def lstm():
data, targets, filenames, embedding_matrix, word_index = preprocess_embedding()
EMBEDDING_DIM = 300
MAX_SEQUENCE_LENGTH = 50
embedding_layer = Embedding(len(word_index) + 1,
EMBEDDING_DIM,
weights=[embedding_matrix],
input_length=MAX_SEQUENCE_LENGTH,
trainable= False,
name='layer_embedding') #mask_zero=True,
sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')
embedded_sequences = embedding_layer(sequence_input)
x1 = LSTM(150, return_sequences=True,name='lstm_1')(embedded_sequences)
#x2 = LSTM(75, return_sequences=True,name='lstm_2')(x1)
encoded = LSTM(30,name='lstm_3')(x1)
x3 = RepeatVector(MAX_SEQUENCE_LENGTH,name='layer_repeat')(encoded)
# x4 = LSTM(75, return_sequences=True,name='lstm_4')(x3)
x5 = LSTM(150, return_sequences=True,name='lstm_5')(x3)
decoded = LSTM(300, return_sequences=True,activation='linear',name='lstm_6')(x5)
sequence_autoencoder = Model(sequence_input, decoded)
#print sequence_autoencoder.get_layer('lstm_6').output
encoder = Model(sequence_input, encoded)
sequence_autoencoder.compile(loss='cosine_proximity',
optimizer='sgd')#, metrics=['acc'])
embedding_layer = Model(inputs=sequence_autoencoder.input,
outputs=sequence_autoencoder.get_layer('layer_embedding').output)
sequence_autoencoder.fit(data, embedding_layer.predict(data), epochs=5)
# for i in sequence_autoencoder.layers[3].get_weights()[0]:
# print i
#
# print sequence_autoencoder.layers[3].get_weights()[0][1]
# print sequence_autoencoder.layers[1].get_weights()[0][1].shape
# print sequence_autoencoder.layers[2].get_weights()[0][1].shape
# print sequence_autoencoder.layers[3].get_weights()[0][1].shape
# print sequence_autoencoder.layers[4].get_weights()[0][1].shape
# #print sequence_autoencoder.layers[5].get_weights()[0][1].shape
# print sequence_autoencoder.layers[6].get_weights()[0][1].shape
# print sequence_autoencoder.layers[7].get_weights()[0][1].shape
csvname = 'lstm_autoencoder_weight'
write_vec_to_csv(sequence_autoencoder.layers[3].get_weights()[0],targets,filenames,csvname)
#Custom metric calculations
metrics = Metrics_Approx(tag_index=all_labels, k=k)
#Train model
if not trans_learn:
history = model.fit(X_train,
np.array(y_train),
batch_size=batch_size,
epochs=int(args.num_epochs),
verbose=1,
callbacks=[metrics],
validation_data=(X_test, np.array(y_test)))
#Perform prediction to calculate scores for K iteration
y_pred = np.asarray(model.predict(X_test))
y_pred = np.argmax(y_pred, axis=-1)
targ = y_test
targ = np.argmax(targ, axis=-1)
k_scores = evaluate_approx_match(y_pred, targ, all_labels)
k_metrics_val = model.evaluate(X_test, y_test)
k_metrics_train = model.evaluate(X_train, y_train)
#Get timestamp for log
timestamp = str(datetime.datetime.now())
#Print scores for current K iteration
with open(args.log_name, 'a', encoding='UTF-8') as writer:
writer.write(timestamp + "|" + str(k) + "|" + str(len(X_train)) + "|" +
str(k_scores["p"]) + "|" + str(k_scores["r"]) + "|" +
str(k_scores["f1"]) + "|" + str(k_metrics_val[0]) + "|" +
seq.append(word2index["UNK"])
XX = sequence.pad_sequences(sequences=[seq], maxlen=MAX_SENTENCE_LENGTH) # list假設多個，所以這邊要放多層
label_pred = int(round(model_LSTM.predict(XX)[0][0]))
label2word = {1: 'Positive', 0: 'Negative'}
print("Your sentence: '{}' belongs to {} emotion.".format(INPUT_SENTENCES, label2word[label_pred]))
interact_pred()
# 多個句子
INPUT_SENTENCES = ['I love reading.', 'You are so boring.']
XX = np.empty(len(INPUT_SENTENCES), dtype=list)
i = 0
for sentence in INPUT_SENTENCES:
words = nltk.word_tokenize(sentence.lower())
seq = []
for word in words:
if word in word2index:
seq.append(word2index[word])
else:
seq.append(word2index['UNK'])
XX[i] = seq
i += 1
XX = sequence.pad_sequences(XX, maxlen=MAX_SENTENCE_LENGTH)
labels = [int(round(x[0])) for x in model.predict(XX)]
label2word = {1: '积极', 0: '消极'}
for i in range(len(INPUT_SENTENCES)):
print('{} {}'.format(label2word[labels[i]], INPUT_SENTENCES[i]))
#build model
input = Input(shape=(X_train.shape[1],))
model = Embedding(input_dim=vocab_size, output_dim=word_embed_size,
input_length=seq_maxlen, weights=[embedding_weight_matrix],
trainable = False) (input)
model = LSTM(100, return_sequences=True)(model)
output = TimeDistributed(Dense(num_tags, activation="softmax"))(model)
model = Model(inputs=input, outputs=output)
print(model.summary())
model.compile(optimizer="rmsprop", loss="categorical_crossentropy", metrics=["accuracy"])
save_weights = ModelCheckpoint('model.h5', monitor='val_loss', save_best_only=True)
model.fit(x=X_train, y=y_train, batch_size=batch_size,
epochs=epochs, validation_data=(X_validation, y_validation), callbacks=[save_weights])
sentences_test, tags_test = load_data(os.path.join(data_dir, 'test.txt'))
X_test = getSequences(sentences_test, tokenizer_words, seq_maxlen, 0)
y_test = getSequences(tags_test, tokenizer_tags, seq_maxlen, tokenizer_tags.word_index['o'])
y_pred = model.predict(X_test)
y_pred_tags, y_test_tags = decode_output(X_test, y_pred, y_test, tokenizer_words, tokenizer_tags, seq_maxlen)
report = classification_report(y_pred=y_pred_tags, y_true=y_test_tags)
print(report)
def lstm():
data, targets, filenames, embedding_matrix, word_index = preprocess_embedding(
)
EMBEDDING_DIM = 300
MAX_SEQUENCE_LENGTH = 500
keras.callbacks.TensorBoard(log_dir='./Graph_lstm_embedding',
histogram_freq=0,
write_graph=True,
write_images=True)
tbCallBack = keras.callbacks.TensorBoard(
log_dir='./Graph_lstm_embedding',
histogram_freq=10,
embeddings_layer_names='layer_embedding',
embeddings_freq=100,
write_graph=True,
write_images=True)
embedding_layer = Embedding(len(word_index) + 1,
EMBEDDING_DIM,
weights=[embedding_matrix],
input_length=MAX_SEQUENCE_LENGTH,
trainable=False,
name='layer_embedding') #mask_zero=True,
# STep 1 Training
sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH, ), dtype='int32')
embedded_sequences = embedding_layer(sequence_input)
#x1 = LSTM(150, return_sequences=True,name='lstm_1')(embedded_sequences) # [ batchsize, timesteps, input_dimension ]
# [batchsize, timesteps, output_dimension ]
#x2 = LSTM(75, return_sequences=True,name='lstm_2')(x1)
encoded = LSTM(300, name='lstm_3')(embedded_sequences)
x3 = RepeatVector(MAX_SEQUENCE_LENGTH, name='layer_repeat')(encoded)
# x4 = LSTM(75, return_sequences=True,name='lstm_4')(x3)
#x5 = LSTM(150, return_sequences=True,name='lstm_5')(x3)
decoded = LSTM(300,
return_sequences=True,
activation='softmax',
name='lstm_6')(x3)
sequence_autoencoder = Model(sequence_input, decoded)
#print sequence_autoencoder.get_layer('lstm_6').output
encoder = Model(sequence_input, encoded) # two functions that you learn
sequence_autoencoder.compile(loss='cosine_proximity',
optimizer='sgd') #, metrics=['acc'])
embedding_layer = Model(
inputs=sequence_autoencoder.input,
outputs=sequence_autoencoder.get_layer('layer_embedding').output)
sequence_autoencoder.fit(data,
embedding_layer.predict(data),
epochs=20,
callbacks=[tbCallBack])
# Training is done
# define the encoding function using the trained encoded weights
#encoder = Model(sequence_input, encoded) # a function that you learn
print '**************************************************'
encoded_data = encoder.predict(data)
print encoded_data
print '**************************************************'
print encoded_data.shape
csvname = 'lstm_autoencoder_representation'
write_vec_to_csv(encoded_data, targets, filenames, csvname)