tflearn中lstm文本分类相关实现

最近偶然看到tflearn这个东西，相比tensorflow已经是对原始相关接口做了很大的封装，api通俗易懂去源码写的还真是不错，本例子实现一个lstm网络，看看吧，就几十行代码就可以实现，要部署到线上也应该及其简单，对于tensorflow不熟悉的人可以说是个很好的东西：

   
   
import os
import tflearn
print(os.path.dirname(__file__))
from   tflearn.data_utils import  to_categorical,pad_sequences
from  tflearn.datasets  import  imdb
"""该数据集如果没有的话会自动下载
    如果程序下载失败，建议用命令行下载
    #curl -O "http://www.iro.umontreal.ca/~lisa/deep/data/imdb.pkl"
"""
train,test,_=imdb.load_data(path="imdb.pkl",n_words=10000,valid_portion=0.1)
trainx,trainy=train
testx,testy=test
print(trainx[0:2])
print(testx[0:2])
"""
pad_sequences是一个把序列转化为固定长度的函数，可以去看源码，有两种展开方式，
第一种向前填充，一种是可以向后填充，填充值可以自己指定，可以指定是0，也可以指定是1
主要这里的输入是一个int类型的矩阵，
"""
trainx=pad_sequences(trainx,maxlen=100,value=0.)
testx=pad_sequences(testx,maxlen=100,value=0.)
"""
to_categorical 是一个把分类弄成[0,1]
                              [1,0]的形式
"""
trainy=to_categorical(trainy,nb_classes=2)
testy=to_categorical(testy,nb_classes=2)
"""
输出指定数据大小，前面已经把固定长度弄成100，这里None是批次
"""
net=tflearn.input_data([None,100])
"""
input_dim就是看分语料中取了多少个单词，一般取top多少
"""
net=tflearn.embedding(net,input_dim=10000,output_dim=128)
"""
这里n_units=128就是网络大小情况  dropout=0.8
"""
net=tflearn.lstm(net,128,dropout=0.8)
"""
输出目标是2个
"""
net = tflearn.fully_connected(net, 2, activation='softmax')
net = tflearn.regression(net, optimizer='adam', learning_rate=0.001,
                         loss='categorical_crossentropy')
# Training
#模型初始化
model = tflearn.DNN(net, tensorboard_verbose=0,tensorboard_dir="/tmp/tflearn_logs/")
"""
show_metric=True可以看到过程中的准确率
"""
show_metric=True
model.fit(trainx, trainy, validation_set=(testx, testy), show_metric=True,
          batch_size=32)
#model.predict()

运行结果情况：

   
   
Training Step: 7032  | total loss: 0.12957 | time: 92.288s
| Adam | epoch: 010 | loss: 0.12957 - acc: 0.9725 -- iter: 22272/22500
0.11797 | time: 92.396s
| Adam | epoch: 010 | loss: 0.11797 - acc: 0.9752 -- iter: 22304/22500
Training Step: 7034  | total loss: 0.11303 | time: 92.513s
| Adam | epoch: 010 | loss: 0.11303 - acc: 0.9746 -- iter: 22336/22500
Training Step: 7035  | total loss: 0.11608 | time: 92.632s
| Adam | epoch: 010 | loss: 0.11608 - acc: 0.9740 -- iter: 22368/22500
Training Step: 7036  | total loss: 0.10833 | time: 92.764s
| Adam | epoch: 010 | loss: 0.10833 - acc: 0.9766 -- iter: 22400/22500
Training Step: 7037  | total loss: 0.09968 | time: 92.877s
| Adam | epoch: 010 | loss: 0.09968 - acc: 0.9789 -- iter: 22432/22500
Training Step: 7038  | total loss: 0.09557 | time: 93.015s
| Adam | epoch: 010 | loss: 0.09557 - acc: 0.9779 -- iter: 22464/22500
Training Step: 7039  | total loss: 0.09463 | time: 93.138s
| Adam | epoch: 010 | loss: 0.09463 - acc: 0.9739 -- iter: 22496/22500
Training Step: 7040  | total loss: 0.09200 | time: 95.779s
| Adam | epoch: 010 | loss: 0.09200 - acc: 0.9734 | val_loss: 0.76267 - val_acc: 0.7944 -- iter: 22500/22500
--

查看tensorboard的情况:

  
  
tensorboard --logdir="/tmp/tflearn_logs/"