pytorch+lstm实现的pos示例
学了几天终于大概明白pytorch怎么用了
这个是直接搬运的官方文档的代码
之后会自己试着实现其他nlp的任务
#Author:RobertGuthrie
importtorch
importtorch.autogradasautograd
importtorch.nnasnn
importtorch.nn.functionalasF
importtorch.optimasoptim
torch.manual_seed(1)
lstm=nn.LSTM(3,3)#Inputdimis3,outputdimis3
inputs=[autograd.Variable(torch.randn((1,3)))
for_inrange(5)]#makeasequenceoflength5
#initializethehiddenstate.
hidden=(autograd.Variable(torch.randn(1,1,3)),
autograd.Variable(torch.randn((1,1,3))))
foriininputs:
#Stepthroughthesequenceoneelementatatime.
#aftereachstep,hiddencontainsthehiddenstate.
out,hidden=lstm(i.view(1,1,-1),hidden)
#alternatively,wecandotheentiresequenceallatonce.
#thefirstvaluereturnedbyLSTMisallofthehiddenstatesthroughout
#thesequence.thesecondisjustthemostrecenthiddenstate
#(comparethelastsliceof"out"with"hidden"below,theyarethesame)
#Thereasonforthisisthat:
#"out"willgiveyouaccesstoallhiddenstatesinthesequence
#"hidden"willallowyoutocontinuethesequenceandbackpropagate,
#bypassingitasanargumenttothelstmatalatertime
#Addtheextra2nddimension
inputs=torch.cat(inputs).view(len(inputs),1,-1)
hidden=(autograd.Variable(torch.randn(1,1,3)),autograd.Variable(
torch.randn((1,1,3))))#cleanouthiddenstate
out,hidden=lstm(inputs,hidden)
#print(out)
#print(hidden)
#准备数据
defprepare_sequence(seq,to_ix):
idxs=[to_ix[w]forwinseq]
tensor=torch.LongTensor(idxs)
returnautograd.Variable(tensor)
training_data=[
("Thedogatetheapple".split(),["DET","NN","V","DET","NN"]),
("Everybodyreadthatbook".split(),["NN","V","DET","NN"])
]
word_to_ix={}
forsent,tagsintraining_data:
forwordinsent:
ifwordnotinword_to_ix:
word_to_ix[word]=len(word_to_ix)
print(word_to_ix)
tag_to_ix={"DET":0,"NN":1,"V":2}
#Thesewillusuallybemorelike32or64dimensional.
#Wewillkeepthemsmall,sowecanseehowtheweightschangeaswetrain.
EMBEDDING_DIM=6
HIDDEN_DIM=6
#继承自nn.module
classLSTMTagger(nn.Module):
def__init__(self,embedding_dim,hidden_dim,vocab_size,tagset_size):
super(LSTMTagger,self).__init__()
self.hidden_dim=hidden_dim
#一个单词数量到embedding维数的矩阵
self.word_embeddings=nn.Embedding(vocab_size,embedding_dim)
#传入两个维度参数
#TheLSTMtakeswordembeddingsasinputs,andoutputshiddenstates
#withdimensionalityhidden_dim.
self.lstm=nn.LSTM(embedding_dim,hidden_dim)
#线性layer从隐藏状态空间映射到tag便签
#Thelinearlayerthatmapsfromhiddenstatespacetotagspace
self.hidden2tag=nn.Linear(hidden_dim,tagset_size)
self.hidden=self.init_hidden()
definit_hidden(self):
#Beforewe'vedoneanything,wedonthaveanyhiddenstate.
#RefertothePytorchdocumentationtoseeexactly
#whytheyhavethisdimensionality.
#Theaxessemanticsare(num_layers,minibatch_size,hidden_dim)
return(autograd.Variable(torch.zeros(1,1,self.hidden_dim)),
autograd.Variable(torch.zeros(1,1,self.hidden_dim)))
defforward(self,sentence):
embeds=self.word_embeddings(sentence)
lstm_out,self.hidden=self.lstm(embeds.view(len(sentence),1,-1),self.hidden)
tag_space=self.hidden2tag(lstm_out.view(len(sentence),-1))
tag_scores=F.log_softmax(tag_space)
returntag_scores
#embedding维度，hidden维度，词语数量，标签数量
model=LSTMTagger(EMBEDDING_DIM,HIDDEN_DIM,len(word_to_ix),len(tag_to_ix))
#optim中存了各种优化算法
loss_function=nn.NLLLoss()
optimizer=optim.SGD(model.parameters(),lr=0.1)
#Seewhatthescoresarebeforetraining
#Notethatelementi,joftheoutputisthescorefortagjforwordi.
inputs=prepare_sequence(training_data[0][0],word_to_ix)
tag_scores=model(inputs)
print(tag_scores)
forepochinrange(300):#again,normallyyouwouldNOTdo300epochs,itistoydata
forsentence,tagsintraining_data:
#Step1.RememberthatPytorchaccumulatesgradients.
#Weneedtoclearthemoutbeforeeachinstance
model.zero_grad()
#Also,weneedtoclearoutthehiddenstateoftheLSTM,
#detachingitfromitshistoryonthelastinstance.
model.hidden=model.init_hidden()
#Step2.Getourinputsreadyforthenetwork,thatis,turntheminto
#Variablesofwordindices.
sentence_in=prepare_sequence(sentence,word_to_ix)
targets=prepare_sequence(tags,tag_to_ix)
#Step3.Runourforwardpass.
tag_scores=model(sentence_in)
#Step4.Computetheloss,gradients,andupdatetheparametersby
#callingoptimizer.step()
loss=loss_function(tag_scores,targets)
loss.backward()
optimizer.step()
#Seewhatthescoresareaftertraining
inputs=prepare_sequence(training_data[0][0],word_to_ix)
tag_scores=model(inputs)
#Thesentenceis"thedogatetheapple".i,jcorrespondstoscorefortagj
#forwordi.Thepredictedtagisthemaximumscoringtag.
#Here,wecanseethepredictedsequencebelowis01201
#since0isindexofthemaximumvalueofrow1,
#1istheindexofmaximumvalueofrow2,etc.
#WhichisDETNOUNVERBDETNOUN,thecorrectsequence!
print(tag_scores)
以上这篇pytorch+lstm实现的pos示例就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持毛票票。
声明：本文内容来源于网络，版权归原作者所有，内容由互联网用户自发贡献自行上传，本网站不拥有所有权，未作人工编辑处理，也不承担相关法律责任。如果您发现有涉嫌版权的内容，欢迎发送邮件至：czq8825#qq.com（发邮件时，请将#更换为@）进行举报，并提供相关证据，一经查实，本站将立刻删除涉嫌侵权内容。