当前位置： 首页 > news >正文

小程序推广费用一个月需要多少钱西安seo顾问

news 2025/7/21 15:29:57
小程序推广费用一个月需要多少钱,西安seo顾问,视频分享网站怎么做,用dede.com网站上传文件1. 背景 有些场景下，开始的时候数据量很小，如果我们用一个几千条数据训练一个全新的深度机器学习的文本分类模型，效果不会很好。这个时候你有两种选择，1.用传统的机器学习训练，2.利用迁移学习在一个预训练的模型上训练…

1. 背景

有些场景下，开始的时候数据量很小，如果我们用一个几千条数据训练一个全新的深度机器学习的文本分类模型，效果不会很好。这个时候你有两种选择，1.用传统的机器学习训练，2.利用迁移学习在一个预训练的模型上训练。本博客教你怎么用tensorflow Hub和keras 在少量的数据上训练一个文本分类模型。

2. 实践

2.1. 下载IMDB 数据集，参考下面博客。

Imdb影评的数据集介绍与下载_imdb影评数据集-CSDN博客

2.2.  预处理数据

替换掉imdb目录 （imdb_raw_data_dir）. 创建dataset目录。

import numpy as np
import os as osimport re
from sklearn.model_selection import train_test_splitvocab_size = 30000
maxlen = 200
imdb_raw_data_dir = "/Users/harry/Documents/apps/ml/aclImdb"
save_dir = "dataset"def get_data(datapath =r'D:\train_data\aclImdb\aclImdb\train' ):pos_files = os.listdir(datapath + '/pos')neg_files = os.listdir(datapath + '/neg')print(len(pos_files))print(len(neg_files))pos_all = []neg_all = []for pf, nf in zip(pos_files, neg_files):with open(datapath + '/pos' + '/' + pf, encoding='utf-8') as f:s = f.read()s = process(s)pos_all.append(s)with open(datapath + '/neg' + '/' + nf, encoding='utf-8') as f:s = f.read()s = process(s)neg_all.append(s)print(len(pos_all))# print(pos_all[0])print(len(neg_all))X_orig= np.array(pos_all + neg_all)# print(X_orig)Y_orig = np.array([1 for _ in range(len(pos_all))] + [0 for _ in range(len(neg_all))])print("X_orig:", X_orig.shape)print("Y_orig:", Y_orig.shape)return X_orig, Y_origdef generate_dataset():X_orig, Y_orig = get_data(imdb_raw_data_dir + r'/train')X_orig_test, Y_orig_test = get_data(imdb_raw_data_dir + r'/test')X_orig = np.concatenate([X_orig, X_orig_test])Y_orig = np.concatenate([Y_orig, Y_orig_test])X = X_origY = Y_orignp.random.seed = 1random_indexs = np.random.permutation(len(X))X = X[random_indexs]Y = Y[random_indexs]X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.3)print("X_train:", X_train.shape)print("y_train:", y_train.shape)print("X_test:", X_test.shape)print("y_test:", y_test.shape)np.savez(save_dir + '/train_test', X_train=X_train, y_train=y_train, X_test= X_test, y_test=y_test )def rm_tags(text):re_tag = re.compile(r'<[^>]+>')return re_tag.sub(' ', text)def clean_str(string):string = re.sub(r"[^A-Za-z0-9(),!?\'\`]", " ", string)string = re.sub(r"\'s", " \'s", string)  # it's -> it 'sstring = re.sub(r"\'ve", " \'ve", string) # I've -> I 'vestring = re.sub(r"n\'t", " n\'t", string) # doesn't -> does n'tstring = re.sub(r"\'re", " \'re", string) # you're -> you arestring = re.sub(r"\'d", " \'d", string)  # you'd -> you 'dstring = re.sub(r"\'ll", " \'ll", string) # you'll -> you 'llstring = re.sub(r"\'m", " \'m", string) # I'm -> I 'mstring = re.sub(r",", " , ", string)string = re.sub(r"!", " ! ", string)string = re.sub(r"\(", " \( ", string)string = re.sub(r"\)", " \) ", string)string = re.sub(r"\?", " \? ", string)string = re.sub(r"\s{2,}", " ", string)return string.strip().lower()def process(text):text = clean_str(text)text = rm_tags(text)#text = text.lower()return  textif __name__ == '__main__':generate_dataset()

执行完后，产生train_test.npz 文件

2.3.  训练模型

1. 取数据集

def get_dataset_to_train():train_test = np.load('dataset/train_test.npz', allow_pickle=True)x_train =  train_test['X_train']y_train = train_test['y_train']x_test =  train_test['X_test']y_test = train_test['y_test']return x_train, y_train, x_test, y_test

2. 创建模型

基于nnlm-en-dim50/2 预训练的文本嵌入向量，在模型外面加了两层全连接。

def get_model():hub_layer = hub.KerasLayer(embedding_url, input_shape=[], dtype=tf.string, trainable=True)# Build the modelmodel = Sequential([hub_layer,Dense(16, activation='relu'),Dropout(0.5),Dense(2, activation='softmax')])print(model.summary())model.compile(optimizer=keras.optimizers.Adam(),loss=keras.losses.SparseCategoricalCrossentropy(),metrics=[keras.metrics.SparseCategoricalAccuracy()])return model

还可以使用来自 TFHub 的许多其他预训练文本嵌入向量：

  • google/nnlm-en-dim128/2 - 基于与 google/nnlm-en-dim50/2 相同的数据并使用相同的 NNLM 架构进行训练，但具有更大的嵌入向量维度。更大维度的嵌入向量可以改进您的任务，但可能需要更长的时间来训练您的模型。
  • google/nnlm-en-dim128-with-normalization/2 - 与 google/nnlm-en-dim128/2 相同，但具有额外的文本归一化，例如移除标点符号。如果您的任务中的文本包含附加字符或标点符号，这会有所帮助。
  • google/universal-sentence-encoder/4 - 一个可产生 512 维嵌入向量的更大模型，使用深度平均网络 (DAN) 编码器训练。

还有很多！在 TFHub 上查找更多文本嵌入向量模型。

3. 评估你的模型

def evaluate_model(test_data, test_labels):model = load_trained_model()# Evaluate the modelresults = model.evaluate(test_data, test_labels, verbose=2)print("Test accuracy:", results[1])def load_trained_model():# model = get_model()# model.load_weights('./models/model_new1.h5')model = tf.keras.models.load_model('models_pb')return model

4. 测试几个例子

def predict(real_data):model  = load_trained_model()probabilities = model.predict([real_data]);print("probabilities :",probabilities)result =  get_label(probabilities)return resultdef get_label(probabilities):index = np.argmax(probabilities[0])print("index :" + str(index))result_str =  index_dic.get(str(index))# result_str = list(index_dic.keys())[list(index_dic.values()).index(index)]return result_strdef predict_my_module():# review = "I don't like it"# review = "this is bad movie "# review = "This is good movie"review = " this is terrible movie"# review = "This isn‘t great movie"# review = "i think this is bad movie"# review = "I'm not very disappoint for this movie"# review = "I'm not very disappoint for this movie"# review = "I am very happy for this movie"#neg:0 postive:1s = predict(review)print(s)if __name__ == '__main__':x_train, y_train, x_test, y_test = get_dataset_to_train()model = get_model()model = train(model, x_train, y_train, x_test, y_test)evaluate_model(x_test, y_test)predict_my_module()

完整代码

import numpy as np
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense, Dropout
import keras as keras
from keras.callbacks import EarlyStopping, ModelCheckpoint
import tensorflow_hub as hubembedding_url = "https://tfhub.dev/google/nnlm-en-dim50/2"index_dic = {"0":"negative", "1": "positive"}def get_dataset_to_train():train_test = np.load('dataset/train_test.npz', allow_pickle=True)x_train =  train_test['X_train']y_train = train_test['y_train']x_test =  train_test['X_test']y_test = train_test['y_test']return x_train, y_train, x_test, y_testdef get_model():hub_layer = hub.KerasLayer(embedding_url, input_shape=[], dtype=tf.string, trainable=True)# Build the modelmodel = Sequential([hub_layer,Dense(16, activation='relu'),Dropout(0.5),Dense(2, activation='softmax')])print(model.summary())model.compile(optimizer=keras.optimizers.Adam(),loss=keras.losses.SparseCategoricalCrossentropy(),metrics=[keras.metrics.SparseCategoricalAccuracy()])return modeldef train(model , train_data, train_labels, test_data, test_labels):# train_data, train_labels, test_data, test_labels = get_dataset_to_train()train_data = [tf.compat.as_str(tf.compat.as_bytes(str(x))) for x in train_data]test_data = [tf.compat.as_str(tf.compat.as_bytes(str(x))) for x in test_data]train_data = np.asarray(train_data)  # Convert to numpy arraytest_data = np.asarray(test_data)  # Convert to numpy arrayprint(train_data.shape, test_data.shape)early_stop = EarlyStopping(monitor='val_sparse_categorical_accuracy', patience=4, mode='max', verbose=1)# 定义ModelCheckpoint回调函数# checkpoint = ModelCheckpoint( './models/model_new1.h5', monitor='val_sparse_categorical_accuracy', save_best_only=True,#                              mode='max', verbose=1)checkpoint_pb = ModelCheckpoint(filepath="./models_pb/",  monitor='val_sparse_categorical_accuracy', save_weights_only=False, save_best_only=True)history = model.fit(train_data[:2000], train_labels[:2000], epochs=45, batch_size=45, validation_data=(test_data, test_labels), shuffle=True,verbose=1, callbacks=[early_stop, checkpoint_pb])print("history", history)return modeldef evaluate_model(test_data, test_labels):model = load_trained_model()# Evaluate the modelresults = model.evaluate(test_data, test_labels, verbose=2)print("Test accuracy:", results[1])def predict(real_data):model  = load_trained_model()probabilities = model.predict([real_data]);print("probabilities :",probabilities)result =  get_label(probabilities)return resultdef get_label(probabilities):index = np.argmax(probabilities[0])print("index :" + str(index))result_str =  index_dic.get(str(index))# result_str = list(index_dic.keys())[list(index_dic.values()).index(index)]return result_strdef load_trained_model():# model = get_model()# model.load_weights('./models/model_new1.h5')model = tf.keras.models.load_model('models_pb')return modeldef predict_my_module():# review = "I don't like it"# review = "this is bad movie "# review = "This is good movie"review = " this is terrible movie"# review = "This isn‘t great movie"# review = "i think this is bad movie"# review = "I'm not very disappoint for this movie"# review = "I'm not very disappoint for this movie"# review = "I am very happy for this movie"#neg:0 postive:1s = predict(review)print(s)if __name__ == '__main__':x_train, y_train, x_test, y_test = get_dataset_to_train()model = get_model()model = train(model, x_train, y_train, x_test, y_test)evaluate_model(x_test, y_test)predict_my_module()

http://www.zhongyajixie.com/news/44722.html

相关文章：

  • 武汉手机网站建设咨询关键词排名怎么做上去
  • 做网站挣钱快吗网站推广的渠道有
  • 常用个人网站中国新闻网
  • 网站定制开发微信运营色盲能治好吗
  • 上海地区网站备案网络营销方式有哪几种
  • 济南 网站建设公司 医疗外贸公司如何做推广
  • 湖北省住房和城乡建设厅官方网站打开百度搜索引擎
  • 如何开发自己的app龙岩seo
  • 成都网站建设排名百度推广是什么工作
  • 05网电子书国外网站谷歌seo推广
  • 做婚纱网站的意义网站维护的主要内容
  • 网站首页轮播图片做网站流程
  • 孝感网站开发优搏快有哪些平台可以免费发广告
  • 深圳网站建设 网络推广关键词排名推广公司
  • html 网站建设中百度快照是怎么做上去的
  • 博兴网站建设seo搜索引擎优化视频
  • 商务网站规划与建设心得淘宝流量助手平台
  • 万能素材网站下载百度推广效果不好怎么办
  • 东风地区网站建设价格低上海企业网站推广
  • 一搜网站制作crm软件
  • 广东网站建设联系电话google年度关键词
  • 二手房发布网站怎么做营销渠道名词解释
  • 厦门做网站多南京关键词seo公司
  • 山东住房与城乡建设网站百度推广登陆首页
  • 徐州网站制作费用百度网站优化排名
  • 上海公司注册一网通办人员优化方案怎么写
  • 卢湾做网站网络营销包括的主要内容有
  • 做汽车新闻哪个网站好windows优化大师电脑版
  • 阿里云网站建设方案书怎么写搜索网站的软件
  • b2b电子商务网站开发应用商店aso