def sentiment_analysis(load_model, label_type, embs_convert_type,
label_type_folder, target_data_folder, save_folder):
check_type(label_type,
types_list=['tonality', 'toxicity'],
type_name='label')
check_type(embs_convert_type,
types_list=['mean', 'length_64'],
type_name='embeddings convert')
x_target, y_target = load_target_data(label_type=label_type,
convert_type=embs_convert_type,
data_folder=target_data_folder)
if not load_model:
x_source, x_source_test, y_source, y_source_test = load_source_data(
label_type=label_type,
label_data_folder=label_type_folder,
convert_type=embs_convert_type)
model = train_lstm(x_source=x_source,
y_source=y_source,
label_type=label_type,
convert_type=embs_convert_type,
save_folder=save_folder,
epochs=5)
predict(model=model, x=x_source_test, y=y_source_test, title='Source')
else:
model = load_lstm(label_type=label_type,
convert_type=embs_convert_type,
folder=save_folder)
y_pred = predict(model=model, x=x_target, y=y_target, title='Target')
return y_pred
def main():
model_filename = 'models/lstm_model_3.json'
weights_filename = 'models/lstm_model_3.h5'
kaggle_filename = 'output/kaggle_lstm_256.csv'
lstm.train_lstm(model_filename,
weights_filename,
l1_d=256,
l2_d=256,
b_s=128)
run_lstm_on_test_data(model_filename, weights_filename, kaggle_filename)
model_filename = 'models/bi_lstm_model_128.json'
weights_filename = 'models/bi_lstm_model_128.h5'
kaggle_filename = 'output/bi_kaggle_lstm_128.csv'
lstm.train_lstm(model_filename,
weights_filename,
l1_d=128,
l2_d=128,
b_s=128,
bi=True)
run_lstm_on_test_data(model_filename, weights_filename, kaggle_filename)
model_filename = 'models/bi_lstm_model_265.json'
weights_filename = 'models/bi_lstm_model_256.h5'
kaggle_filename = 'output/bi_kaggle_lstm_256.csv'
lstm.train_lstm(model_filename,
weights_filename,
l1_d=256,
l2_d=256,
b_s=128,
bi=True)
run_lstm_on_test_data(model_filename, weights_filename, kaggle_filename)
def test_lstm():
lstm.train_lstm(max_epochs=1, test_size=1000, saveto='')
vocab = len(dic)
print("Vocab=%d" %vocab)
X_train_batch = drop_tail(X_train_batch, seq_len)
X_val_batch = drop_tail(X_val_batch, seq_len)
#print 'ddddd'
print X_train_batch.shape
print X_val_batch.shape
model = lstm.setup_rnn_model(mx.cpu(),
num_lstm_layer=num_lstm_layer,
seq_len=seq_len,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=vocab,
batch_size=batch_size,
input_size=vocab,
initializer=mx.initializer.Uniform(0.1),dropout=0.5)
# max_grad_norm=5.0 | update_period=1 | wd=0 | learning_rate=0.1 | num_roud=25
lstm.train_lstm(model, X_train_batch, X_val_batch,
num_round=num_round,
half_life=2,
max_grad_norm = max_grad_norm,
update_period=update_period,
learning_rate=learning_rate,
wd=wd)
# momentum=momentum)
# whether do group-wise concat
concat_decode = False
use_loss = True
model = lstm.setup_rnn_model(mx.gpu(),
group2ctx=group2ctx,
concat_decode=concat_decode,
use_loss=use_loss,
num_lstm_layer=num_lstm_layer,
seq_len=X_train_batch.default_bucket_key,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=vocab,
batch_size=batch_size,
input_size=vocab,
initializer=mx.initializer.Uniform(0.1),
dropout=0.5,
buckets=buckets)
lstm.train_lstm(model,
X_train_batch,
X_val_batch,
num_round=num_round,
concat_decode=concat_decode,
use_loss=use_loss,
half_life=2,
max_grad_norm=max_grad_norm,
update_period=update_period,
learning_rate=learning_rate,
batch_size=batch_size,
wd=wd)
X_val, _ = load_data("./data/ptb.valid.txt", dic)
X_train_batch = replicate_data(X_train, batch_size)
X_val_batch = replicate_data(X_val, batch_size)
vocab = len(dic)
print("Vocab=%d" % vocab)
X_train_batch = drop_tail(X_train_batch, seq_len)
X_val_batch = drop_tail(X_val_batch, seq_len)
model = lstm.setup_rnn_model(mx.gpu(),
num_lstm_layer=num_lstm_layer,
seq_len=seq_len,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=vocab,
batch_size=batch_size,
input_size=vocab,
initializer=mx.initializer.Uniform(0.1),
dropout=0.5)
lstm.train_lstm(model,
X_train_batch,
X_val_batch,
num_round=num_round,
half_life=2,
max_grad_norm=max_grad_norm,
update_period=update_period,
learning_rate=learning_rate,
wd=wd,
momentum=momentum)
def test_lstm():
lstm.train_lstm(max_epochs=1, test_size=1000, saveto='')
LIN_valid.save_LSTM_input(mat_indexes_valid, valid_indexes_file)
X_val = LIN_valid.load_LSTM_input(valid_indexes_file)
model = lstm.setup_rnn_model(mx.gpu(),
num_lstm_layer=num_lstm_layer,
seq_len=sentence_len,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=max_vocab,
batch_size=batch_size,
input_size=max_vocab,
initializer=mx.initializer.Uniform(0.1),
dropout=0.5)
for i in xrange(nb_split):
input_small_file = small_file_prefix + '.' + str(i)
mat_indexes_file = mat_indexes_file_prefix + '.' + str(i)
LIN = LSTMinput(input_small_file, sentence_len)
X_train = LIN.load_LSTM_input(mat_indexes_file)
lstm.train_lstm(model,
X_train,
X_val,
num_round=num_round,
half_life=2,
max_grad_norm=max_grad_norm,
update_period=update_period,
learning_rate=learning_rate,
wd=wd)
del X_train
del LIN
'decode': mx.gpu(ngpu - 1)}
for i in range(num_lstm_layer):
group2ctx['layer%d' % i] = mx.gpu(i * ngpu // num_lstm_layer)
# whether do group-wise concat
concat_decode = False
use_loss=True
model = lstm.setup_rnn_model(mx.gpu(), group2ctx=group2ctx,
concat_decode=concat_decode,
use_loss=use_loss,
num_lstm_layer=num_lstm_layer,
seq_len=X_train_batch.default_bucket_key,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=vocab,
batch_size=batch_size,
input_size=vocab,
initializer=mx.initializer.Uniform(0.1),dropout=0.5, buckets=buckets)
lstm.train_lstm(model, X_train_batch, X_val_batch,
num_round=num_round,
concat_decode=concat_decode,
use_loss=use_loss,
half_life=2,
max_grad_norm = max_grad_norm,
update_period=update_period,
learning_rate=learning_rate,
batch_size = batch_size,
wd=wd)
# import data
f = open('dataset/dataset_1.csv')
# read stock info
df = pd.read_csv(f)
# get col 3-10
data = df.iloc[:, 2:10].values
# set parameter
input_size = 7 # input size
output_size = 1 # output size
lr = 0.001 # learning rate
time_step = 20 # time step for lstm
choice = 1
if choice == 0:
res1, acc1, mae1 = lstm.train_lstm(data, input_size, output_size, lr, time_step, choice=0)
res2, acc2, mae2 = bp.train_bp(data, input_size, output_size, lr, choice=0)
# plot acc and mae
plt.figure()
plt.title('acc')
plt.xlabel('Number of iterations')
plt.ylabel('value')
plt.plot(list(range(len(acc1))), acc1, color='green')
plt.plot(list(range(len(acc2))), acc2, color='red')
plt.show()
plt.figure()
plt.title('mae')
plt.xlabel('Number of iterations')
plt.ylabel('value')
plt.plot(list(range(len(mae1))), mae1, color='green')
X_train, dic = load_data("./data/ptb.train.txt")
X_val, _ = load_data("./data/ptb.valid.txt", dic)
X_train_batch = replicate_data(X_train, batch_size)
X_val_batch = replicate_data(X_val, batch_size)
vocab = len(dic)
print("Vocab=%d" %vocab)
X_train_batch = drop_tail(X_train_batch, seq_len)
X_val_batch = drop_tail(X_val_batch, seq_len)
model = lstm.setup_rnn_model(mx.gpu(),
num_lstm_layer=num_lstm_layer,
seq_len=seq_len,
num_hidden=num_hidden,
num_embed=num_embed,
num_label=vocab,
batch_size=batch_size,
input_size=vocab,
initializer=mx.initializer.Uniform(0.1))
lstm.train_lstm(model, X_train_batch, X_val_batch,
num_round=num_round,
half_life=2,
update_period=update_period,
learning_rate=learning_rate,
wd=wd,
momentum=momentum,
clip_gradient=clip_gradient)