怎样做商业网站平台宽带营销策略

# 本段代码构建类BiLSTM, 完成初始化和网络结构的搭建
# 总共3层: 词嵌入层, 双向LSTM层, 全连接线性层

# 本段代码构建类BiLSTM, 完成初始化和网络结构的搭建
# 总共3层: 词嵌入层, 双向LSTM层, 全连接线性层
import torch
import torch.nn as nn# 本函数实现将中文文本映射为数字化张量
def sentence_map(sentence_list, char_to_id, max_length):"""将句子中的每一个字符映射到码表中:param sentence_list: 待映射的句子，类型为字符串或列表:param char_to_id: 码表，类型为字典，格式为格式为{"字1": 1, "字2": 2}，例如：码表与id对照：char_to_id = {"双": 0, "肺": 1, "见": 2, "多": 3, "发": 4, "斑": 5, "片": 6,"状": 7, "稍": 8, "高": 9, "密": 10, "度": 11, "影": 12, "。": 13}:param max_length::return: 每一个字对应的编码，类型为tensor"""# 字符串按照逆序进行排序，不是必须操作sentence_list.sort(key=lambda c:len(c), reverse = True)# 定义句子映射列表sentence_map_list = []for sentence in sentence_list:# 生成句子中每个字对应的id列表sentence_id_list =[char_to_id[c] for c in sentence]# 计算所要填充0的长度padding = [0] * (max_length-len(sentence))# 组合sentence_map_list.append(sentence_id_list)# 返回句子映射集合，转为标量return torch.tensor(sentence_map_list, dtype= torch.long)class BiLSTM(nn.Module):"""BiLSTM模型定义"""def __init__(self, vocab_size, tag_to_id, input_feature_size, hidden_size,batch_size, sentence_length, num_layers=1, batch_first=True):"""description: 模型初始化:param vocab_size:          所有句子包含字符大小:param tag_to_id:           标签与 id 对照:param input_feature_size:  字嵌入维度( 即LSTM输入层维度 input_size ):param hidden_size:         隐藏层向量维度:param batch_size:          批训练大小:param sentence_length      句子长度:param num_layers:          堆叠 LSTM 层数:param batch_first:         是否将batch_size放置到矩阵的第一维度"""# 类继承初始化函数super(BiLSTM, self).__init__()# 设置标签与id对照self.tag_to_id = tag_to_id# 设置标签大小, 对应BiLSTM最终输出分数矩阵宽度self.tag_size = len(tag_to_id)# 设定LSTM输入特征大小, 对应词嵌入的维度大小self.embedding_size = input_feature_size# 设置隐藏层维度, 若为双向时想要得到同样大小的向量, 需要除以2self.hidden_size = hidden_size // 2# 设置批次大小, 对应每个批次的样本条数, 可以理解为输入张量的第一个维度self.batch_size = batch_size# 设定句子长度self.sentence_length = sentence_length# 设定是否将batch_size放置到矩阵的第一维度, 取值True, 或Falseself.batch_first = batch_first# 设置网络的LSTM层数self.num_layers = num_layers"""构建词嵌入层: 字向量, 维度为总单词数量与词嵌入维度参数: 总体字库的单词数量, 每个字被嵌入的维度"""self.embedding = nn.Embedding(vocab_size, self.embedding_size)self.bilstm = nn.LSTM(input_size=input_feature_size,hidden_size=self.hidden_size,num_layers=num_layers,bidirectional=True,batch_first=batch_first)# 构建全连接线性层: 将BiLSTM的输出层进行线性变换self.linear = nn.Linear(hidden_size, self.tag_size)print("=" * 100)
# 参数1:码表与id对照
char_to_id = {"双": 0, "肺": 1, "见": 2, "多": 3, "发": 4, "斑": 5, "片": 6,"状": 7, "稍": 8, "高": 9, "密": 10, "度": 11, "影": 12, "。": 13}# 参数2:标签码表对照
tag_to_id = {"O": 0, "B-dis": 1, "I-dis": 2, "B-sym": 3, "I-sym": 4}
# 参数3:字向量维度
EMBEDDING_DIM = 200
# 参数4:隐层维度
HIDDEN_DIM = 100
# 参数5:批次大小
BATCH_SIZE = 8
# 参数6:句子长度
SENTENCE_LENGTH = 20
# 参数7:堆叠 LSTM 层数
NUM_LAYERS = 1# 初始化模型
"""
model = BiLSTM(vocab_size=len(char_to_id),tag_to_id=tag_to_id,input_feature_size=EMBEDDING_DIM,hidden_size=HIDDEN_DIM,batch_size= BATCH_SIZE,sentence_length= SENTENCE_LENGTH,num_layers=NUM_LAYERS)print(model)
"""