,emb_size = 128
,hidden_size = 128
,nb_filter = 250
,filter_length = 3
,pool_length = 2
,init_type = 'random'
,emb_file = "../data/unlabeled_corpus.vec"
,tune_emb = True
)
options,args = parser.parse_args(sys.argv)
print('Loading data...')
(X_train, y_train), (X_test, y_test), (X_dev, y_dev), max_features, E, label_id = aidr.load_and_numberize_data(path=options.data_dir,
nb_words=options.max_features, init_type=options.init_type,
embfile=options.emb_file, dev_train_merge=0, map_labels_to_five_class=0)
# print("Padding sequences....")
X_train = sequence.pad_sequences(X_train, maxlen=options.maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=options.maxlen)
X_dev = sequence.pad_sequences(X_dev, maxlen=options.maxlen)
#build model...
nb_classes = np.max(y_train) + 1
print('............................')
print(len(X_train), 'train tweets')
print(len(X_dev), 'dev tweets')
print(max_features - 3, 'vocabulary size')
,nb_filter = 250
,init_type = 'random'
,filter_length = 3
,pool_length = 2
,add_feat = 0
,maxlen = 80
,tune_emb = True
,max_features = 80
,map_class = 0
)
options,args = parser.parse_args(sys.argv)
print('Loading data...')
(X_train, y_train), (X_test, y_test), (X_dev, y_dev), vocab_size, E, label_id = aidr.load_and_numberize_data(path=options.data_dir, seed=113,
nb_words=options.max_features, init_type=options.init_type,
embfile=options.emb_file)
X_train_f, X_test_f, X_dev_f = aidr_feat.load_tfidf_vectors(path=options.data_dir, seed=113) # load features
assert len(X_train) == X_train_f.shape[0] and len(X_test) == X_test_f.shape[0]
print("Padding sequences....")
X_train = sequence.pad_sequences(X_train, maxlen=options.maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=options.maxlen)
X_dev = sequence.pad_sequences(X_dev, maxlen=options.maxlen)
#build model...
nb_classes = len(label_id)
max_features = X_train_f.shape[1]
,emb_size = 128
,hidden_size = 128
,nb_filter = 250
,filter_length = 3
,pool_length = 2
,init_type = 'random'
,emb_file = "../data/unlabeled_corpus.vec"
,tune_emb = True
)
options,args = parser.parse_args(sys.argv)
print('Loading data...')
(X_train, y_train), (X_test, y_test), (X_dev, y_dev), max_features, E, label_id = aidr.load_and_numberize_data(path=options.data_dir,
nb_words=options.max_features, init_type=options.init_type,
embfile=options.emb_file, dev_train_merge=0, map_labels_to_five_class=0)
# print("Padding sequences....")
X_train = sequence.pad_sequences(X_train, maxlen=options.maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=options.maxlen)
X_dev = sequence.pad_sequences(X_dev, maxlen=options.maxlen)
#build model...
nb_classes = np.max(y_train) + 1
print('............................')
print(len(X_train), 'train tweets')
print(len(X_test), 'test tweets')
print(len(X_dev), 'dev tweets')
,learn_alg = "adam" # sgd, adagrad, rmsprop, adadelta, adam (default)
,loss = "hinge"#"binary_crossentropy" # hinge, squared_hinge, binary_crossentropy (default)
,minibatch_size = 32
,dropout_ratio = 0.2
,epochs = 25
,hidden_size = 128
,nb_layers = 1
,model_type = 'mlp'
,add_feat = 0
,map_class = 0
)
options,args = parser.parse_args(sys.argv)
print('Loading data...')
(X_train, y_train), (X_test, y_test), (X_dev, y_dev), max_features, E, label_id = aidr.load_and_numberize_data(path=options.data_dir, seed=113)
X_train_f, X_test_f, X_dev_f = aidr_feat.load_tfidf_vectors(path=options.data_dir, seed=113) # load features
assert len(X_train) == X_train_f.shape[0] and len(X_test) == X_test_f.shape[0]
#build model...
nb_classes = len(label_id)
max_features = X_train_f.shape[1]
print('............................')
print(len(X_train), 'train tweets')
print(len(X_test), 'test tweets')
print(len(X_dev), 'dev tweets')
print(max_features, 'features')
print(nb_classes, 'different classes')
print('............................')
,
recur_type='lstm' # gru, simplernn, lstm (default)
,
init_type=
'conv_glove' # 'random', 'word2vec', 'glove', 'conv_word2vec', 'conv_glove', 'meta_conv', 'meta_orig'
,
emb_file="../data/unlabeled_corpus.vec",
tune_emb=True,
map_class=1,
numClasses=5,
evalMinibatches=100)
options, args = parser.parse_args(sys.argv)
(X_train, y_train), (X_test, y_test), (X_dev, y_dev), max_features, E, label_id, sequence_len = \
aidr.load_and_numberize_data(path=options.data_dir, nb_words=options.max_features, maxlen=options.maxlen,
init_type=options.init_type,
dev_train_merge=1, embfile=None, map_labels_to_five_class=options.map_class)
# Placeholders
input_data = tf.placeholder(tf.int32, [None, options.maxlen],
name="input_data")
sequence_lengths = tf.placeholder(tf.int32,
shape=[None],
name="sequence_lengths")
y_values = tf.placeholder(tf.int32, [None])
labels = tf.one_hot(y_values, options.numClasses)
prediction = forward_propagation_bidirectional(input_data,
sequence_lengths, E)
#prediction = forward_propagation_unidirectional(input_data, sequence_lengths, E)
#prediction = forward_propagation_averaging(input_data, sequence_lengths, E)
pool_length=2,
init_type='random',
emb_file="../data/unlabeled_corpus.vec",
tune_emb=True)
options, args = parser.parse_args(sys.argv)
#print('Loading data...')
print('LOADING DATA...')
print('----------------------------------------------------------------')
(X_train, y_train), (X_test, y_test), (X_validate, y_validate), (
X_newinput,
y_newinput), max_features, E, label_id = aidr.load_and_numberize_data(
path=options.data_dir,
nb_words=options.max_features,
init_type=options.init_type,
embfile=options.emb_file,
validate_train_merge=0,
map_labels_to_five_class=0)
# print("Padding sequences....")
X_train = sequence.pad_sequences(X_train, maxlen=options.maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=options.maxlen)
X_validate = sequence.pad_sequences(X_validate, maxlen=options.maxlen)
X_newinput = sequence.pad_sequences(
X_newinput, maxlen=options.maxlen) #Quan them vao--------------------
#print(X_train[0]) #quan them vao
#build model...
nb_classes = np.max(y_train) + 1