                 min_samples_leaf=1, min_weight_fraction_leaf=0.,
                 max_depth=3, min_impurity_decrease=0.,
                 min_impurity_split=None, init=None,
                 random_state=None, max_features=None, verbose=0,
                 max_leaf_nodes=None, warm_start=False,
                 presort='auto')

 

示例

from sklearn.ensemble import GradientBoostingClassifier, GradientBoostingRegressor
from sklearn.model_selection import GridSearchCV, train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import accuracy_score, mean_squared_error
from time import time
import numpy as np
import pandas as pd
import mnist


if __name__ == "__main__":
    # 读取Mnist数据集, 测试GBDT的分类模型
    mnistSet = mnist.loadLecunMnistSet()
    train_X, train_Y, test_X, test_Y = mnistSet[0], mnistSet[1], mnistSet[2], mnistSet[3]

    m, n = np.shape(train_X)
    idx = range(m)
    np.random.shuffle(idx)

    # 使用PCA降维
    # num = 30000
    # pca = PCA(n_components=0.9, whiten=True, random_state=0)
    # for i in range(int(np.ceil(1.0 * m / num))):
    #     minEnd = min((i + 1) * num, m)
    #     sub_idx = idx[i * num:minEnd]
    #     train_pca_X = pca.fit_transform(train_X[sub_idx])
    #     print np.shape(train_pca_X)

    print "\n**********测试GradientBoostingClassifier类**********"
    t = time()
    # param_grid1 = {"n_estimators": range(1000, 2001, 100)}
    # param_grid2 = {'max_depth': range(30, 71, 10), 'min_samples_split': range(4, 9, 2)}
    # param_grid3 = {'min_samples_split': range(4, 9, 2), 'min_samples_leaf': range(3, 12, 2)}
    # param_grid4 = {'subsample': np.arange(0.6, 1.0, 0.05)}
    # model = GridSearchCV(
    #     estimator=GradientBoostingClassifier(max_features=90, max_depth=40, min_samples_split=8, learning_rate=0.1,
    #                                          n_estimators=1800),
    #     param_grid=param_grid4, cv=3)
    # # 拟合训练数据集
    # model.fit(train_X, train_Y)
    # print "最好的参数是:%s, 此时的得分是:%0.2f" % (model.best_params_, model.best_score_)
    model = GradientBoostingClassifier(max_features=90, max_depth=40, min_samples_split=8, min_samples_leaf=3,
                                       n_estimators=1200, learning_rate=0.05, subsample=0.95)
    # 拟合训练数据集
    model.fit(train_X, train_Y)
    # 预测训练集
    train_Y_hat = model.predict(train_X[idx])
    print "训练集精确度: ", accuracy_score(train_Y[idx], train_Y_hat)
    # 预测测试集
    test_Y_hat = model.predict(test_X)
    print "测试集精确度: ", accuracy_score(test_Y, test_Y_hat)
    print "总耗时:", time() - t, "秒"

 

 

参考资料：

https://github.com/haidawyl/Mnist　　各种模型的用法