Esempio n. 1
0
 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])
Esempio n. 2
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))