├── README.md
└── Keras_TextCNN.py


/README.md:
--------------------------------------------------------------------------------
1 | # Keras_TextCNN
2 | Use python to learn how to use Keras to define the TextCNN model
3 | 


--------------------------------------------------------------------------------
/Keras_TextCNN.py:
--------------------------------------------------------------------------------
 1 | from keras.layers import Input, Dense, Embedding, Conv2D, MaxPool2D, Conv1D, MaxPooling1D, concatenate
 2 | from keras.layers import Reshape, Flatten, Dropout, Concatenate
 3 | from keras.callbacks import ModelCheckpoint
 4 | from keras.optimizers import Adam
 5 | from keras.models import Model
 6 | from keras import backend as K
 7 | import keras.layers as L
 8 | import keras
 9 | 
10 | from sklearn.model_selection import train_test_split
11 | 
12 | def text_cnn(sequence_length, num_classes, vocab_size, embedding_size, 
13 |              filter_sizes, num_filters, embedding_matrix, drop_out=0.5 ,l2_reg_lambda=0.0):
14 |     input_x = L.Input(shape=(sequence_length,), name='input_x')
15 |     
16 |     # embedding layer
17 |     if embedding_matrix is None:
18 |         embedding = L.Embedding(vocab_size, embedding_size, name='embedding')(input_x)
19 |     else:
20 |         embedding = L.Embedding(vocab_size, embedding_size, weights=[embedding_matrix], name='embedding')(input_x)
21 |     expend_shape = [embedding.get_shape().as_list()[1], embedding.get_shape().as_list()[2], 1]
22 |     # embedding_chars = K.expand_dims(embedding, -1)    
23 |     # 4D tensor [batch_size, seq_len, embeding_size, 1] seems like a gray picture
24 |     embedding_chars = L.Reshape(expend_shape)(embedding)
25 |     
26 |     # conv->max pool
27 |     pooled_outputs = []
28 |     for i, filter_size in enumerate(filter_sizes):
29 |         conv = L.Conv2D(filters=num_filters, 
30 |                         kernel_size=[filter_size, embedding_size],
31 |                         strides=1,
32 |                         padding='valid',
33 |                         activation='relu',
34 |                         kernel_initializer=keras.initializers.TruncatedNormal(mean=0.0, stddev=0.1),
35 |                         bias_initializer=keras.initializers.constant(value=0.1),
36 |                         name=('conv_%d' % filter_size))(embedding_chars)
37 |         # print("conv-%d: " % i, conv)
38 |         max_pool = L.MaxPool2D(pool_size=[sequence_length - filter_size + 1, 1],
39 |                                strides=(1, 1),
40 |                                padding='valid',
41 |                                name=('max_pool_%d' % filter_size))(conv)
42 |         pooled_outputs.append(max_pool)
43 |         # print("max_pool-%d: " % i, max_pool)
44 |     
45 |     # combine all the pooled features
46 |     num_filters_total = num_filters * len(filter_sizes)
47 |     h_pool = L.Concatenate(axis=3)(pooled_outputs)
48 |     h_pool_flat = L.Reshape([num_filters_total])(h_pool)
49 |     # add dropout
50 |     dropout = L.Dropout(drop_out)(h_pool_flat)
51 |     
52 |     # output layer
53 |     output = L.Dense(num_classes,
54 |                      kernel_initializer='glorot_normal',
55 |                      bias_initializer=keras.initializers.constant(0.1),
56 |                      activation='softmax',
57 |                      name='output')(dropout)
58 |     
59 |     model = keras.models.Model(inputs=input_x, outputs=output)
60 |     print('Use TextCNN model')
61 |     return model


--------------------------------------------------------------------------------