# -*- coding: utf-8 -*- """ @Time : 2023/1/31 19:02 @Author : @FileName: @Software: @Describe: """ import os # os.environ["TF_KERAS"] = "1" os.environ["CUDA_VISIBLE_DEVICES"] = "0" import json import numpy as np from bert4keras.backend import keras, set_gelu from bert4keras.tokenizers import Tokenizer, load_vocab from bert4keras.models import build_transformer_model from bert4keras.optimizers import Adam, extend_with_piecewise_linear_lr from bert4keras.snippets import sequence_padding, DataGenerator from bert4keras.snippets import open from keras.layers import Lambda, Dense import tensorflow as tf from keras.backend import set_session from sklearn.metrics.pairwise import cosine_similarity from rouge import Rouge # pip install rouge from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction from tqdm import tqdm import jieba from gensim.models import KeyedVectors, word2vec, Word2Vec import random config = tf.ConfigProto() config.gpu_options.allow_growth = True set_session(tf.Session(config=config)) # 此处不同 class Word2vecModel: def __init__(self): self.path = "E:\pycharm_workspace\查重分析\word2vec_model\\word2vec_add_new_18.model" self.model = Word2Vec.load(self.path) def word2vec_res(self,seg_0_list, seg_1_list): sentence_0_list = [] sentence_1_list = [] for i in seg_0_list: a = self.model.wv[i] sentence_0_list.append(a) for i in seg_1_list: a = self.model.wv[i] sentence_1_list.append(a) return sentence_0_list, sentence_1_list class Evaluator(keras.callbacks.Callback): """评估与保存 """ def __init__(self): self.rouge = Rouge() self.smooth = SmoothingFunction().method1 self.best_bleu = 0. # def on_epoch_end(self, epoch, logs=None): # metrics = self.evaluate(valid_data) # 评测模型 # if metrics['bleu'] > self.best_bleu: # self.best_bleu = metrics['bleu'] # model.save_weights('./best_model.weights') # 保存模型 # metrics['best_bleu'] = self.best_bleu # print('valid_data:', metrics) def evaluate_t(self, data_1, data_2, topk=1): total = 0 rouge_1, rouge_2, rouge_l, bleu = 0, 0, 0, 0 scores = self.rouge.get_scores(hyps=[data_1], refs=[data_2]) rouge_1 += scores[0]['rouge-1']['f'] rouge_2 += scores[0]['rouge-2']['f'] rouge_l += scores[0]['rouge-l']['f'] bleu += sentence_bleu( references=[data_1.split(' ')], hypothesis=data_2.split(' '), smoothing_function=self.smooth ) # rouge_1 /= total # rouge_2 /= total # rouge_l /= total # bleu /= total return [rouge_1, rouge_2, rouge_l, bleu] class bertModel: def __init__(self): self.config_path = '../chinese_roberta_wwm_ext_L-12_H-768_A-12/bert_config.json' self.checkpoint_path = '../chinese_roberta_wwm_ext_L-12_H-768_A-12/bert_model.ckpt' self.dict_path = '../chinese_roberta_wwm_ext_L-12_H-768_A-12/vocab.txt' self.token_dict, self.keep_tokens = load_vocab( dict_path=self.dict_path, simplified=True, startswith=['[PAD]', '[UNK]', '[CLS]', '[SEP]'], ) self.tokenizer = Tokenizer(self.token_dict, do_lower_case=True) self.buildmodel() def buildmodel(self): bert = build_transformer_model( config_path=self.config_path, checkpoint_path=self.checkpoint_path, return_keras_model=False, ) output = Lambda(lambda x: x[:, 0], name='CLS-token')(bert.model.output) self.model = keras.models.Model(bert.model.input, output) self.model.summary() def predict(self,text): batch_token_ids, batch_segment_ids = [], [] token_ids, segment_ids = self.tokenizer.encode(text, maxlen=256) batch_token_ids.append(token_ids) batch_segment_ids.append(segment_ids) return self.model.predict([batch_token_ids, batch_segment_ids]) def simbert(data_1, data_2): pass def word2vec(): pass def bleu(): pass if __name__ == '__main__': file = "../data/train_yy_1.txt" model = bertModel() eval_class = Evaluator() # word2vecmodel = Word2vecModel() try: with open(file, 'r', encoding="utf-8") as f: lines = [x.strip() for x in f if x.strip() != ''] except: with open(file, 'r', encoding="gbk") as f: lines = [x.strip() for x in f if x.strip() != ''] bertsim_list = [] bleusim_list = [] word2vecsim_list = [] data_train_text = [] random.shuffle(lines) print(len(lines)) for txt in tqdm(lines): text = txt.split('\t') if len(text) == 3: data_1 = text[0] data_2 = text[2] y1 = model.predict(data_1)[0] y2 = model.predict(data_2)[0] cos_sim = cosine_similarity(y1.reshape(1, -1), y2.reshape(1, -1)) bertsim_list.append(cos_sim[0][0]) bertsim_value = cos_sim[0][0] eval_list = eval_class.evaluate_t(' '.join(data_1), ' '.join(data_2)) bleusim_list.append(eval_list[3]) bleusim_value = eval_list[3] if bertsim_value <= 0.94 and bleusim_value <= 0.4: data_train_text.append("\t".join([data_1, "to", data_2])) # eval_list = eval_class.evaluate_t(' '.join(data_1), ' '.join(data_2)) # bleusim_list.append(eval_list[3]) # word2vec # seg_0_list = jieba.cut(data_1, cut_all=False) # seg_1_list = jieba.cut(data_2, cut_all=False) # seg_0_list = [char for char in seg_0_list] # seg_1_list = [char for char in seg_1_list] # # sentence_0_list, sentence_1_list = word2vecmodel.word2vec_res(seg_0_list, seg_1_list) # sentence_0_result = np.array(sentence_0_list) # sentence_1_result = np.array(sentence_1_list) # sentence_0_array = sentence_0_result.sum(axis=0) # sentence_1_array = sentence_1_result.sum(axis=0) # print(sentence_1_array) # print(sentence_0_array) # cos_sim = cosine_similarity(sentence_0_array.reshape(1, -1), sentence_1_array.reshape(1, -1)) # word2vecsim_list.append(cos_sim[0][0]) # bertsim_list = sorted(bertsim_list) # zong_num = len(bertsim_list) # print(bertsim_list) # print(bertsim_list[int(zong_num/2)]) # print(sum(bertsim_list)/zong_num) # bleusim_list = sorted(bleusim_list) # zong_num = len(bleusim_list) # print(bleusim_list) # print(bleusim_list[int(zong_num / 2)]) # print(sum(bleusim_list) / zong_num) fileName = 'train_yy_1_sim_09.txt' with open(fileName, 'w', encoding='utf-8') as file: for i in data_train_text: file.write(str(i) + '\n') file.close()