def load_data(self, data_dir):
self.data_loader = TextLoader(data_dir, self.batch_size, self.seq_length)
# self.data_loader = Reader(filename, batch_size,seq_length)
# vocabulary_size.append(data_loader.vocab_size)
self.vocab_size = self.data_loader.vocab_size
def load_data(self, data_dir):
self.data_loader = TextLoader(data_dir, self.batch_size, self.seq_length)
#self.data_loader = Reader(filename, batch_size,seq_length)
#vocabulary_size.append(data_loader.vocab_size)
self.vocab_size = self.data_loader.vocab_size
class Model:
def __init__(
self,
infer=False,
rnn_size=512,
batch_size=50,
seq_length=50,
num_layers=3,
vocab_size=300,
grad_clip=5,
learning_rate=0.02,
decay_rate=0.95,
num_epocs=50,
display_steps=100,
):
self.batch_size = batch_size
self.seq_length = seq_length
self.rnn_size = rnn_size
self.num_layers = num_layers
# vocabulary_size =[500, 2000, 8000, 16000, 32000, 64000] # hyper-parameter
# self.vocab_size = vocab_size
self.grad_clip = grad_clip
self.learning_rate = learning_rate # hyper-parameter
self.decay_rate = decay_rate
self.num_epocs = num_epocs
self.display_steps = display_steps
self.is_training = True
self.infer = infer
if infer:
self.batch_size = 1
self.seq_length = 1
self.load_data(data_dir)
def load_data(self, data_dir):
self.data_loader = TextLoader(data_dir, self.batch_size, self.seq_length)
# self.data_loader = Reader(filename, batch_size,seq_length)
# vocabulary_size.append(data_loader.vocab_size)
self.vocab_size = self.data_loader.vocab_size
def create_model(self):
self.input_data = tf.placeholder(tf.int32, [self.batch_size, self.seq_length], name="input_data")
self.target_data = tf.placeholder(tf.int32, [self.batch_size, self.seq_length], name="target_data")
# define hyper_parameters
self.keep_prob = tf.Variable(0.3, trainable=False, name="keep_prob")
self.lr = tf.Variable(0.0, trainable=False, name="lr")
softmax_weights = tf.get_variable("softmax_weights", [self.rnn_size, self.vocab_size])
softmax_biases = tf.get_variable("softmax_biases", [self.vocab_size])
lstm_cell = rnn_cell.BasicLSTMCell(self.rnn_size)
# if self.is_training and self.keep_prob < 1:
# lstm_cell = rnn_cell.DropoutWrapper(lstm_cell, output_keep_prob=self.keep_prob)
multilayer_cell = rnn_cell.MultiRNNCell([lstm_cell] * self.num_layers)
self.initial_state = multilayer_cell.zero_state(self.batch_size, tf.float32)
with tf.device("/cpu:0"):
# define the embedding matrix for the whole vocabulary
self.embedding = tf.get_variable("embeddings", [self.vocab_size, self.rnn_size])
# take the vector representation for each word in the embeddings
embeds = tf.nn.embedding_lookup(self.embedding, self.input_data)
if self.is_training and self.keep_prob < 1:
embeds = tf.nn.dropout(embeds, self.keep_prob)
def loop(prev, _):
prev = tf.nn.xw_plus_b(prev, softmax_weights, softmax_biases)
prev_symbol = tf.stop_gradient(tf.argmax(prev, 1))
return tf.nn.embedding_lookup(self.embedding, prev_symbol)
# convert input to a list of seq_length
inputs = tf.split(1, self.seq_length, embeds)
# after splitting the shape becomes (batch_size,1,rnn_size). We need to modify it to [batch*rnn_size]
inputs = [tf.squeeze(input_, [1]) for input_ in inputs]
output, states = seq2seq.rnn_decoder(
inputs, self.initial_state, multilayer_cell, loop_function=loop if self.infer else None, scope="rnnlm"
)
output = tf.reshape(tf.concat(1, output), [-1, self.rnn_size])
self.logits = tf.nn.xw_plus_b(output, softmax_weights, softmax_biases)
self.probs = tf.nn.softmax(self.logits, name="probability")
loss = seq2seq.sequence_loss_by_example(
[self.logits],
[tf.reshape(self.target_data, [-1])],
[tf.ones([self.batch_size * self.seq_length])],
self.vocab_size,
)
self.cost = tf.reduce_sum(loss) / (self.batch_size * self.seq_length)
self.final_state = states[-1]
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(self.cost, tvars), self.grad_clip)
optimizer = tf.train.AdamOptimizer(0.01)
self.train_op = optimizer.apply_gradients(zip(grads, tvars))
def train_model(self):
# load data
self.create_model()
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
saver = tf.train.Saver(tf.all_variables())
training_cost = float("inf")
trainingcost_lastepoch = float("inf")
with open(os.path.join("chars_vocab.pkl"), "w") as f:
cPickle.dump((self.data_loader.chars, self.data_loader.vocab), f)
# for prob in keep_probaties:
# sess.run(tf.assign(self.keep_prob, prob))
for e in xrange(self.num_epocs):
sess.run(tf.assign(self.lr, self.learning_rate * (self.decay_rate ** e)))
self.data_loader.reset_batch_pointer()
state = self.initial_state.eval(session=sess)
for b in xrange(self.data_loader.num_batches):
x, y = self.data_loader.next_batch()
# print "X:", x, "Y:", y
feed = {self.input_data: x, self.target_data: y, self.initial_state: state}
effective_cost, state, _ = sess.run([self.cost, self.final_state, self.train_op], feed_dict=feed)
if effective_cost < training_cost:
training_cost = effective_cost
if b % self.display_steps == 0:
print " Batch Number: ", b, "/", self.data_loader.num_batches
print " Epoch :", e,
print " and Cost : ", effective_cost
if training_cost < trainingcost_lastepoch:
trainingcost_lastepoch = training_cost
saver.save(sess, "model.ckpt")
def sample(self, num_of_samples, starting_text="The"):
self.create_model()
with open(os.path.join("chars_vocab.pkl")) as f:
chars, vocab = cPickle.load(f)
def weighted_pick(weights):
# print "weights :", len(weights)
t = np.cumsum(weights)
s = np.sum(weights)
return int(np.searchsorted(t, np.random.rand(1) * s))
# store arguments in in a file and load it
with tf.Session() as sess:
tf.initialize_all_variables().run()
saver = tf.train.Saver(tf.all_variables())
ckpt = tf.train.get_checkpoint_state("model.ckpt")
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
# print "Length of chars : ", len(chars)
# print "vocab", vocab
ret = starting_text
char = starting_text[-1]
state = self.initial_state.eval()
for i in range(num_of_samples):
x = np.zeros((1, 1))
x[0, 0] = vocab[char]
feed = {self.input_data: x, self.initial_state: state}
[prob, state] = sess.run([self.probs, self.final_state], feed_dict=feed)
p = prob[0]
# sample = int(np.random.choice(len(p), p=p))
sample = weighted_pick(p)
# print "Sample is : ",sample
pred = chars[sample]
ret += pred
char = pred
return ret
class Model():
def __init__(self,infer= False, rnn_size = 512, batch_size=50, seq_length = 50, num_layers=3,
vocab_size = 300, grad_clip = 5, learning_rate = 0.02,
decay_rate = 0.95, num_epocs = 50, display_steps = 100):
self.batch_size= batch_size
self.seq_length = seq_length
self.rnn_size = rnn_size
self.num_layers = num_layers
#vocabulary_size =[500, 2000, 8000, 16000, 32000, 64000] # hyper-parameter
#self.vocab_size = vocab_size
self.grad_clip = grad_clip
self.learning_rate = learning_rate # hyper-parameter
self.decay_rate = decay_rate
self.num_epocs = num_epocs
self.display_steps = display_steps
self.is_training = True
self.infer = infer
if infer:
self.batch_size = 1
self.seq_length = 1
self.load_data(data_dir)
def load_data(self, data_dir):
self.data_loader = TextLoader(data_dir, self.batch_size, self.seq_length)
#self.data_loader = Reader(filename, batch_size,seq_length)
#vocabulary_size.append(data_loader.vocab_size)
self.vocab_size = self.data_loader.vocab_size
def create_model(self):
self.input_data = tf.placeholder(tf.int32, [self.batch_size, self.seq_length], name="input_data")
self.target_data = tf.placeholder(tf.int32,[self.batch_size, self.seq_length], name="target_data")
# define hyper_parameters
self.keep_prob = tf.Variable(0.3, trainable=False, name='keep_prob')
self.lr = tf.Variable(0.0, trainable=False, name="lr")
softmax_weights = tf.get_variable("softmax_weights",[self.rnn_size, self.vocab_size])
softmax_biases = tf.get_variable("softmax_biases", [self.vocab_size])
lstm_cell = rnn_cell.BasicLSTMCell(self.rnn_size)
# if self.is_training and self.keep_prob < 1:
# lstm_cell = rnn_cell.DropoutWrapper(lstm_cell, output_keep_prob=self.keep_prob)
multilayer_cell = rnn_cell.MultiRNNCell([lstm_cell] * self.num_layers)
self.initial_state = multilayer_cell.zero_state(self.batch_size, tf.float32)
with tf.device("/cpu:0"):
# define the embedding matrix for the whole vocabulary
self.embedding = tf.get_variable("embeddings", [self.vocab_size, self.rnn_size])
# take the vector representation for each word in the embeddings
embeds = tf.nn.embedding_lookup(self.embedding, self.input_data)
if self.is_training and self.keep_prob < 1:
embeds = tf.nn.dropout(embeds, self.keep_prob)
def loop(prev, _):
prev = tf.nn.xw_plus_b(prev, softmax_weights, softmax_biases)
prev_symbol = tf.stop_gradient(tf.argmax(prev, 1))
return tf.nn.embedding_lookup(self.embedding, prev_symbol)
#convert input to a list of seq_length
inputs = tf.split(1,self.seq_length, embeds)
#after splitting the shape becomes (batch_size,1,rnn_size). We need to modify it to [batch*rnn_size]
inputs = [ tf.squeeze(input_, [1]) for input_ in inputs]
output,states= seq2seq.rnn_decoder(inputs,self.initial_state, multilayer_cell, loop_function=loop if self.infer else None, scope='rnnlm')
output = tf.reshape(tf.concat(1, output), [-1, self.rnn_size])
self.logits = tf.nn.xw_plus_b(output, softmax_weights, softmax_biases)
self.probs = tf.nn.softmax(self.logits, name= "probability")
loss = seq2seq.sequence_loss_by_example([self.logits], [tf.reshape(self.target_data, [-1])], [tf.ones([self.batch_size * self.seq_length])], self.vocab_size )
self.cost = tf.reduce_sum(loss) / ( self.batch_size * self.seq_length )
self.final_state= states[-1]
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(self.cost, tvars),self.grad_clip)
optimizer = tf.train.AdamOptimizer(0.01)
self.train_op = optimizer.apply_gradients(zip(grads, tvars))
def train_model(self):
# load data
self.create_model()
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
saver = tf.train.Saver(tf.all_variables())
training_cost = float("inf")
trainingcost_lastepoch = float("inf")
with open(os.path.join('chars_vocab.pkl'), 'w') as f:
cPickle.dump((self.data_loader.chars, self.data_loader.vocab), f)
# for prob in keep_probaties:
# sess.run(tf.assign(self.keep_prob, prob))
for e in xrange(self.num_epocs):
sess.run(tf.assign(self.lr, self.learning_rate * (self.decay_rate ** e)))
self.data_loader.reset_batch_pointer()
state = self.initial_state.eval(session = sess)
for b in xrange(self.data_loader.num_batches):
x, y = self.data_loader.next_batch()
#print "X:", x, "Y:", y
feed ={self.input_data : x, self.target_data:y, self.initial_state:state }
effective_cost,state,_= sess.run([self.cost, self.final_state, self.train_op], feed_dict=feed)
if effective_cost < training_cost:
training_cost = effective_cost
if (b % self.display_steps == 0):
print " Batch Number: ", b, "/", self.data_loader.num_batches
print " Epoch :", e ,
print " and Cost : ", effective_cost
if training_cost < trainingcost_lastepoch:
trainingcost_lastepoch = training_cost
saver.save(sess, 'model.ckpt')
def sample(self, num_of_samples, starting_text = 'The'):
self.create_model()
with open(os.path.join('chars_vocab.pkl')) as f:
chars, vocab = cPickle.load(f)
def weighted_pick(weights):
#print "weights :", len(weights)
t = np.cumsum(weights)
s = np.sum(weights)
return(int(np.searchsorted(t, np.random.rand(1)*s)))
#store arguments in in a file and load it
with tf.Session() as sess:
tf.initialize_all_variables().run()
saver = tf.train.Saver(tf.all_variables())
ckpt = tf.train.get_checkpoint_state('model.ckpt')
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
#print "Length of chars : ", len(chars)
#print "vocab", vocab
ret = starting_text
char = starting_text[-1]
state= self.initial_state.eval()
for i in range(num_of_samples):
x = np.zeros((1,1))
x[0,0]= vocab[char]
feed = {self.input_data :x, self.initial_state:state}
[prob, state] = sess.run([self.probs, self.final_state], feed_dict=feed)
p = prob[0]
# sample = int(np.random.choice(len(p), p=p))
sample = weighted_pick(p)
#print "Sample is : ",sample
pred = chars[sample]
ret += pred
char = pred
return ret