def setUp(self):
tree = ((0, 1), ((2, 3), 4))
self.func = functions.BinaryHierarchicalSoftmax(3, tree)
self.x = numpy.random.uniform(-1, 1, (2, 3)).astype(numpy.float32)
self.t = numpy.array([0, 2]).astype(numpy.int32)
self.gy = numpy.random.uniform(-1, 1, ()).astype(numpy.float32)
self.W = self.func.W.copy()
print('n_vocab: %d' % n_vocab)
print('data length: %d' % len(dataset))
if args.model == 'skipgram':
train_model = skip_gram
elif args.model == 'cbow':
train_model = continuous_bow
else:
raise Exception('Unknown model type: {}'.format(args.model))
model = chainer.FunctionSet(embed=F.EmbedID(n_vocab, args.unit), )
if args.out_type == 'hsm':
tree = F.create_huffman_tree(counts)
model.l = F.BinaryHierarchicalSoftmax(args.unit, tree)
loss_func = model.l
elif args.out_type == 'ns':
cs = [counts[w] for w in range(len(counts))]
model.l = F.NegativeSampling(args.unit, cs, 20)
loss_func = model.l
elif args.out_type == 'original':
model.l = F.Linear(args.unit, n_vocab)
loss_func = lambda h, t: F.softmax_cross_entropy(model.l(h), t)
else:
raise Exception('Unknown output type: {}'.format(args.out_type))
if args.gpu >= 0:
model.to_gpu()
dataset = np.array(dataset, dtype=np.int32)