keras遇到bert實戰一（bert實現分類）

說明：最近一直在做關系抽取的任務，此次僅僅是記錄一個實用的簡單示例

參考https://www.cnblogs.com/jclian91/p/12301056.html

參考https://blog.csdn.net/asialee_bird/article/details/102747435

import pandas as pd
import codecs, gc
import numpy as np
from sklearn.model_selection import KFold
from keras_bert import load_trained_model_from_checkpoint, Tokenizer
from keras.metrics import top_k_categorical_accuracy
from keras.layers import *
from keras.callbacks import *
from keras.models import Model
import keras.backend as K
from keras.optimizers import Adam
from keras.utils import to_categorical

# 讀取訓練集和測試集
train_df = pd.read_csv(r'D:\Program Files\FileRecv\情感分析數據集/data_train.csv', sep='\t', names=['id', 'type', 'contents', 'labels']).astype(str)
test_df = pd.read_csv(r'D:\Program Files\FileRecv\情感分析數據集/data_test.csv', sep='\t', names=['id', 'type', 'contents']).astype(str)
train_df = train_df[:200]
test_df = test_df[:20]

maxlen = 100  # 設置序列長度為120，要保證序列長度不超過512

# 預訓練好的模型
config_path = r'C:\Users\Downloads\chinese_L-12_H-768_A-12/bert_config.json'
checkpoint_path = r'C:\Users\Downloads\chinese_L-12_H-768_A-12/bert_model.ckpt'
dict_path = r'C:\Users\Downloads\chinese_L-12_H-768_A-12/vocab.txt'

# 將詞表中的詞編號轉換為字典
token_dict = {}
with codecs.open(dict_path, 'r', 'utf8') as reader:
    for line in reader:
        token = line.strip()
        token_dict[token] = len(token_dict)


# 重寫tokenizer
class OurTokenizer(Tokenizer):
    def _tokenize(self, text):
        R = []
        for c in text:
            if c in self._token_dict:
                R.append(c)
            elif self._is_space(c):
                R.append('[unused1]')  # 用[unused1]來表示空格類字符
            else:
                R.append('[UNK]')  # 不在列表的字符用[UNK]表示
        return R


tokenizer = OurTokenizer(token_dict)


# 讓每條文本的長度相同，用0填充
def seq_padding(X, padding=0):
    L = [len(x) for x in X]
    ML = max(L)
    return np.array([
        np.concatenate([x, [padding] * (ML - len(x))]) if len(x) < ML else x for x in X
    ])


# data_generator只是一種為了節約內存的數據方式
class data_generator:
    def __init__(self, data, batch_size=32, shuffle=True):
        self.data = data
        self.batch_size = batch_size
        self.shuffle = shuffle
        self.steps = len(self.data) // self.batch_size
        if len(self.data) % self.batch_size != 0:
            self.steps += 1

    def __len__(self):
        return self.steps

    def __iter__(self):
        while True:
            idxs = list(range(len(self.data)))

            if self.shuffle:
                np.random.shuffle(idxs)

            X1, X2, Y = [], [], []
            for i in idxs:
                d = self.data[i]
                text = d[0][:maxlen]
                x1, x2 = tokenizer.encode(first=text)
                y = d[1]
                X1.append(x1)
                X2.append(x2)
                Y.append([y])
                if len(X1) == self.batch_size or i == idxs[-1]:
                    X1 = seq_padding(X1)
                    X2 = seq_padding(X2)
                    Y = seq_padding(Y)
                    yield [X1, X2], Y[:, 0, :]
                    [X1, X2, Y] = [], [], []


# 計算top-k正確率,當預測值的前k個值中存在目標類別即認為預測正確
def acc_top2(y_true, y_pred):
    return top_k_categorical_accuracy(y_true, y_pred, k=2)


# bert模型設置
def build_bert(nclass):
    bert_model = load_trained_model_from_checkpoint(config_path, checkpoint_path, seq_len=None)  # 加載預訓練模型

    for l in bert_model.layers:
        l.trainable = True

    x1_in = Input(shape=(None,))
    x2_in = Input(shape=(None,))

    x = bert_model([x1_in, x2_in])
    x = Lambda(lambda x: x[:, 0])(x)  # 取出[CLS]對應的向量用來做分類
    p = Dense(nclass, activation='softmax')(x)

    model = Model([x1_in, x2_in], p)
    model.compile(loss='categorical_crossentropy',
                  optimizer=Adam(1e-5),  # 用足夠小的學習率
                  metrics=['accuracy', acc_top2])
    print(model.summary())
    return model


# 訓練數據、測試數據和標簽轉化為模型輸入格式
DATA_LIST = []
for data_row in train_df.iloc[:].itertuples():
    DATA_LIST.append((data_row.contents, to_categorical(data_row.labels, 3)))
DATA_LIST = np.array(DATA_LIST)

DATA_LIST_TEST = []
for data_row in test_df.iloc[:].itertuples():
    DATA_LIST_TEST.append((data_row.contents, to_categorical(0, 3)))
DATA_LIST_TEST = np.array(DATA_LIST_TEST)


# 交叉驗證訓練和測試模型
def run_cv(nfold, data, data_labels, data_test):
    kf = KFold(n_splits=nfold, shuffle=True, random_state=520).split(data)
    train_model_pred = np.zeros((len(data), 3))
    test_model_pred = np.zeros((len(data_test), 3))

    for i, (train_fold, test_fold) in enumerate(kf):
        X_train, X_valid, = data[train_fold, :], data[test_fold, :]

        model = build_bert(3)
        early_stopping = EarlyStopping(monitor='val_acc', patience=3)  # 早停法，防止過擬合
        plateau = ReduceLROnPlateau(monitor="val_acc", verbose=1, mode='max', factor=0.5,
                                    patience=2)  # 當評價指標不在提升時，減少學習率
        checkpoint = ModelCheckpoint('./bert_dump/' + str(i) + '.hdf5', monitor='val_acc', verbose=2,
                                     save_best_only=True, mode='max', save_weights_only=True)  # 保存最好的模型

        train_D = data_generator(X_train, shuffle=True)
        valid_D = data_generator(X_valid, shuffle=True)
        test_D = data_generator(data_test, shuffle=False)
        # 模型訓練
        model.fit_generator(
            train_D.__iter__(),
            steps_per_epoch=len(train_D),
            epochs=5,
            validation_data=valid_D.__iter__(),
            validation_steps=len(valid_D),
            callbacks=[early_stopping, plateau, checkpoint],
        )

        # model.load_weights('./bert_dump/' + str(i) + '.hdf5')

        # return model
        train_model_pred[test_fold, :] = model.predict_generator(valid_D.__iter__(), steps=len(valid_D), verbose=1)
        test_model_pred += model.predict_generator(test_D.__iter__(), steps=len(test_D), verbose=1)

        del model
        gc.collect()  # 清理內存
        K.clear_session()  # clear_session就是清除一個session
        # break

    return train_model_pred, test_model_pred


# n折交叉驗證
train_model_pred, test_model_pred = run_cv(2, DATA_LIST, None, DATA_LIST_TEST)

test_pred = [np.argmax(x) for x in test_model_pred]

# 將測試集預測結果寫入文件
output = pd.DataFrame({'id': test_df.id, 'sentiment': test_pred})
output.to_csv('results.csv', index=None)