def pred_input(str_input, path_hyper_parameter=path_hyper_parameters): # 输入预测 # 加载超参数 hyper_parameters = load_json(path_hyper_parameter) pt = PreprocessTextMulti() # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding ques = str_input # str to token ques_embed = ra_ed.sentence2idx(ques) if hyper_parameters['embedding_type'] == 'bert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # 预测 pred = graph.predict(x_val) print(pred) # 取id to label and pred pre = pt.prereocess_idx(pred[0]) ls_nulti = [] for ls in pre[0]: if ls[1] >= 0.73: ls_nulti.append(ls) print(str_input) print(pre[0]) print(ls_nulti)
def train(hyper_parameters=None, rate=1.0): if not hyper_parameters: hyper_parameters = { 'len_max': 150, # 句子最大长度, 固定推荐20-50, bert越长会越慢, 占用空间也会变大, 本地win10-4G设为20就好, 过大小心OOM 'embed_size': 200, # 字/词向量维度, bert取768, word取300, char可以更小些 'vocab_size': 21128, # 这里随便填的,会根据代码里修改 'trainable': True, # embedding是静态的还是动态的, 即控制可不可以微调 'level_type': 'char', # 级别, 最小单元, 字/词, 填 'char' or 'word', 注意:word2vec模式下训练语料要首先切好 'embedding_type': 'random', # 级别, 嵌入类型, 还可以填'xlnet'、'random'、 'bert'、 'albert' or 'word2vec" 'gpu_memory_fraction': 0.66, # gpu使用率 'ifChangeOutput': True, 'model': {'label': 19, # 类别数 'batch_size': 100, # 批处理尺寸, 感觉原则上越大越好,尤其是样本不均衡的时候, batch_size设置影响比较大 'dropout': 0.5, # 随机失活, 概率 'decay_step': 100, # 学习率衰减step, 每N个step衰减一次 'decay_rate': 0.9, # 学习率衰减系数, 乘法 'epochs': 50, # 训练最大轮次 'patience': 3, # 早停,2-3就好 'lr': 1e-3, # 学习率, bert取5e-5, 其他取1e-3, 对训练会有比较大的影响, 如果准确率一直上不去,可以考虑调这个参数 'l2': 1e-9, # l2正则化 'activate_classify': 'sigmoid', # 'sigmoid', # 最后一个layer, 即分类激活函数 'loss': 'binary_crossentropy', # 损失函数, 可能有问题, 可以自己定义 # 'metrics': 'top_k_categorical_accuracy', # 1070个类, 太多了先用topk, 这里数据k设置为最大:33 'metrics': 'accuracy', # 保存更好模型的评价标准 'is_training': True, # 训练后者是测试模型 'model_path': path_model, # 模型地址, loss降低则保存的依据, save_best_only=True, save_weights_only=True 'path_hyper_parameters': path_hyper_parameters, # 模型(包括embedding),超参数地址, 'path_fineture': path_fineture, # 保存embedding trainable地址, 例如字向量、词向量、bert向量等 }, 'embedding': {'layer_indexes': [12], # bert取的层数 # 'corpus_path': '', # embedding预训练数据地址,不配则会默认取conf里边默认的地址, keras-bert可以加载谷歌版bert,百度版ernie(需转换,https://github.com/ArthurRizar/tensorflow_ernie),哈工大版bert-wwm(tf框架,https://github.com/ymcui/Chinese-BERT-wwm) }, 'data': {'train_data': "./data/train.csv", # 训练数据 'val_data': "./data/val.csv", # 验证数据 'test_data':"./data/test.csv" }, } # 删除先前存在的模型和embedding微调模型等 delete_file(path_model_dir) time_start = time.time() # graph初始化 graph = Graph(hyper_parameters) print("graph init ok!") ra_ed = graph.word_embedding # 数据预处理 pt = PreprocessTextMulti() print(ra_ed,rate) x_train, y_train,_,_ = pt.preprocess_label_ques_to_idx(hyper_parameters['embedding_type'], hyper_parameters['data']['train_data'], ra_ed, rate=rate, shuffle=True) print('train data progress ok!') x_val, y_val,_,_ = pt.preprocess_label_ques_to_idx(hyper_parameters['embedding_type'], hyper_parameters['data']['val_data'], ra_ed, rate=rate, shuffle=True) print("data progress ok!") print(len(y_train)) # 训练 graph.fit(x_train, y_train, x_val, y_val) print("耗时:" + str(time.time() - time_start))
def train(hyper_parameters=None, rate=1.0): if not hyper_parameters: hyper_parameters = { 'len_max': 50, # 句子最大长度, 固定推荐20-50, bert越长会越慢, 占用空间也会变大, 本地win10-4G设为20就好, 过大小心OOM 'embed_size': 300, # 字/词向量维度, bert取768, word取300, char可以更小些 'vocab_size': 20000, # 这里随便填的,会根据代码里修改 'trainable': True, # embedding是静态的还是动态的, 即控制可不可以微调 'level_type': 'char', # 级别, 最小单元, 字/词, 填 'char' or 'word', 注意:word2vec模式下训练语料要首先切好 'embedding_type': 'random', # 级别, 嵌入类型, 还可以填'xlnet'、'random'、 'bert'、 'albert' or 'word2vec" 'gpu_memory_fraction': 0.76, #gpu使用率 'model': { 'label': 17, # 类别数 'batch_size': 2, # 批处理尺寸, 感觉原则上越大越好,尤其是样本不均衡的时候, batch_size设置影响比较大 'dropout': 0.5, # 随机失活, 概率 'decay_step': 100, # 学习率衰减step, 每N个step衰减一次 'decay_rate': 0.9, # 学习率衰减系数, 乘法 'epochs': 20, # 训练最大轮次 'patience': 3, # 早停,2-3就好 'lr': 1e-3, # 学习率,bert取5e-5,其他取1e-3, 对训练会有比较大的影响, 如果准确率一直上不去,可以考虑调这个参数 'l2': 1e-9, # l2正则化 'activate_classify': 'softmax', # 最后一个layer, 即分类激活函数 'loss': 'categorical_crossentropy', # 损失函数 'metrics': 'accuracy', # 保存更好模型的评价标准 'is_training': True, # 训练后者是测试模型 'model_path': path_model, # 模型地址, loss降低则保存的依据, save_best_only=True, save_weights_only=True 'path_hyper_parameters': path_hyper_parameters, # 模型(包括embedding),超参数地址, 'path_fineture': path_fineture, # 保存embedding trainable地址, 例如字向量、词向量、bert向量等 }, 'embedding': { 'layer_indexes': [1, 2, 3, 12, 13], # bert取的层数,1为embedding层,未处理 # 'corpus_path': 'Y:/BaiduNetdiskDownload/DataSet/bert-model/chinese_bert_chinese_wwm_L-12_H-768_A-12', # embedding预训练数据地址,不配则会默认取conf里边默认的地址 # 'corpus_path':'Y:/BaiduNetdiskDownload/DataSet/bert-model/baidu_ernie', # keras - bert可以加载谷歌版bert, 百度版ernie(需转换,https: // github.com / ArthurRizar / tensorflow_ernie), 哈工大版bert - wwm(tf框架,https: // github.com / ymcui / Chinese - BERT - wwm) }, 'data': { 'train_data': path_baidu_qa_2019_train, # 训练数据 'val_data': path_baidu_qa_2019_valid # 验证数据 }, } # 删除先前存在的模型和embedding微调模型等 delete_file(path_model_dir) time_start = time.time() # graph初始化 graph = Graph(hyper_parameters) print("graph init ok!") ra_ed = graph.word_embedding # 训练 graph.fit_generator(embed=ra_ed, rate=rate) print("耗时:" + str(time.time() - time_start))
def pred_tet(path_hyper_parameter=path_hyper_parameters, path_test=None, rate=1.0): # 测试集的准确率 hyper_parameters = load_json(path_hyper_parameter) if path_test: # 从外部引入测试数据地址 hyper_parameters['data']['val_data'] = path_test time_start = time.time() # graph初始化 graph = Graph(hyper_parameters) print("graph init ok!") graph.load_model() print("graph load ok!") ra_ed = graph.word_embedding # 数据预处理 pt = PreprocessText() y, x = read_and_process(hyper_parameters['data']['val_data']) # 取该数据集的百分之几的语料测试 len_rate = int(len(y) * rate) x = x[1:len_rate] y = y[1:len_rate] y_pred = [] count = 0 for x_one in x: count += 1 ques_embed = ra_ed.sentence2idx(x_one) if hyper_parameters['embedding_type'] in ['bert', 'albert']: x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] elif hyper_parameters['embedding_type'] == 'xlnet': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val_3 = np.array([ques_embed[2]]) x_val = [x_val_1, x_val_2, x_val_3] else: x_val = ques_embed # 预测 pred = graph.predict(x_val) pre = pt.prereocess_idx(pred[0]) label_pred = pre[0][0][0] if count % 1000 == 0: print(label_pred) y_pred.append(label_pred) print("data pred ok!") # 预测结果转为int类型 index_y = [pt.l2i_i2l['l2i'][i] for i in y] index_pred = [pt.l2i_i2l['l2i'][i] for i in y_pred] target_names = [ pt.l2i_i2l['i2l'][str(i)] for i in list(set((index_pred + index_y))) ] # 评估 report_predict = classification_report(index_y, index_pred, target_names=target_names, digits=9) print(report_predict) print("耗时:" + str(time.time() - time_start))
def train(hyper_parameters=None, rate=1.0): if not hyper_parameters: hyper_parameters = { 'len_max': 50, # 句子最大长度, 固定推荐20-50, bert越长会越慢, 占用空间也会变大, 本地win10-4G设为20就好, 过大小心OOM 'embed_size': 300, # 字/词向量维度, bert取768, word取300, char可以更小些 'vocab_size': 20000, # 这里随便填的,会根据代码里修改 'trainable': True, # embedding是静态的还是动态的, 即控制可不可以微调 'level_type': 'word', # 级别, 最小单元, 字/词, 填 'char' or 'word', 注意:word2vec模式下训练语料要首先切好 'embedding_type': 'word2vec', # 级别, 嵌入类型, 还可以填'random'、 'bert' or 'word2vec" 'gpu_memory_fraction': 0.66, #gpu使用率 'model': {'label': 17, # 类别数 'batch_size': 32, # 批处理尺寸, 感觉原则上越大越好,尤其是样本不均衡的时候, batch_size设置影响比较大 'dropout': 0.5, # 随机失活, 概率 'decay_step': 100, # 学习率衰减step, 每N个step衰减一次 'decay_rate': 0.9, # 学习率衰减系数, 乘法 'epochs': 20, # 训练最大轮次 'patience': 3, # 早停,2-3就好 'lr': 1e-3, # 学习率,bert取5e-5,其他取1e-3, 对训练会有比较大的影响, 如果准确率一直上不去,可以考虑调这个参数 'l2': 1e-9, # l2正则化 'activate_classify': 'softmax', # 最后一个layer, 即分类激活函数 'loss': 'categorical_crossentropy', # 损失函数 'metrics': 'accuracy', # 保存更好模型的评价标准 'is_training': True, # 训练后者是测试模型 'model_path': path_model, # 模型地址, loss降低则保存的依据, save_best_only=True, save_weights_only=True 'path_hyper_parameters': path_hyper_parameters, # 模型(包括embedding),超参数地址, 'path_fineture': path_fineture, # 保存embedding trainable地址, 例如字向量、词向量、bert向量等 }, 'embedding': {'layer_indexes': [12], # bert取的层数 'corpus_path': '', # embedding预训练数据地址,不配则会默认取conf里边默认的地址 }, 'data':{'train_data': path_baidu_qa_2019_train, # 训练数据 'val_data': path_baidu_qa_2019_valid # 验证数据 }, } # 删除先前存在的模型和embedding微调模型等 delete_file(path_model_dir) time_start = time.time() # graph初始化 graph = Graph(hyper_parameters) print("graph init ok!") ra_ed = graph.word_embedding # 数据预处理 pt = PreprocessText() x_train, y_train = pt.preprocess_label_ques_to_idx(hyper_parameters['embedding_type'], hyper_parameters['data']['train_data'], ra_ed, rate=rate, shuffle=True) x_val, y_val = pt.preprocess_label_ques_to_idx(hyper_parameters['embedding_type'], hyper_parameters['data']['val_data'], ra_ed, rate=rate, shuffle=True) print("data propress ok!") print(len(y_train)) # 训练 graph.fit(x_train, y_train, x_val, y_val) print("耗时:" + str(time.time()-time_start))
def pred_input(path_hyper_parameter=path_hyper_parameters): # 输入预测 # 加载超参数 hyper_parameters = load_json(path_hyper_parameter) pt = PreprocessTextMulti(path_model_dir) # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding ques = '我要打王者荣耀' # str to token ques_embed = ra_ed.sentence2idx(ques) if hyper_parameters['embedding_type'] in ['bert', 'albert']: x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # 预测 pred = graph.predict(x_val) print(pred) # 取id to label and pred pre = pt.prereocess_idx(pred[0]) ls_nulti = [] for ls in pre[0]: if ls[1] >= 0.5: ls_nulti.append(ls) print(pre[0]) print(ls_nulti) while True: print("请输入: ") ques = input() ques_embed = ra_ed.sentence2idx(ques) print(ques_embed) if hyper_parameters['embedding_type'] in ['bert', 'albert']: x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed pred = graph.predict(x_val) pre = pt.prereocess_idx(pred[0]) ls_nulti = [] for ls in pre[0]: if ls[1] >= 0.5: ls_nulti.append(ls) print(pre[0]) print(ls_nulti)
def pred_input(path_hyper_parameter=path_hyper_parameters): # 输入预测 # 加载超参数 hyper_parameters = load_json(path_hyper_parameter) pt = PreprocessText() # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding ques = '我要打王者荣耀' # str to token ques_embed = ra_ed.sentence2idx(ques) if hyper_parameters['embedding_type'] == 'bert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # 预测 pred = graph.predict(x_val) # 取id to label and pred pre = pt.prereocess_idx(pred[0]) print(pre) while True: print("请输入: ") ques = input() ques_embed = ra_ed.sentence2idx(ques) print(ques_embed) if hyper_parameters['embedding_type'] == 'bert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed pred = graph.predict(x_val) pre = pt.prereocess_idx(pred[0]) print(pre)
def evaluate(path_hyper_parameter=path_hyper_parameters, rate=1.0): # 输入预测 # 加载超参数 hyper_parameters = load_json(path_hyper_parameter) pt = PreprocessTextMulti() # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding # get validation data ques_list, val_list, que, val = pt.preprocess_label_ques_to_idx( hyper_parameters['embedding_type'], hyper_parameters['data']['val_data'], ra_ed, rate=rate, shuffle=True) print(len(ques_list)) print("que:", len(que)) # print(val) # str to token ques_embed_list = [] count = 0 acc_count = 0 not_none_count = 0 not_none_acc_count = 0 sum_iou = 0 sum_all_iou = 0 for index, que___ in enumerate(que): # print("原句 ", index, que[index]) # print("真实分类 ", index, val[index]) # print("ques: ", ques) ques_embed = ra_ed.sentence2idx(que[index]) if hyper_parameters['embedding_type'] == 'albert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) ques_embed = [x_val_1, x_val_2] else: x_val = ques_embed # print("ques_embed: ", ques_embed) if hyper_parameters['embedding_type'] == 'bert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # print("x_val", x_val) ques_embed_list.append(x_val) # 预测 pred = graph.predict(x_val) # print(pred) # 取id to label and pred pre = pt.prereocess_idx(pred[0]) # print("pre",pre) ls_nulti = [] threshold = 0.44 top_threshold = 0 for i, ls in enumerate(pre[0]): if i == 0 or ls[1] > threshold: ls_nulti.append(ls) top_threshold = ls[1] elif abs(ls[1] - top_threshold) < top_threshold / 4.0: ls_nulti.append(ls) # print("预测结果", index, pre[0]) # print(ls_nulti) res = cal_acc(ls_nulti, val[index].split(",")) res_iou, res_all_iou = cal_iou(ls_nulti, val[index].split(",")) sum_iou += res_iou sum_all_iou += res_all_iou if res: if val[index] != "无": not_none_acc_count += 1 acc_count += 1 else: print("原句 ", index, que[index]) print("真实分类 ", index, val[index]) print("pre ", pre) print("iou ", res_iou) count += 1 if val[index] != "无": not_none_count += 1 print("acc: ", acc_count / count) print("not none acc: ", not_none_acc_count / not_none_count) print("average iou: ", sum_iou / sum_all_iou) # log append_log(hyper_parameters, acc_count / count, not_none_acc_count / not_none_count, threshold)
'lr': 1e-4, 'l2': 1e-9, 'activate_classify': 'softmax', # 还可以填'random'、 'bert' or 'word2vec" 'embedding_type': 'random', 'is_training': False, 'model_path': path_model_fast_text_baiduqa_2019,}, 'embedding':{ 'embedding_type': 'random', 'corpus_path': path_embedding_random_char, 'level_type': 'char', 'embed_size': 300, 'len_max': 50,}, } # ns = np.array([1,2,3,4]) # print(type(ns)) pt = PreprocessText graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding ques = '你好呀' ques_embed = ra_ed.sentence2idx(ques) pred = graph.predict(np.array([ques_embed])) pre = pt.prereocess_idx(pred[0]) print(pre) while True: print("请输入: ") ques = input() ques_embed = ra_ed.sentence2idx(ques) pred = graph.predict(np.array([ques_embed])) pre = pt.prereocess_idx(pred[0]) print(pre)
from keras_textclassification.conf.path_config import path_model, path_fineture, path_model_dir, path_hyper_parameters # 训练验证数据地址 from keras_textclassification.conf.path_config import path_baidu_qa_2019_train, path_baidu_qa_2019_valid # 数据预处理, 删除文件目录下文件 from keras_textclassification.data_preprocess.text_preprocess import PreprocessText, read_and_process, load_json # 模型图 from keras_textclassification.m02_TextCNN.graph import TextCNNGraph as Graph import numpy as np # flask from flask import Flask, request, jsonify app = Flask(__name__) hyper_parameters = load_json(path_hyper_parameters) pt = PreprocessText(path_model_dir) # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding ques = '我要打王者荣耀' # str to token ques_embed = ra_ed.sentence2idx(ques) if hyper_parameters['embedding_type'] in ['bert', 'albert']: x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # 预测 pred = graph.predict(x_val) # 取id to label and pred pre = pt.prereocess_idx(pred[0])
def evaluate(path_hyper_parameter=path_hyper_parameters, rate=1.0): # 输入预测 # 加载超参数 hyper_parameters = load_json(path_hyper_parameter) pt = PreprocessTextMulti() # 模式初始化和加载 graph = Graph(hyper_parameters) graph.load_model() ra_ed = graph.word_embedding print(hyper_parameters['data']['test_data']) # init confusion table dict_all=initConfusion() # get validation data ques_list, val_list, que, val = pt.preprocess_label_ques_to_idx(hyper_parameters['embedding_type'], hyper_parameters['data']['test_data'], ra_ed, rate=rate, shuffle=True) print(len(ques_list)) print("que:",len(que)) # print(val) # str to token ques_embed_list = [] count = 0 acc_count = 0 not_none_count = 0 not_none_acc_count = 0 sum_iou = 0 density_correct=0 sum_all_iou=0 for index, que___ in enumerate(que): # print("原句 ", index, que[index]) # print("真实分类 ", index, val[index]) # print("ques: ", ques) ques_embed = ra_ed.sentence2idx(que[index]) if hyper_parameters['embedding_type'] == 'albert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) ques_embed = [x_val_1, x_val_2] else: x_val = ques_embed # print("ques_embed: ", ques_embed) if hyper_parameters['embedding_type'] == 'bert': x_val_1 = np.array([ques_embed[0]]) x_val_2 = np.array([ques_embed[1]]) x_val = [x_val_1, x_val_2] else: x_val = ques_embed # print("x_val", x_val) ques_embed_list.append(x_val) # 预测 pred = graph.predict(x_val) # print(pred) # 取id to label and pred pre = pt.prereocess_idx(pred[0]) # print("pre",pre) ls_nulti = [] threshold = 0.60 top_threshold = 0 has_scope=False for i, ls in enumerate(pre[0]): if ls[0] in ['多发','散发','无']: if not has_scope: has_scope=True ls_nulti.append(ls) if ls[0] in val[index].split(","): dict_all[ls[0]]['TP'] += 1 else: dict_all[ls[0]]['FN'] += 1 else: if ls[0] in val[index].split(","): dict_all[ls[0]]['FP']+=1 else: dict_all[ls[0]]['TN']+=1 if ls[0] not in ['多发','散发','无']: if ls[1] > threshold: ls_nulti.append(ls) if ls[0] in val[index].split(","): dict_all[ls[0]]['TP']+=1 else: dict_all[ls[0]]['FP']+=1 else: if ls[0] in val[index].split(","): dict_all[ls[0]]['FN']+=1 else: dict_all[ls[0]]['TN']+=1 # print("预测结果", index, pre[0]) # print(ls_nulti) res = cal_acc(ls_nulti, val[index].split(",")) res_iou = cal_iou(ls_nulti, val[index].split(",")) sum_iou += res_iou # sum_all_iou+=res_all_iou density_res=checkDensity(ls_nulti,val[index].split(",")) if density_res: density_correct+=1 if res: acc_count += 1 else: print("原句 ", index, que[index]) print("真实分类 ", index, val[index]) print("pre ", pre) print("iou ", res_iou) print(ls_nulti) count += 1 if val[index] != "无": not_none_count += 1 # print("softmax count",softmax_count) print("acc: ", acc_count / count) # print("not none acc: ", not_none_acc_count / not_none_count) print("average iou: ", sum_iou / count) print('density correct', density_correct/count) print(dict_all) import prettytable as pt tb = pt.PrettyTable() tb.field_names = [" ", "Recall", "Precision", "TP","FP","TN","FN"] for item in dict_all: if dict_all[item]['TP']+dict_all[item]['FN']==0: recall=1 else: recall = dict_all[item]['TP']/(dict_all[item]['TP']+dict_all[item]['FN']) if dict_all[item]['TP']+dict_all[item]['FP']==0: precision=1 else: precision = dict_all[item]['TP'] / (dict_all[item]['TP'] + dict_all[item]['FP']) # print(item,recall,precision) tb.add_row([item, recall, precision, dict_all[item]['TP'],dict_all[item]['FP'],dict_all[item]['TN'],dict_all[item]['FN']]) print(tb) # log append_log(hyper_parameters, acc_count / count, not_none_acc_count / not_none_count, threshold)
'epochs': 20, 'len_max': 50, 'vocab_size': 20000, #这里随便填的,会根据代码里修改 'lr': 1e-3, 'l2': 1e-6, 'activate_classify': 'softmax', 'embedding_type': 'random', # 还可以填'random'、 'bert' or 'word2vec" 'is_training': True, 'model_path': path_model_fast_text_baiduqa_2019,}, 'embedding':{ 'embedding_type': 'random', 'corpus_path': path_embedding_random_char, 'level_type': 'char', 'embed_size': 300, 'len_max': 50,}, } time_start = time.time() graph = Graph(hyper_parameters) ra_ed = graph.word_embedding pt = PreprocessText() x_train, y_train = pt.preprocess_baidu_qa_2019_idx(path_baidu_qa_2019_train, ra_ed) x_val, y_val = pt.preprocess_baidu_qa_2019_idx(path_baidu_qa_2019_valid, ra_ed) print(len(y_train)) graph.fit(x_train, y_train, x_val, y_val) print("耗时:" + str(time.time()-time_start)) # 1425170/1425170 [==============================] - 648s 455us/step - loss: 0.8771 - acc: 0.7317 - val_loss: 1.2533 - val_acc: 0.7069 # Epoch 00001: val_loss improved from inf to 1.25335, saving model to D:\workspace\pythonMyCode\django_project\ClassificationTextChinese/data/model/fast_text/model_fast_text.f5 # Epoch 2/20 # TIME: 20 * 5轮= 100 (min) # acc: 0.7317