def batch_sgd_step_disk(self, lines, lrate): # Get nn input and output from line batch token_list, relation_list = utils.parse_processed_lines(lines) inp = [] for tokens in token_list: inp.append(np.asarray([self.word_to_ind.get(token, self.default_index) for token in tokens])) inp = np.asarray(inp) rel = [self.rel_to_ind[relation] for relation in relation_list] out = np.zeros((len(lines), self.odim)) out[range(len(lines)), rel] = 1 out = np.asarray(out) print 'Example input', type(inp), inp.size, type(inp[0]), inp[0].size #print 'Example output', rel hist = self.model.fit(inp, out, batch_size = len(lines), nb_epoch = 1, verbose = 0) return hist['loss'][0]*len(lines)
def batch_sgd(self, lines): # Get nn input and output from line batch token_list, relation_list = utils.parse_processed_lines(lines) inp = [] for tokens in token_list: inp.append(np.asarray([self.word_to_ind.get(token, self.default_index) for token in tokens])) inp = np.asarray(inp) inp = sequence.pad_sequences(inp) print 'Example input', type(inp), inp.size, type(inp[0]), inp[0].size rel = [self.rel_to_ind[relation] for relation in relation_list] #print 'Example output', rel out = np.zeros((len(lines), self.odim)) out[range(len(lines)), rel] = 1 out = np.asarray(out) hist = self.model.fit(inp, out, batch_size = self.batch_size, nb_epoch = self.nepochs, verbose = 1)
def batch_sgd_step_disk(self, lines, lrate):
# Get nn input and output from line batch
token_list, relation_list = utils.parse_processed_lines(lines)
inp = []
for tokens in token_list:
inp.append(
np.asarray([
self.word_to_ind.get(token, self.default_index)
for token in tokens
]))
inp = np.asarray(inp)
rel = [self.rel_to_ind[relation] for relation in relation_list]
out = np.zeros((len(lines), self.odim))
out[range(len(lines)), rel] = 1
out = np.asarray(out)
print 'Example input', type(inp), inp.size, type(inp[0]), inp[0].size
#print 'Example output', rel
hist = self.model.fit(inp,
out,
batch_size=len(lines),
nb_epoch=1,
verbose=0)
return hist['loss'][0] * len(lines)
def batch_sgd(self, lines):
# Get nn input and output from line batch
token_list, relation_list = utils.parse_processed_lines(lines)
inp = []
for tokens in token_list:
inp.append(
np.asarray([
self.word_to_ind.get(token, self.default_index)
for token in tokens
]))
inp = np.asarray(inp)
inp = sequence.pad_sequences(inp)
print 'Example input', type(inp), inp.size, type(inp[0]), inp[0].size
rel = [self.rel_to_ind[relation] for relation in relation_list]
#print 'Example output', rel
out = np.zeros((len(lines), self.odim))
out[range(len(lines)), rel] = 1
out = np.asarray(out)
hist = self.model.fit(inp,
out,
batch_size=self.batch_size,
nb_epoch=self.nepochs,
verbose=1)