xgb, lgb, Keras, LR(二分类、多分类代码)

preprocess

# 通用的预处理框架

import pandas as pd

import numpy as np

import scipy as sp

# 文件读取

def read_csv_file(f, logging=False):

    print("==========读取数据=========")

    data =  pd.read_csv(f)

    if logging:

        print(data.head(5))

        print(f, "包含以下列")

        print(data.columns.values)

        print(data.describe())

        print(data.info())

    return data

Logistic Regression

# 通用的LogisticRegression框架

import pandas as pd

import numpy as np

from scipy import sparse

from sklearn.preprocessing import OneHotEncoder

from sklearn.linear_model import LogisticRegression

from sklearn.preprocessing import StandardScaler

# 1. load data

df_train = pd.DataFrame()

df_test  = pd.DataFrame()

y_train = df_train['label'].values

# 2. process data

ss = StandardScaler()

# 3. feature engineering/encoding

# 3.1 For Labeled Feature

enc = OneHotEncoder()

feats = ["creativeID", "adID", "campaignID"]

for i, feat in enumerate(feats):

    x_train = enc.fit_transform(df_train[feat].values.reshape(-1, 1))

    x_test = enc.fit_transform(df_test[feat].values.reshape(-1, 1))

    if i == 0:

        X_train, X_test = x_train, x_test

    else:

        X_train, X_test = sparse.hstack((X_train, x_train)), sparse.hstack((X_test, x_test))

# 3.2 For Numerical Feature

# It must be a 2-D Data for StandardScalar, otherwise reshape(-1, len(feats)) is required

feats = ["price", "age"]

x_train = ss.fit_transform(df_train[feats].values)

x_test  = ss.fit_transform(df_test[feats].values)

X_train, X_test = sparse.hstack((X_train, x_train)), sparse.hstack((X_test, x_test))

# model training

lr = LogisticRegression()

lr.fit(X_train, y_train)

proba_test = lr.predict_proba(X_test)[:, 1]

LightGBM

1. 二分类

import lightgbm as lgb

import pandas as pd

import numpy as np

import pickle

from sklearn.metrics import roc_auc_score

from sklearn.model_selection import train_test_split

print("Loading Data ... ")

# 导入数据

train_x, train_y, test_x = load_data()

# 用sklearn.cross_validation进行训练数据集划分，这里训练集和交叉验证集比例为7：3，可以自己根据需要设置

X, val_X, y, val_y = train_test_split(

    train_x,

    train_y,

    test_size=0.05,

    random_state=1,

    stratify=train_y ## 这里保证分割后y的比例分布与原数据一致

)

X_train = X

y_train = y

X_test = val_X

y_test = val_y

# create dataset for lightgbm

lgb_train = lgb.Dataset(X_train, y_train)

lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)

# specify your configurations as a dict

params = {

    'boosting_type': 'gbdt',

    'objective': 'binary',

    'metric': {'binary_logloss', 'auc'},

    'num_leaves': 5,

    'max_depth': 6,

    'min_data_in_leaf': 450,

    'learning_rate': 0.1,

    'feature_fraction': 0.9,

    'bagging_fraction': 0.95,

    'bagging_freq': 5,

    'lambda_l1': 1,

    'lambda_l2': 0.001,  # 越小l2正则程度越高

    'min_gain_to_split': 0.2,

    'verbose': 5,

    'is_unbalance': True

}

# train

print('Start training...')

gbm = lgb.train(params,

                lgb_train,

                num_boost_round=10000,

                valid_sets=lgb_eval,

                early_stopping_rounds=500)

print('Start predicting...')

preds = gbm.predict(test_x, num_iteration=gbm.best_iteration)  # 输出的是概率结果

# 导出结果

threshold = 0.5

for pred in preds:

    result = 1 if pred > threshold else 0

# 导出特征重要性

importance = gbm.feature_importance()

names = gbm.feature_name()

with open('./feature_importance.txt', 'w+') as file:

    for index, im in enumerate(importance):

        string = names[index] + ', ' + str(im) + '\n'

        file.write(string)

2.多分类

import lightgbm as lgb

import pandas as pd

import numpy as np

import pickle

from sklearn.metrics import roc_auc_score

from sklearn.model_selection import train_test_split

print("Loading Data ... ")

# 导入数据

train_x, train_y, test_x = load_data()

# 用sklearn.cross_validation进行训练数据集划分，这里训练集和交叉验证集比例为7：3，可以自己根据需要设置

X, val_X, y, val_y = train_test_split(

    train_x,

    train_y,

    test_size=0.05,

    random_state=1,

    stratify=train_y ## 这里保证分割后y的比例分布与原数据一致

)

X_train = X

y_train = y

X_test = val_X

y_test = val_y

# create dataset for lightgbm

lgb_train = lgb.Dataset(X_train, y_train)

lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)

# specify your configurations as a dict

params = {

    'boosting_type': 'gbdt',

    'objective': 'multiclass',

    'num_class': 9,

    'metric': 'multi_error',

    'num_leaves': 300,

    'min_data_in_leaf': 100,

    'learning_rate': 0.01,

    'feature_fraction': 0.8,

    'bagging_fraction': 0.8,

    'bagging_freq': 5,

    'lambda_l1': 0.4,

    'lambda_l2': 0.5,

    'min_gain_to_split': 0.2,

    'verbose': 5,

    'is_unbalance': True

}

# train

print('Start training...')

gbm = lgb.train(params,

                lgb_train,

                num_boost_round=10000,

                valid_sets=lgb_eval,

                early_stopping_rounds=500)

print('Start predicting...')

preds = gbm.predict(test_x, num_iteration=gbm.best_iteration)  # 输出的是概率结果

# 导出结果

for pred in preds:

    result = prediction = int(np.argmax(pred))

# 导出特征重要性

importance = gbm.feature_importance()

names = gbm.feature_name()

with open('./feature_importance.txt', 'w+') as file:

    for index, im in enumerate(importance):

        string = names[index] + ', ' + str(im) + '\n'

        file.write(string)

XGBoost

1. 二分类

import numpy as np

import pandas as pd

import xgboost as xgb

import time

from sklearn.model_selection import StratifiedKFold

from sklearn.model_selection import train_test_split

train_x, train_y, test_x = load_data()

# 构建特征

# 用sklearn.cross_validation进行训练数据集划分，这里训练集和交叉验证集比例为7：3，可以自己根据需要设置

X, val_X, y, val_y = train_test_split(

    train_x,

    train_y,

    test_size=0.01,

    random_state=1,

    stratify=train_y

)

# xgb矩阵赋值

xgb_val = xgb.DMatrix(val_X, label=val_y)

xgb_train = xgb.DMatrix(X, label=y)

xgb_test = xgb.DMatrix(test_x)

# xgboost模型 #####################

params = {

    'booster': 'gbtree',

    # 'objective': 'multi:softmax',  # 多分类的问题、

    # 'objective': 'multi:softprob',   # 多分类概率

    'objective': 'binary:logistic',

    'eval_metric': 'logloss',

    # 'num_class': 9,  # 类别数，与 multisoftmax 并用

    'gamma': 0.1,  # 用于控制是否后剪枝的参数,越大越保守，一般0.1、0.2这样子。

    'max_depth': 8,  # 构建树的深度，越大越容易过拟合

    'alpha': 0,   # L1正则化系数

    'lambda': 10,  # 控制模型复杂度的权重值的L2正则化项参数，参数越大，模型越不容易过拟合。

    'subsample': 0.7,  # 随机采样训练样本

    'colsample_bytree': 0.5,  # 生成树时进行的列采样

    'min_child_weight': 3,

    # 这个参数默认是 1，是每个叶子里面 h 的和至少是多少，对正负样本不均衡时的 0-1 分类而言

    # ，假设 h 在 0.01 附近，min_child_weight 为 1 意味着叶子节点中最少需要包含 100 个样本。

    # 这个参数非常影响结果，控制叶子节点中二阶导的和的最小值，该参数值越小，越容易 overfitting。

    'silent': 0,  # 设置成1则没有运行信息输出，最好是设置为0.

    'eta': 0.03,  # 如同学习率

    'seed': 1000,

    'nthread': -1,  # cpu 线程数

    'missing': 1,

    'scale_pos_weight': (np.sum(y==0)/np.sum(y==1))  # 用来处理正负样本不均衡的问题,通常取：sum(negative cases) / sum(positive cases)

    # 'eval_metric': 'auc'

}

plst = list(params.items())

num_rounds = 2000  # 迭代次数

watchlist = [(xgb_train, 'train'), (xgb_val, 'val')]

# 交叉验证

result = xgb.cv(plst, xgb_train, num_boost_round=200, nfold=4, early_stopping_rounds=200, verbose_eval=True, folds=StratifiedKFold(n_splits=4).split(X, y))

# 训练模型并保存

# early_stopping_rounds 当设置的迭代次数较大时，early_stopping_rounds 可在一定的迭代次数内准确率没有提升就停止训练

model = xgb.train(plst, xgb_train, num_rounds, watchlist, early_stopping_rounds=200)

model.save_model('../data/model/xgb.model')  # 用于存储训练出的模型

preds = model.predict(xgb_test)

# 导出结果

threshold = 0.5

for pred in preds:

    result = 1 if pred > threshold else 0

处理正负样本不均匀的案例

# 计算正负样本比例

positive_num = df_train[df_train['label']==1].values.shape[0]

negative_num = df_train[df_train['label']==0].values.shape[0]

print(float(positive_num)/float(negative_num))

主要思路

手动调整正负样本比例
过采样 Over-Sampling

对训练集里面样本数量较少的类别（少数类）进行过采样，合成新的样本来缓解类不平衡，比如SMOTE算法
欠采样 Under-Sampling
将样本按比例一一组合进行训练，训练出多个弱分类器，最后进行集成

巴特西

xgb, lgb, Keras, LR(二分类、多分类代码)

preprocess

Logistic Regression

LightGBM

1. 二分类

2.多分类

XGBoost

1. 二分类

处理正负样本不均匀的案例

主要思路

最新文章

热门文章