def make_structure(self):
"""
Creates and allocates the network structure
"""
#Model structure
self.model = dy.ParameterCollection()
#Lex input
self.E = self.model.add_lookup_parameters(
(self.lexicon.size(), self.embedding_size))
#Lex output
self.O = dy.ClassFactoredSoftmaxBuilder(self.hidden_size,
self.brown_file,
self.lexicon.words2i,
self.model,
bias=True)
#RNN
self.rnn = dy.LSTMBuilder(
2, self.embedding_size + self.char_embedding_size,
self.hidden_size, self.model)
#char encodings
self.char_rnn = CharRNNBuilder(self.char_embedding_size,
self.char_memory_size, self.charset,
self.model)
def setUp(self):
# create model
self.pc = dy.ParameterCollection()
dic = dict()
with open('cluster_file.txt', 'w+') as f:
for i in range(5):
f.write(str(i) + " " + str(2 * i) + "\n")
f.write(str(i) + " " + str(2 * i + 1) + "\n")
dic[str(2 * i)] = len(dic)
dic[str(2 * i + 1)] = len(dic)
self.sm = dy.ClassFactoredSoftmaxBuilder(3, 'cluster_file.txt', dic, self.pc, True)
def allocate_params(self):
"""
Allocates memory for the model parameters.
"""
self.model = dy.ParameterCollection()
self.word_embeddings = self.model.add_lookup_parameters(
(self.lexicon.size(), self.word_embedding_size))
self.rnn = dy.LSTMBuilder(
2, self.word_embedding_size + self.char_embedding_size,
self.hidden_size, self.model)
self.char_rnn = CharRNNBuilder(self.char_embedding_size,
self.char_embedding_size, self.charset,
self.model)
self.word_softmax = dy.ClassFactoredSoftmaxBuilder(
self.hidden_size,
self.brown_file,
self.lexicon.words2i,
self.model,
bias=True)