def fit_lstm_model(X_train, y_train, n_words, n_tags, seq_len, class_weights,
epochs):
'''Set up LSTM model with one input - equal length sequences of encoded text'''
input_seq = Input(shape=(seq_len, ))
'''Pass the GloVe pretrained model weights into the embedding layer'''
embedding = Embedding(input_dim=n_words,
output_dim=300,
weights=[embedding_matrix],
trainable=True)(input_seq)
embedding = Dropout(0.1)(embedding)
'''Add Bidirectional LSTM layer, dense hidden layer, and final output layer'''
model = Bidirectional(
LSTM(units=64, return_sequences=True,
recurrent_dropout=0.1))(embedding)
model = TimeDistributed(Dense(64, activation='relu'))(model)
output = Dense(n_tags, activation="softmax")(model)
'''Compile and fit deep neural network'''
model = Model(inputs=input_seq, outputs=output)
model.compile(optimizer="adam",
loss="categorical_crossentropy",
metrics=["accuracy"])
history = model.fit(X_train,
y_train,
epochs=epochs,
batch_size=32,
validation_split=0.1,
verbose=1,
class_weight=[class_weights])
'''Create simple performance report for the model'''
val_loss, val_acc = model.evaluate(X_test, y_test)
print(f'Model validation loss was {val_loss}')
print(f'Model validation accuracy was {val_acc}')
return model, history
def train(self, epochs, embedding=None):
# Embedded Words
txt_input = Input(shape=(None, ), name='txt_input')
txt_embed = Embedding(input_dim=self.num_words,
output_dim=MAX_LEN,
input_length=None,
name='txt_embedding',
trainable=False,
weights=([embedding]))(txt_input)
txt_drpot = Dropout(0.1, name='txt_dropout')(txt_embed)
# Embedded Part of Speech
pos_input = Input(shape=(None, ), name='pos_input')
pos_embed = Embedding(input_dim=self.num_pos,
output_dim=MAX_LEN,
input_length=None,
name='pos_embedding')(pos_input)
pos_drpot = Dropout(0.1, name='pos_dropout')(pos_embed)
# Embedded Characters
char_in = Input(shape=(
None,
MAX_LEN_CHAR,
), name="char_input")
emb_char = TimeDistributed(
Embedding(input_dim=self.num_chars,
output_dim=MAX_LEN_CHAR,
input_length=None))(char_in)
char_enc = TimeDistributed(
LSTM(units=20, return_sequences=False,
recurrent_dropout=0.5))(emb_char)
# Concatenate inputs
x = concatenate([txt_drpot, pos_drpot, char_enc], axis=2)
x = SpatialDropout1D(0.3)(x)
# Deep Layers
model = Bidirectional(
LSTM(units=100, return_sequences=True, recurrent_dropout=0.1))(x)
model = Bidirectional(
LSTM(units=100, return_sequences=True,
recurrent_dropout=0.1))(model)
# Output
out = TimeDistributed(Dense(self.num_entities,
activation="softmax"))(model)
model = Model(inputs=[txt_input, pos_input, char_in], outputs=[out])
model.compile(optimizer="rmsprop",
loss='categorical_crossentropy',
metrics=['accuracy'])
plot_model(model, to_file=self.save_path + 'model_structure.png')
print(model.summary())
history = model.fit(
[self.X_train, self.train_pos, self.train_characters],
np.array(self.Y_train),
batch_size=32,
epochs=epochs,
validation_data=([
self.X_validation, self.valid_pos, self.valid_characters
], np.array(self.Y_validation)),
verbose=1)
model.save(self.save_path + 'model_ner')
test_eval = model.evaluate(
[self.X_test, self.test_pos, self.test_characters],
np.array(self.Y_test))
print('Test loss:', test_eval[0])
print('Test accuracy:', test_eval[1])
return model, history
# set titles
sub_fig1.set_title('Accuracy')
sub_fig2.set_title('Loss')
print(hist)
# set values and labels
sub_fig1.plot(hist["crf_viterbi_accuracy"],label='acc')
sub_fig1.plot(hist["val_crf_viterbi_accuracy"], label='val_acc')
sub_fig1.legend(loc="lower right")
sub_fig2.plot(hist["loss"],label='loss')
sub_fig2.plot(hist["val_loss"],label='val_loss')
sub_fig2.legend(loc="upper right")
plt.xlabel('epoch')
# show figure
plt.show()
score = model.evaluate([X_w_te,np.array(X_c_te).reshape((len(X_c_te), max_len, max_len_char))], np.array(y_te), batch_size=batch_size,verbose=1)
print(model.metrics_names)
print("Score:")
print(score)
# ## Prediction on test set
from seqeval.metrics import precision_score, recall_score, f1_score, classification_report
# print("Input:")
# print(X_te[0])
# print("Supposed output:")
# print(y_te)
# print(np.array(y_te))
test_pred = model.predict([X_w_te,np.array(X_c_te).reshape((len(X_c_te), max_len, max_len_char))], verbose=1)
# print("Prediction result:")
# print(test_pred[0])
idx2tag = {i: w for w,i in tags2idx.items()}