达观杯_构建模型（四）贝叶斯

"""
1.特征：linearsvm-tfidf(word)+lr-tfidf(article) / doc2vec_word
2.模型：bayes"""
from sklearn.calibration import CalibratedClassifierCV
from sklearn.model_selection import StratifiedKFold
import pandas as pd
import pickle
import time
import random
from scipy.stats import randint as sp_randint
from sklearn.naive_bayes import MultinomialNB,BernoulliNB,BaseDiscreteNB
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import RandomizedSearchCV
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.ensemble import RandomForestClassifier
import numpy as np
import scipydef get_time_diff(start_time,str):print("%s 用时:%f min" % (str,(time.time() - start_time)/60))return time.time() - start_timeDATA_PATH = '..\达观杯_特征\\feature_word\Doc2vec_word\Doc2vec_word\data_doc2vec.pkl'
SAVE_PATH = './models/bayes/HashingVectorizer_word'def read_data(path):with open(path, 'rb') as f:x_train, y_train, x_test = pickle.load(f)X_train, X_val, Y_train, Y_val = train_test_split(x_train, y_train, test_size=0.25)return X_train, X_val, Y_train, Y_val, x_testdef model_search_random(X_train, Y_train, param_dist, n_iter_search = 20,clf = MultinomialNB()):random_search = RandomizedSearchCV(clf, param_distributions=param_dist, n_iter=n_iter_search)random_search.fit(X_train, Y_train)return random_searchdef model_search_grid(X_train, Y_train, param_dist,clf = MultinomialNB()):random_search = GridSearchCV(estimator=clf,param_grid=param_dist)random_search.fit(X_train, Y_train)return random_searchdef model_best(model_search):print('随机搜索-度量记录：', model_search.cv_results_)  # 包含每次训练的相关信息print('随机搜索-最佳度量值:', model_search.best_score_)  # 获取最佳度量值print('随机搜索-最佳参数：', model_search.best_params_)  # 获取最佳度量值时的代定参数的值。是一个字典print('随机搜索-最佳模型：', model_search.best_estimator_)  # 获取最佳度量时的分类器模型return model_search.cv_results_, model_search.best_score_,model_search.best_params_,model_search.best_estimator_def predict(X_val,Y_val,x_test,model):Y_val_test = model.predict(X_val)score = accuracy_score(Y_val, Y_val_test)y_test = model.predict(x_test)y_test = [i + 1 for i in y_test.tolist()]df_result = pd.DataFrame({'id': range(102277), 'class': y_test})y_test_proba = model.predict_proba(x_test)y_test_proba = y_test_proba.tolist()df_proba = pd.DataFrame({'id': range(102277), 'proba': y_test_proba})return df_result,df_proba,scoredef save(score,path_save,df_result,df_proba,model,alpha = None):if alpha is not None:class_name = '/bayes_%0.2f_%s_class_%0.4f.csv' % (alpha,"data_tf_word",score)proba_name = '/bayes_%0.2f_%s_proba_%0.4f.csv' % (alpha,"data_tf_word",score)model_name = '/bayes_%0.2f_%s_model_%0.4f.csv' % (alpha,"data_tf_word",score)else:class_name = '/bayes_%s_class_%0.4f.csv' % ("data_tf_word",score)proba_name = '/bayes_%s_proba_%0.4f.csv' % ("data_tf_word",score)model_name = '/bayes_%s_model_%0.4f.csv' % ("data_tf_word",score)with open(path_save + model_name, 'wb')as f:pickle.dump(model, f)df_result.to_csv(path_save + class_name, index=False)df_proba.to_csv(path_save + proba_name, index=False)def test():_start = time.time()__start = _start # 记录总的时间X_train, X_val, Y_train, Y_val, x_test = read_data(DATA_PATH)get_time_diff(_start,"读取数据")_start = time.time()params = {"alpha": np.arange(0.75,0.86,0.01)}# search_model = model_search_random(X_train,Y_train,param_dist=params,n_iter_search=15,clf=MultinomialNB())search_model = model_search_grid(X_train, Y_train, param_dist=params,clf = MultinomialNB())get_time_diff(_start,"模型搜索")_start = time.time()cv_results_, best_score_, best_params_, model_best_estimator_ = model_best(search_model)alpha = best_params_['alpha']# fit_prior = best_params_['fit_prior']model = CalibratedClassifierCV(base_estimator=MultinomialNB(alpha=alpha))model.fit(X_train,Y_train)get_time_diff(_start,"最佳模型")_start = time.time()df_res, df_prob, score = predict(X_val,Y_val,x_test,model)get_time_diff(_start, "预测结果")print("验证集分数：%0.4f"%score)_start = time.time()save(score,SAVE_PATH,df_res,df_prob,model,alpha=alpha)get_time_diff(_start,"保存模型")get_time_diff(__start,"共用")if __name__ == '__main__':test()

达观杯_构建模型（四）贝叶斯相关推荐

达观杯_构建模型（一）linearSVM
特征:tfidf(word)+tfidf(article) """ 1.特征:tfidf(word)+tfidf(article) 2.模型:linearsvm 3.参数 ...
达观杯_构建模型（三）lightGBM
countvector(a)+doc(a)+hash(a) """ 1.特征:countvector(a)+doc(a)+hash(a) 2.模型:lgb "& ...
达观杯_构建模型（二）逻辑回归
特征:tfidf(word+article) """ 1.特征:tfidf(word+article) 2.模型:lr 3.参数:C=120 ""&q ...
决策树模型朴素贝叶斯模型_有关决策树模型的概述
决策树模型朴素贝叶斯模型 Decision Trees are one of the highly interpretable models and can perform both classif ...
如何用python进行建模_用 Python 进行贝叶斯模型建模（1）
本系列: 第1节:估计模型参数在这一节,我们将讨论贝叶斯方法是如何思考数据的,我们怎样通过 MCMC 技术估计模型参数. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 ...
判别两棵树是否相等设计算法_机器学习算法-朴素贝叶斯
一.概述概率分类器在许多分类算法应用中,特征和标签之间的关系并非是决定性的.比如说,我们想预测一个人究竟是否会在泰坦尼克号海难中生存下来,那我们可以建一棵决策树来学习我们的训练集.在训练中,其中一 ...
pymc3 贝叶斯线性回归_使用PyMC3进行贝叶斯媒体混合建模，带来乐趣和收益
pymc3 贝叶斯线性回归 Michael Johns, Zhenyu Wang, Bruno Dupont, and Luca Fiaschi 迈克尔·约翰斯,王振宇,布鲁诺·杜邦和卢卡·菲亚斯基 ...
【NLP】经典分类模型朴素贝叶斯解读
贝叶斯分类器在早期的自然语言处理任务中有着较多实际的应用,例如大部分的垃圾邮件处理都是用的贝叶斯分类器.贝叶斯分类器的理论对于理解后续的NLP模型有很大的进益,感兴趣的小伙伴一定要好好看看,本文会详细 ...
PGM：有向图模型：贝叶斯网络
http://blog.csdn.net/pipisorry/article/details/52489270 为什么用贝叶斯网络联合分布的显式表示 Note: n个变量的联合分布,每个x对应两个值 ...

达观杯_构建模型（四）贝叶斯

达观杯_构建模型（四）贝叶斯相关推荐

最新文章

热门文章