NLP入门之新闻文本分类竞赛——FastText

    from fastText import train_unsupervised
    model = train_unsupervised(input, model='skipgram', lr=0.05, 
                            dim=100, ws=5, epoch=5, minCount=5, 
                            wordNgrams=1, loss='ns', bucket=2000000, 
                            thread=12, lrUpdateRate=100, t=0.0001, 
                            label='__label__', verbose=2, 
                            pretrainedVectors='')

    import pandas as pd
    from sklearn.metrics import f1_score
    import fasttext
    from sklearn.model_selection import StratifiedKFold
    from sklearn.externals import joblib
    
    def tunningFasttext():
    # 调参
    train_df = pd.read_csv('../dataset/train_set.csv', sep='\t', nrows=50000)
    
    train_df['label_ft'] = '__label__' + train_df['label'].astype(str)
    train_df[['text','label_ft']].iloc[:-5000].to_csv('train.csv', index=None, header=None, sep='\t') 
    
    #lr_list = [0.8, 0.9, 1, 1.1, 1.2, 1.3]
    #minCount_list = [1,2,3,4,5]
    #epoch_list = [20, 25, 30, 35]
    
    # lr:学习率
    # wordNgrams:n-gram设置
    # minCount:最低出现的词频
    # loss:损失函数（ns, hs, softmax)
    model = fasttext.train_supervised('train.csv', lr=1.0, wordNgrams=2,
    verbose=2, minCount=1, epoch=25, loss="hs")
    
    val_pred = [model.predict(x)[0][0].split('__')[-1] for x in train_df.iloc[-5000:]['text']]  
    fscore = f1_score(train_df['label'].values[-5000:].astype(str), val_pred, average='macro')
    print("the f1_score of training is: " + str(fscore))
    
    """
    # 调整部分参数f1_score的变化
    minCount is 1 which the f1_score of training is: 0.8237674862710784
    minCount is 2 which the f1_score of training is: 0.8249374300159736
    minCount is 3 which the f1_score of training is: 0.8240444453594258
    minCount is 4 which the f1_score of training is: 0.823250595891796
    minCount is 5 which the f1_score of training is: 0.8239732463432038
    
    lr is 0.9 which the f1_score of training is: 0.8251838864793
    lr is 1.0 which the f1_score of training is: 0.8286428017584191
    lr is 1.1 which the f1_score of training is: 0.8272700396899009
    lr is 1.2 which the f1_score of training is: 0.8262960227484231
    """
    
    """
    #  取15000条数据
    the f1_score of training is: 0.8244907044064147
    #  取50000条数据
    the f1_score of training is: 0.8904843116417097
    """

    def creatData():

    # 构建10折交叉验证训练和测试数据集
    train_df = pd.read_csv('../dataset/train_set.csv', sep='\t', nrows=200000) #  取200000条数据
    sfolder = StratifiedKFold(n_splits = 10)
    X = train_df['text']
    y = train_df['label']
    for index, (train_index, val_index) in enumerate(sfolder.split(X,y)):
        train_x, train_y= train_df['text'].iloc[train_index], train_df['label'].iloc[train_index]
        val_x, val_y = train_df['text'].iloc[val_index], train_df['label'].iloc[val_index]
        train_y = '__label__' + train_y.astype(str)
    
        traindata = pd.DataFrame(list(zip(train_x.values, train_y.values))) 
        traindata.to_csv(f'./data/sfalltrain{index}.csv', index=None, header=['text', 'label_ft'], sep='\t')
        testdata = pd.DataFrame(list(zip(val_x.values, val_y.values)))
        testdata.to_csv(f'./data/sfalltest{index}.csv', index=None, header=['text', 'label'], sep='\t')

    ############进行10折交叉验证############
    def trainFasttext():
    #进行10折交叉验证
    fscore_list = []
    for i in range(10):
        train_path = './data/sfalltrain' + str(i) +'.csv'
        test_path = './data/sfalltest' + str(i) +'.csv'
        test_df = pd.read_csv(test_path, sep='\t')
    
        #model = fasttext.train_supervised(train_path, lr=1.0, wordNgrams=2, verbose=2, minCount=1, epoch=25, loss="hs")
        model = fasttext.train_supervised(train_path, lr=1.0, wordNgrams=2, verbose=2, minCount=1, epoch=25, loss='softmax')
    
        test_pred = [model.predict(x)[0][0].split('__')[-1] for x in test_df['text']]
        fscore = f1_score(test_df['label'].values.astype(str), test_pred, average='macro')
        fscore_list.append(fscore)
        print(f"the f1_score of training{i+1} is: " , fscore)
        
        if i == 1:
            break;
    print('The average f1_score is', sum(fscore_list)/len(fscore_list))
    
    """
    # 参数：loss='hs'
    the f1_score of training1 is:  0.911607846104518
    the f1_score of training2 is:  0.9164789318984321
    the f1_score of training3 is:  0.9152974954237019
    the f1_score of training4 is:  0.911595813010921
    the f1_score of training5 is:  0.9172344278973873
    the f1_score of training6 is:  0.9140821540157751
    the f1_score of training7 is:  0.9156893038604335
    the f1_score of training8 is:  0.9101087268398959
    the f1_score of training9 is:  0.912208373997352
    the f1_score of training10 is: 0.9158255016919351
    
    The average f1_score is 0.9140128574740352
    
    # 参数：loss='softmax'
    the f1_score of training1 is:  0.9288426310944516
    the f1_score of training2 is:  0.9314477168650777
    the f1_score of training3 is:  0.9304964586973928
    the f1_score of training4 is:  0.9306809809860263
    the f1_score of training5 is:  0.9299938215659551
    the f1_score of training6 is:  0.9295759220278372
    the f1_score of training7 is:  0.9306571061874928
    the f1_score of training8 is:  0.9276190929199206
    the f1_score of training9 is:  0.9283404333208727
    the f1_score of training10 is:  0.9324812547890577
    The average f1_score is 0.9300135418454085
    """
    # 可发现设置loss='softmax'精确度更高