def test_alignments_to_paths(self):
exp_out = tf.convert_to_tensor(
[[[1., 0., 0., 0., 0., 0., 0.], [0., 2., 5., 0., 0., 0., 0.],
[0., 0., 0., 3., 0., 0., 0.], [0., 0., 0., 7., 0., 0., 0.],
[0., 0., 0., 9., 0., 0., 0.], [0., 0., 0., 0., 4., 0., 0.],
[0., 0., 0., 0., 0., 2., 0.], [0., 0., 0., 0., 0., 0., 2.]]],
tf.float32)
paths = alignment.alignments_to_paths(self.alignments, self.len_x,
self.len_y)
sq_paths = alignment.path_label_squeeze(paths)
self.assertAllEqual(sq_paths, exp_out)
# Invariance to padding.
padded_alignments = tf.concat(
[self.alignments, tf.zeros([1, 3, 3], tf.int32)], 2)
paths = alignment.alignments_to_paths(padded_alignments, self.len_x,
self.len_y)
sq_paths = alignment.path_label_squeeze(paths)
self.assertAllEqual(sq_paths, exp_out)
# Pads correctly via length arguments.
paths = alignment.alignments_to_paths(self.alignments, self.len_x + 3,
self.len_y + 3)
sq_paths = alignment.path_label_squeeze(paths)
self.assertAllEqual(sq_paths[:, :self.len_x, :self.len_y], exp_out)
self.assertAllEqual(sq_paths[:, self.len_x:, :],
tf.zeros([1, 3, self.len_y + 3], tf.float32))
self.assertAllEqual(sq_paths[Ellipsis, self.len_y:],
tf.zeros([1, self.len_x + 3, 3], tf.float32))
# Deals with empty ground-truth alignments.
paths = alignment.alignments_to_paths(tf.zeros_like(self.alignments),
self.len_x, self.len_y)
sq_paths = alignment.path_label_squeeze(paths)
self.assertAllEqual(sq_paths, tf.zeros_like(exp_out))
def update_state(self,
alignments_true,
alignments_pred,
sample_weight=None):
"""Updates mean squared error for a batch of true vs pred alignments."""
if alignments_pred[1] is None:
return
sol_paths_pred = alignments_pred[1]
len_x, len_y = tf.shape(sol_paths_pred)[1], tf.shape(sol_paths_pred)[2]
sol_paths_true = alignment.alignments_to_paths(alignments_true, len_x,
len_y)
mse = tf.reduce_sum((sol_paths_pred - sol_paths_true)**2,
axis=[1, 2, 3])
super().update_state(mse, sample_weight)
def test_sw_score(self):
# Test sw_score from sparse representation.
sw_score = alignment.sw_score(self.sw_params, self.alignments)
self.assertAllEqual(sw_score, [95.0])
# Test sw_score from dense representation.
paths = alignment.alignments_to_paths(self.alignments, self.len_x,
self.len_y +
1) # Testing padding too.
sw_score = alignment.sw_score(self.sw_params, paths)
self.assertAllEqual(sw_score, [95.0])
# Test empty alignments / paths
sw_score = alignment.sw_score(self.sw_params,
tf.zeros_like(self.alignments))
self.assertAllEqual(sw_score, [0.0])
sw_score = alignment.sw_score(self.sw_params, tf.zeros_like(paths))
self.assertAllEqual(sw_score, [0.0])