output_layer = Layer.W(2*hiddendim, labelspace, 'Output')
output_bias = Layer.b(labelspace, 'OutputBias')
outputs, fstate = tf.nn.dynamic_rnn(lstm, embeddings.lookup(inputs),
sequence_length=lengths,
dtype=tf.float32)
logits = tf.matmul(fstate, output_layer) + output_bias
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, labels))
## Learning ##
## Optimizer Adam/RMSPropOptimizer tf.train.AdamOptimizer()
optimizer = tf.train.AdamOptimizer()
## Gradient Clipping:
##tvars = tf.trainable_variables()
##grads, _ = tf.clip_by_global_norm(tf.gradients(loss, tvars), 5.0)
##train_op = optimizer.apply_gradients(zip(grads, tvars))
train_op = optimizer.minimize(loss)
correct_prediction = tf.equal(tf.argmax(logits,1), tf.argmax(labels,1))
# Add this to session to see cpu/gpu placement:
# config=tf.ConfigProto(log_device_placement=True)
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
# Model params
Trainer = Training(sess, correct_prediction, train_op, loss,
inputs, labels, lengths, batch_size)
# Run training
Trainer.train(training, training_labels, development, development_labels,
generate_batch, train_lens, dev_lens)
embeddings = Embedding(vocabsize, one_hot=onehot, embedding_size=embeddingdim)
# Input data.
dataset = tf.placeholder(tf.int32, shape=[batch_size, maxlength], name='Train')
labels = tf.placeholder(tf.float32,
shape=[batch_size, labelspace],
name='Label')
# Model
hidden_layer = Layer.W(inputdim, hiddendim, 'Hidden')
hidden_bias = Layer.b(hiddendim, 'HiddenBias')
# Prediction
output_layer = Layer.W(hiddendim, labelspace, 'Output')
output_bias = Layer.b(labelspace, 'OutputBias')
embedded = tf.reshape(embeddings.lookup(dataset), [batch_size, inputdim])
forward = tf.nn.relu(tf.matmul(embedded, hidden_layer) + hidden_bias)
dropout = tf.nn.dropout(forward, 0.5)
logits = tf.matmul(dropout, output_layer) + output_bias
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, labels))
train_op = tf.train.AdamOptimizer().minimize(loss)
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
# Model params
Trainer = Training(sess, correct_prediction, train_op, loss, dataset,
labels)
# Run training
Trainer.train(training, training_labels, development, development_labels,
generate_batch)
return jsonify(output)
else:
return """<html><body>
GET not implemented
</body></html>"""
saver = tf.train.Saver()
# Add this to session to see cpu/gpu placement:
# config=tf.ConfigProto(log_device_placement=True)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
if DoTrain:
print "Train Source"
Trainer = Training(sess, correct_prediction[0], logits[0], train_op[0],
loss[0], inputs, labels[0], lengths, batch_size, 7) #6
Trainer.train(training[0], training_labels[0], testing[0],
testing_labels[0], generate_batch, training_lens[0],
testing_lens[0])
print "Train Reference"
Trainer = Training(sess, correct_prediction[1], logits[1], train_op[1],
loss[1], inputs, labels[1], lengths, batch_size, 5) #2
Trainer.train(training[1], training_labels[1], testing[1],
testing_labels[1], generate_batch, training_lens[1],
testing_lens[1])
print "Train Direction"
Trainer = Training(sess, correct_prediction[2], logits[2], train_op[2],
loss[2], inputs, labels[2], lengths, batch_size, 2) #2
Trainer.train(training[2], training_labels[2], testing[2],
#time.sleep(1) # delays for 5 seconds
return jsonify(output)
else:
return """<html><body>
GET not implemented
</body></html>"""
saver = tf.train.Saver()
# Add this to session to see cpu/gpu placement:
# config=tf.ConfigProto(log_device_placement=True)
sess = tf.Session()
sess.run(tf.initialize_all_variables())
if DoTrain:
print "Train Source"
Trainer = Training(sess, correct_prediction[0], logits[0], train_op[0], loss[0],
inputs, labels[0], lengths, batch_size, 7) #6
Trainer.train(training[0], training_labels[0], testing[0],
testing_labels[0], generate_batch, training_lens[0],
testing_lens[0])
print "Train Reference"
Trainer = Training(sess, correct_prediction[1], logits[1], train_op[1], loss[1],
inputs, labels[1], lengths, batch_size, 5) #2
Trainer.train(training[1], training_labels[1], testing[1],
testing_labels[1], generate_batch, training_lens[1],
testing_lens[1])
print "Train Direction"
Trainer = Training(sess, correct_prediction[2], logits[2], train_op[2], loss[2],
inputs, labels[2], lengths, batch_size, 2) #2
Trainer.train(training[2], training_labels[2], testing[2],
onehot = True inputdim = maxlength*vocabsize if onehot else maxlength*embeddingdim # Define embeddings matrix embeddings = Embedding(vocabsize, one_hot=onehot, embedding_size=embeddingdim) # Input data. dataset = tf.placeholder(tf.int32, shape=[batch_size, maxlength], name='Train') labels = tf.placeholder(tf.float32, shape=[batch_size, labelspace], name='Label') # Model hidden_layer = Layer.W(inputdim, hiddendim, 'Hidden') hidden_bias = Layer.b(hiddendim, 'HiddenBias') # Prediction output_layer = Layer.W(hiddendim, labelspace, 'Output') output_bias = Layer.b(labelspace, 'OutputBias') embedded = tf.reshape(embeddings.lookup(dataset), [batch_size,inputdim]) forward = tf.nn.relu(tf.matmul(embedded, hidden_layer) + hidden_bias) dropout = tf.nn.dropout(forward, 0.5) logits = tf.matmul(dropout, output_layer) + output_bias loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, labels)) train_op = tf.train.AdamOptimizer().minimize(loss) correct_prediction = tf.equal(tf.argmax(logits,1), tf.argmax(labels,1)) with tf.Session() as sess: sess.run(tf.initialize_all_variables()) # Model params Trainer = Training(sess, correct_prediction, train_op, loss, dataset, labels) # Run training Trainer.train(training, training_labels, development, development_labels, generate_batch)