def test_worker():
fout = open('test_label.txt', 'w')
P_input = T.tensor4(dtype=theano.config.floatX)
Cparams = loaddata('./para_learning_rate0.01/0rCparas.pkl')
Mparams = loaddata('./para_learning_rate0.01/12Mparas.pkl')
test_handle = test(P_input=P_input, Cparams=Cparams, Mparams=Mparams)
f = theano.function(inputs=[P_input],
outputs=test_handle.output,
updates=[])
right = 0.0
cnt = 0.0
prep = prepare.prepareData()
block_size = prep.block_size
item = prep.generate_test_from_sentence(1)
for pair in item:
for i in range(block_size):
a = f(pair[i][0])
print a
#fout.write(str(a.mean()))
#fout.write('\n')
cnt = cnt + 1
if pair[i][1].mean() == a.mean():
right = right + 1.0
def train_worker():
P_input = T.tensor4(dtype=theano.config.floatX)
P_y = T.vector(name ='P_y', dtype='int32')
Cparams = parameters.c_random_weights()#loaddata('./Cparams.pkl')
Mparams = parameters.random_weights()
train_handle = train(P_input=P_input, P_y=P_y, Cparams=Cparams, Mparams=Mparams, learning_rate=learning_rate)
f = theano.function(inputs=[P_input, P_y],outputs=train_handle.cost, updates=train_handle.updates)
prep = prepare.prepareData()
block_size = prep.block_size
for epoch in range(startepoch,100):
print "epoch", epoch
cnt = 0.0
cost = 0.0
item = prep.generate_batch_from_sentence(batch_size)
for pair in item :
for i in range(block_size):
a = f(pair[i][0],pair[i][1])
cost = cost + a
cnt = cnt + 1
sys.stdout.flush()
sys.stdout.write(str(cnt)+'\r')
#if cnt*batch_size%10000 == 0:
#train_handle.storedata(cnt)
train_handle.storedata(epoch)
train_handle.c_storedata(epoch)
print cost
def train_worker():
P_input = T.tensor4(dtype=theano.config.floatX)
P_y = T.vector(name='P_y', dtype='int32')
Cparams = parameters.c_random_weights() #loaddata('./Cparams.pkl')
Mparams = parameters.random_weights()
train_handle = train(P_input=P_input,
P_y=P_y,
Cparams=Cparams,
Mparams=Mparams,
learning_rate=learning_rate)
f = theano.function(inputs=[P_input, P_y],
outputs=train_handle.cost,
updates=train_handle.updates)
prep = prepare.prepareData()
block_size = prep.block_size
for epoch in range(startepoch, 100):
print "epoch", epoch
cnt = 0.0
cost = 0.0
item = prep.generate_batch_from_sentence(batch_size)
for pair in item:
for i in range(block_size):
a = f(pair[i][0], pair[i][1])
cost = cost + a
cnt = cnt + 1
sys.stdout.flush()
sys.stdout.write(str(cnt) + '\r')
#if cnt*batch_size%10000 == 0:
#train_handle.storedata(cnt)
train_handle.storedata(epoch)
train_handle.c_storedata(epoch)
print cost
def train_worker(ipath,wpath): params = parameters.random_weights() P_input1 = T.matrix(name='P_input1',dtype=theano.config.floatX) P_input2 = T.matrix(name='P_input2',dtype=theano.config.floatX) P_y = T.vector(name='P_y',dtype='int32') RCNN = train_model(learning_rate,params) p_y,cost = RCNN.training(P_input1,P_input2,P_y) rcnn = theano.function(inputs=[P_input1,P_input2,P_y],outputs = [p_y,cost],updates=RCNN.updates) print 'reading dataset' prep = prepare.prepareData(ipath) prep.Reading_traindata() block_size = prep.block_size session_size = prep.train_session_num for epoch in range(iternum): print "epoch", epoch for k in range(session_size): item = prep.generate_batch_from_sentence(k) out,cost = rcnn(item[0],item[1],item[2]) sys.stdout.flush() sys.stdout.write(str(k)+'\r') RCNN.storedata(epoch,wpath)
def test_worker():
fout = open('test_label.txt','w')
P_input = T.tensor4(dtype=theano.config.floatX)
Cparams = loaddata('./para_learning_rate0.01/0rCparas.pkl')
Mparams = loaddata('./para_learning_rate0.01/12Mparas.pkl')
test_handle = test(P_input=P_input, Cparams=Cparams, Mparams=Mparams)
f = theano.function(inputs=[P_input],outputs=test_handle.output, updates=[])
right = 0.0
cnt = 0.0
prep = prepare.prepareData()
block_size = prep.block_size
item = prep.generate_test_from_sentence(1)
for pair in item :
for i in range(block_size):
a = f(pair[i][0])
print a
#fout.write(str(a.mean()))
#fout.write('\n')
cnt = cnt + 1
if pair[i][1].mean()==a.mean():
right = right + 1.0
x = F.relu(self.fc1(x))
output = torch.sigmoid(self.fc2(x))
return output
print('train dataset: ', dataset_train)
xtrain, train_y = load_dataset_text(dataset_train, False)
dataset_name = 'sp-a'
dataset_name = basePath + dataset_name
print('test dataset', dataset_name)
xvalid, valid_y = load_dataset_text(dataset_name, False)
prepare = prepareData(EMBEDDING_DIM, MAXLEN_CLAIM, TRIM_NUM)
[train_sets,
dev_sets] = prepare.data_process_pair(xtrain, train_y, xvalid, valid_y,
separate, sampleNum)
pretrained_embeddings = torch.tensor(prepare.get_embedding_matrix()).to(device)
dataset = myDataset_pair(train_sets[0], train_sets[1])
trainloader = torch.utils.data.DataLoader(dataset,
batch_size=BATCH_SIZE,
shuffle=False)
dataset = myDataset(dev_sets[0], dev_sets[1])
devloader = torch.utils.data.DataLoader(dataset,
batch_size=BATCH_SIZE,
shuffle=False)
out_layer = tf.matmul(layer_2, weights['out']) + biases['out']
return out_layer
# Construct model
logits = multilayer_perceptron(X)
pred = tf.nn.sigmoid(logits) # Apply softmax to logits
# Define loss and optimizer
loss_op = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
logits=logits, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
train_op = optimizer.minimize(loss_op)
# Initializing the variables
init = tf.global_variables_initializer()
features, label = prepareData()
features = features[:, 0, 0, :]
label = label[:, np.newaxis]
with tf.Session() as sess:
sess.run(init)
# Training cycle
for epoch in range(training_epochs):
avg_cost = 0.
total_batch = 30
# Loop over all batches
for i in range(total_batch):
idx = np.random.choice(features.shape[0], batch_size)
batch_x, batch_y = features[idx], label[idx]
# Run optimization op (backprop) and cost op (to get loss value)
_, c = sess.run([train_op, loss_op], feed_dict={X: batch_x,