class TestGravesPredictionNet(TestCase):
def setUp(self):
self.model = GravesPredictionNet()
self.mini_batch = 1
self.shape = (self.mini_batch, 100)
self.xs, self.es = parse_IAMxml("res/strokesz.xml")
self.xs = numpy.float32(self.xs)
self.es = numpy.float32(self.es)
def check_one_step(self):
x = self.context(numpy.concatenate((self.xs[0], self.es[0]))).reshape(1, 3)
t_x = self.context(self.xs[1]).reshape(1, 2)
t_e = self.context(numpy.array(self.es[1]).astype(numpy.int32)).reshape(1, 1)
self.h = self.model.initial_state(self.mini_batch, self.context, "h")
self.l = self.model.initial_state(self.mini_batch, self.context, "c")
self.h, self.l, self.loss = self.model.forward_one_step(self.h, self.l, x, t_x, t_e)
# self.loss.backward()
@nattr(local=True)
def test_forward_cpu(self):
self.mod = numpy
self.context = lambda x: x
self.check_one_step()
with open("gravesnet.dot", "w") as o:
o.write(c.build_computational_graph((self.loss,), False).dump())
@nattr(local=True)
@attr.gpu
def test_forward_gpu(self):
self.model.to_gpu()
self.mod = cuda
self.context = cuda.to_gpu
self.check_one_step()
def setUp(self):
self.model = GravesPredictionNet()
self.mini_batch = 1
self.shape = (self.mini_batch, 100)
self.xs, self.es = parse_IAMxml("res/strokesz.xml")
self.xs = numpy.float32(self.xs)
self.es = numpy.float32(self.es)