exam-question-classification/classify_service/contrast_experiment.py

#!/usr/bin/python
#coding=utf-8

import os
import numpy as np
import logging
import sklearn
from sklearn.model_selection import train_test_split     #导入切分训练集、测试集模块
from sklearn.neighbors import KNeighborsClassifier
from sklearn import svm
from sklearn.naive_bayes import GaussianNB


fileName = './constract.log'
handler = [logging.FileHandler(filename=fileName,encoding="utf-8")]
logging.basicConfig(level = logging.DEBUG, handlers = handler)

parent_path = os.path.dirname(os.path.realpath(__file__))
grander_path = os.path.dirname(parent_path)
word_list_data_path_base = parent_path + "/word_list_data/"
word2index_path_base = grander_path + "/word2index/"

dataset_name = "航天中认自主可控众包测试练习赛"

max_len = 64
vocab_size = 5000
embedding_size = 64
batch_size = 16
random_state = 15

def contrast():
    logging.info("正在加载初始数据")
    txts = np.load(word_list_data_path_base + str(dataset_name) + ".npy", allow_pickle=True)
    labels = np.load(word_list_data_path_base + str(dataset_name) + "_label.npy", allow_pickle=True)
    labels_new = []
    for label in labels:
        label_new = 0
        for i in range(len(label)):
            label_new += i * label[i]
        labels_new.append(label_new)
    labels_new = np.array(labels_new)

    logging.info("正在加载词表")
    word2index_path = word2index_path_base + str(dataset_name) + ".npy"
    word2index = np.load(word2index_path, allow_pickle=True).item()
    features = []
    for txt in txts:
        text_feature = text_to_feature(txt, word2index, max_len)
        features.append(text_feature)
    #np.save(, features)
    score_knn_lowest = 100
    score_svm_lowest = 100
    score_nb_lowest = 100
    score_knn_all = 0
    score_svm_all = 0
    score_nb_all = 0
    for i in range(random_state):
        train_data, test_data, train_label, test_label = sklearn.model_selection.train_test_split(features, labels_new, random_state = i, train_size = 0.2,test_size = 0.8)
        
        logging.info("正在训练k最近邻分类器")
        knn_classifier = KNeighborsClassifier()
        knn_classifier.fit(train_data, train_label)
        score_knn = knn_classifier.score(test_data, test_label)
        if score_knn < score_knn_lowest:
            score_knn_lowest = score_knn
        score_knn_all = score_knn_all + score_knn
        logging.info("k最近邻分类器准确率为{}".format(score_knn))
        
        logging.info("正在训练SVM分类器")
        svm_classifier = svm.SVC(C=2,kernel='rbf',gamma=10,decision_function_shape='ovr')
        svm_classifier.fit(train_data, train_label)
        score_svm = svm_classifier.score(test_data, test_label)
        if score_svm < score_svm_lowest:
            score_svm_lowest = score_svm
        score_svm_all = score_svm_all + score_svm
        logging.info("SVM分类器准确率为{}".format(score_svm))
        
        logging.info("正在训练朴素贝叶斯分类器")
        muNB_classifier = GaussianNB()
        muNB_classifier.fit(train_data, train_label)
        score_nb = muNB_classifier.score(test_data, test_label)
        if score_nb < score_nb_lowest:
            score_nb_lowest = score_nb
        score_nb_all = score_nb_all + score_nb
        logging.info("朴素贝叶斯分类器准确率为{}".format(score_nb))
    logging.info("k最近邻分类器最低准确率为{}".format(score_knn_lowest))
    logging.info("SVM分类器最低准确率为{}".format(score_svm_lowest))
    logging.info("朴素贝叶斯分类器最低准确率为{}".format(score_nb_lowest))
    logging.info("k最近邻分类器平均准确率为{}".format(score_knn_all / random_state))
    logging.info("SVM分类器平均准确率为{}".format(score_svm_all / random_state))
    logging.info("朴素贝叶斯分类器平均准确率为{}".format(score_nb_all / random_state))
    

    
def text_to_feature(text, word2index, max_len):
    feature = []
    for word in text:
        if word in word2index:
            feature.append(word2index[word])
        else:
            feature.append(word2index["<unk>"])
        if(len(feature) == max_len):
            break
    feature = feature + [word2index["<pad>"]] * (max_len - len(feature))
    return feature

if __name__ == "__main__":
    contrast()