Stacked Bidirectional LSTM in Keras

In [1]:

    

import keras
from keras.datasets import imdb
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Dense, Dropout, Embedding, SpatialDropout1D, LSTM
from keras.layers.wrappers import Bidirectional 
from keras.callbacks import ModelCheckpoint
import os
from sklearn.metrics import roc_auc_score 
import matplotlib.pyplot as plt 
%matplotlib inline


    

    




Using TensorFlow backend.

In [2]:

    

# output directory name:
output_dir = 'model_output/stackedLSTM'

# training:
epochs = 4
batch_size = 128

# vector-space embedding: 
n_dim = 64 
n_unique_words = 10000 
max_review_length = 200 
pad_type = trunc_type = 'pre'
drop_embed = 0.2 

# LSTM layer architecture:
n_lstm_1 = 64 # lower
n_lstm_2 = 64 # new!
drop_lstm = 0.2


    

In [3]:

    

(x_train, y_train), (x_valid, y_valid) = imdb.load_data(num_words=n_unique_words) # removed n_words_to_skip


    

In [4]:

    

x_train = pad_sequences(x_train, maxlen=max_review_length, padding=pad_type, truncating=trunc_type, value=0)
x_valid = pad_sequences(x_valid, maxlen=max_review_length, padding=pad_type, truncating=trunc_type, value=0)


    

In [5]:

    

model = Sequential()
model.add(Embedding(n_unique_words, n_dim, input_length=max_review_length)) 
model.add(SpatialDropout1D(drop_embed))
model.add(Bidirectional(LSTM(n_lstm_1, dropout=drop_lstm, return_sequences=True))) # retain temporal dimension
model.add(Bidirectional(LSTM(n_lstm_2, dropout=drop_lstm)))
model.add(Dense(1, activation='sigmoid'))


    

In [6]:

    

# LSTM layer parameters double due to both reading directions
model.summary()


    

    




_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
embedding_1 (Embedding)      (None, 200, 64)           640000    
_________________________________________________________________
spatial_dropout1d_1 (Spatial (None, 200, 64)           0         
_________________________________________________________________
bidirectional_1 (Bidirection (None, 200, 128)          66048     
_________________________________________________________________
bidirectional_2 (Bidirection (None, 128)               98816     
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 129       
=================================================================
Total params: 804,993
Trainable params: 804,993
Non-trainable params: 0
_________________________________________________________________

In [7]:

    

model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])


    

In [8]:

    

modelcheckpoint = ModelCheckpoint(filepath=output_dir+"/weights.{epoch:02d}.hdf5")
if not os.path.exists(output_dir):
    os.makedirs(output_dir)


    

In [9]:

    

# 87.6% validation accuracy in epoch 2
model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, validation_data=(x_valid, y_valid), callbacks=[modelcheckpoint])


    

    




Train on 25000 samples, validate on 25000 samples
Epoch 1/4
25000/25000 [==============================] - 216s - loss: 0.4400 - acc: 0.7801 - val_loss: 0.3231 - val_acc: 0.8677
Epoch 2/4
25000/25000 [==============================] - 211s - loss: 0.2476 - acc: 0.9040 - val_loss: 0.2927 - val_acc: 0.8764
Epoch 3/4
25000/25000 [==============================] - 213s - loss: 0.1861 - acc: 0.9310 - val_loss: 0.3139 - val_acc: 0.8680
Epoch 4/4
25000/25000 [==============================] - 214s - loss: 0.1465 - acc: 0.9461 - val_loss: 0.3704 - val_acc: 0.8610






    
Out[9]:





<keras.callbacks.History at 0x7f83bb875710>

In [10]:

    

model.load_weights(output_dir+"/weights.01.hdf5") # zero-indexed


    

In [11]:

    

y_hat = model.predict_proba(x_valid)


    

    




25000/25000 [==============================] - 154s   

In [12]:

    

plt.hist(y_hat)
_ = plt.axvline(x=0.5, color='orange')


    

In [13]:

    

"{:0.2f}".format(roc_auc_score(y_valid, y_hat)*100.0)


    

    
Out[13]:





'94.78'

In [ ]: