def test_pruned_logsum(self):
forward_logsum_weights = k2host.DoubleArray1.create_array_with_size(
self.num_states)
backward_logsum_weights = k2host.DoubleArray1.create_array_with_size(
self.num_states)
wfsa = k2host.WfsaWithFbWeights(self.fsa,
k2host.FbWeightType.kLogSumWeight,
forward_logsum_weights,
backward_logsum_weights)
beam = 10.0
determinizer = k2host.DeterminizerPrunedLogSum(
wfsa, beam, 100, k2host.FbWeightType.kNoWeight)
fsa_size = k2host.IntArray2Size()
arc_derivs_size = k2host.IntArray2Size()
determinizer.get_sizes(fsa_size, arc_derivs_size)
fsa_out = k2host.Fsa.create_fsa_with_size(fsa_size)
arc_derivs = k2host.LogSumArcDerivs.create_arc_derivs_with_size(
arc_derivs_size)
arc_weights_out = k2host.FloatArray1.create_array_with_size(
fsa_size.size2)
determinizer.get_output(fsa_out, arc_derivs)
self.assertTrue(k2host.is_deterministic(fsa_out))
self.assertEqual(fsa_out.size1, 7)
self.assertEqual(fsa_out.size2, 9)
self.assertEqual(arc_derivs.size1, 9)
self.assertEqual(arc_derivs.size2, 15)
self.assertTrue(
k2host.is_rand_equivalent_logsum_weight(self.fsa, fsa_out, beam))
# cast float to int
arc_ids = k2host.StridedIntArray1.from_float_tensor(arc_derivs.data[:,
0])
def test_logsum_weight(self):
self.assertTrue(
k2host.is_rand_equivalent_logsum_weight(self.fsa_a, self.fsa_b))
self.assertFalse(
k2host.is_rand_equivalent_logsum_weight(self.fsa_a, self.fsa_c))