def main():
print "Starting evaluation..."
sents = load_sentences("c:\\corpora\\corrected.txt")
for i in xrange(len(sents)):
mid_index = get_check_index(sents[i])
print get_original_word(sents[i], mid_index), "->", get_corrected_word(sents[i], mid_index)
train = list(read("C:\\corpora\\long30k.txt"))
words = []
wc = Counter()
for sent in train:
for w in sent:
words.append(w)
wc[w] += 1
vw = Vocab.from_corpus([words])
nwords = vw.size()
LAYERS = 2
INPUT_DIM = 200 # 50 #256
HIDDEN_DIM = 300 # 50 #1024
print "words", nwords
# DyNet Starts
dy.init()
model = dy.Model()
#W_sm = model.add_parameters((nwords, HIDDEN_DIM))
#b_sm = model.add_parameters(nwords)
#trainer = dy.SimpleSGDTrainer(model)
#WORDS_LOOKUP = model.add_lookup_parameters((nwords, INPUT_DIM))
#RNN = dy.LSTMBuilder(LAYERS, INPUT_DIM, HIDDEN_DIM, model)
(RNN, WORDS_LOOKUP, W_sm, b_sm) = model.load("C:\\corpora\\batch_bigmodel.txt")
for sentence in sents:
evaluate_sentence(sentence, vw, [RNN, WORDS_LOOKUP, W_sm, b_sm])
def __init__(self, c2i, num_lstm_layers=-1,\
char_dim=-1, hidden_dim=-1, word_embedding_dim=-1, file=None):
self.c2i = c2i
self._model = dy.Model()
#self._model = dy.ParameterCollection()
if file == None:
# Char LSTM Parameters
self.char_lookup = self._model.add_lookup_parameters((len(c2i), char_dim))
self.char_fwd_lstm = dy.LSTMBuilder(num_lstm_layers, char_dim, hidden_dim, self._model)
self.char_bwd_lstm = dy.LSTMBuilder(num_lstm_layers, char_dim, hidden_dim, self._model)
# Post-LSTM Parameters
self.lstm_to_rep_params = self._model.add_parameters((word_embedding_dim, hidden_dim * 2))
self.lstm_to_rep_bias = self._model.add_parameters(word_embedding_dim)
self.mlp_out = self._model.add_parameters((word_embedding_dim, word_embedding_dim))
self.mlp_out_bias = self._model.add_parameters(word_embedding_dim)
else:
model_members = iter(self._model.load(file))
#pc2 = dy.ParameterCollection()
#model_members = iter(dy.load(file, pc2))
self.char_lookup = model_members.next()
self.char_fwd_lstm = model_members.next()
self.char_bwd_lstm = model_members.next()
self.lstm_to_rep_params = model_members.next()
self.lstm_to_rep_bias = model_members.next()
self.mlp_out = model_members.next()
self.mlp_out_bias = model_members.next()
def main():
model = dy.Model()
trainer = dy.SimpleSGDTrainer(model)
if sys.argv[1] == '-train':
training_src = read_file(sys.argv[2])
training_tgt = read_file(sys.argv[3])
dev_src = read_file(sys.argv[4])
dev_tgt = read_file(sys.argv[5])
test_src = read_file(sys.argv[6])
blind_src = read_file(sys.argv[7])
attention = Attention(model, training_src, training_tgt, dev_src,
dev_tgt, test_src, blind_src)
attention.train()
elif sys.argv[1] == '-trans':
test_src = read_file(sys.argv[2])
attention = Attention(model, None, None, None, None, test_src, None,
'test')
for w in sent:
words.append(w)
wc[w] += 1
vw = Vocab.from_corpus([words])
STOP = vw.w2i["<stop>"]
START = vw.w2i["<start>"]
nwords = vw.size()
LAYERS = 1
INPUT_DIM = 200 #50 #256
HIDDEN_DIM = 200 # 50 #1024
print "words", nwords
dy.init()
model = dy.Model()
trainer = dy.AdamTrainer(model)
WORDS_LOOKUP = model.add_lookup_parameters((nwords, INPUT_DIM))
W_sm = model.add_parameters((nwords, HIDDEN_DIM * 2))
b_sm = model.add_parameters(nwords)
builders = [
LSTMBuilder(LAYERS, INPUT_DIM, HIDDEN_DIM, model),
LSTMBuilder(LAYERS, INPUT_DIM, HIDDEN_DIM, model),
]
def predict_middle_word(iprefix, ipostfix, builders):
renew_cg()
import _gdynet as G
print()
import _dynet as C
cm = C.Model()
gm = G.Model()
cpW = cm.add_parameters((1000,1000))
gpW = gm.add_parameters((1000,1000))
def do_cpu():
C.renew_cg()
W = C.parameter(cpW)
W = W*W*W*W*W*W*W
z = C.squared_distance(W,W)
z.value()
z.backward()
def do_gpu():
G.renew_cg()
W = G.parameter(gpW)
W = W*W*W*W*W*W*W
z = G.squared_distance(W,W)
z.value()
z.backward()
import time
s = time.time()
do_cpu()
print("CPU time:",time.time() - s)