"...git@developer.sourcefind.cn:OpenDAS/mmdetection3d.git" did not exist on "92018ce133423492d34a26bf1bef47de7799852a"
Commit 4b81982e authored by Hongkun Yu's avatar Hongkun Yu Committed by A. Unique TensorFlower
Browse files

Make two beam search utils function as public. They are helpful.

PiperOrigin-RevId: 341953641
parent 4b3214c0
...@@ -188,8 +188,8 @@ class SequenceBeamSearch(tf.Module): ...@@ -188,8 +188,8 @@ class SequenceBeamSearch(tf.Module):
tf.slice(alive_seq, [0, 0, i], [batch_size, self.beam_size, 1]), tf.slice(alive_seq, [0, 0, i], [batch_size, self.beam_size, 1]),
[batch_size * self.beam_size, -1]) [batch_size * self.beam_size, -1])
else: else:
flat_ids = _flatten_beam_dim(alive_seq) # [batch_size * beam_size] flat_ids = flatten_beam_dim(alive_seq) # [batch_size * beam_size]
flat_cache = tf.nest.map_structure(_flatten_beam_dim, alive_cache) flat_cache = tf.nest.map_structure(flatten_beam_dim, alive_cache)
flat_logits, flat_cache = self.symbols_to_logits_fn( flat_logits, flat_cache = self.symbols_to_logits_fn(
flat_ids, i, flat_cache) flat_ids, i, flat_cache)
...@@ -404,7 +404,7 @@ class SequenceBeamSearch(tf.Module): ...@@ -404,7 +404,7 @@ class SequenceBeamSearch(tf.Module):
cur_index = tf.constant(0) cur_index = tf.constant(0)
# Create alive sequence with shape [batch_size, beam_size, 1] # Create alive sequence with shape [batch_size, beam_size, 1]
alive_seq = _expand_to_beam_size(initial_ids, self.beam_size) alive_seq = expand_to_beam_size(initial_ids, self.beam_size)
alive_seq = tf.expand_dims(alive_seq, axis=2) alive_seq = tf.expand_dims(alive_seq, axis=2)
if self.padded_decode: if self.padded_decode:
alive_seq = tf.tile(alive_seq, [1, 1, self.max_decode_length + 1]) alive_seq = tf.tile(alive_seq, [1, 1, self.max_decode_length + 1])
...@@ -419,7 +419,7 @@ class SequenceBeamSearch(tf.Module): ...@@ -419,7 +419,7 @@ class SequenceBeamSearch(tf.Module):
# Expand all values stored in the dictionary to the beam size, so that each # Expand all values stored in the dictionary to the beam size, so that each
# beam has a separate cache. # beam has a separate cache.
alive_cache = tf.nest.map_structure( alive_cache = tf.nest.map_structure(
lambda t: _expand_to_beam_size(t, self.beam_size), initial_cache) lambda t: expand_to_beam_size(t, self.beam_size), initial_cache)
# Initialize tensor storing finished sequences with filler values. # Initialize tensor storing finished sequences with filler values.
finished_seq = tf.zeros(tf.shape(alive_seq), tf.int32) finished_seq = tf.zeros(tf.shape(alive_seq), tf.int32)
...@@ -588,7 +588,7 @@ def _length_normalization(alpha, length, dtype=tf.float32): ...@@ -588,7 +588,7 @@ def _length_normalization(alpha, length, dtype=tf.float32):
return tf.pow(((5. + tf.cast(length, dtype)) / 6.), alpha) return tf.pow(((5. + tf.cast(length, dtype)) / 6.), alpha)
def _expand_to_beam_size(tensor, beam_size): def expand_to_beam_size(tensor, beam_size):
"""Tiles a given tensor by beam_size. """Tiles a given tensor by beam_size.
Args: Args:
...@@ -605,6 +605,21 @@ def _expand_to_beam_size(tensor, beam_size): ...@@ -605,6 +605,21 @@ def _expand_to_beam_size(tensor, beam_size):
return tf.tile(tensor, tile_dims) return tf.tile(tensor, tile_dims)
def flatten_beam_dim(tensor):
"""Reshapes first two dimensions into a single dimension.
Args:
tensor: Tensor to reshape of shape [A, B, ...]
Returns:
Reshaped tensor of shape [A*B, ...]
"""
shape = _shape_list(tensor)
shape[0] *= shape[1]
shape.pop(1) # Remove beam dim
return tf.reshape(tensor, shape)
def _shape_list(tensor): def _shape_list(tensor):
"""Return a list of the tensor's shape, and ensure no None values in list.""" """Return a list of the tensor's shape, and ensure no None values in list."""
# Get statically known shape (may contain None's for unknown dimensions) # Get statically known shape (may contain None's for unknown dimensions)
...@@ -630,21 +645,6 @@ def _get_shape_keep_last_dim(tensor): ...@@ -630,21 +645,6 @@ def _get_shape_keep_last_dim(tensor):
return tf.TensorShape(shape_list) return tf.TensorShape(shape_list)
def _flatten_beam_dim(tensor):
"""Reshapes first two dimensions in to single dimension.
Args:
tensor: Tensor to reshape of shape [A, B, ...]
Returns:
Reshaped tensor of shape [A*B, ...]
"""
shape = _shape_list(tensor)
shape[0] *= shape[1]
shape.pop(1) # Remove beam dim
return tf.reshape(tensor, shape)
def _unflatten_beam_dim(tensor, batch_size, beam_size): def _unflatten_beam_dim(tensor, batch_size, beam_size):
"""Reshapes first dimension back to [batch_size, beam_size]. """Reshapes first dimension back to [batch_size, beam_size].
......
...@@ -24,7 +24,7 @@ class BeamSearchTests(tf.test.TestCase, parameterized.TestCase): ...@@ -24,7 +24,7 @@ class BeamSearchTests(tf.test.TestCase, parameterized.TestCase):
def test_expand_to_beam_size(self): def test_expand_to_beam_size(self):
x = tf.ones([7, 4, 2, 5]) x = tf.ones([7, 4, 2, 5])
x = beam_search._expand_to_beam_size(x, 3) x = beam_search.expand_to_beam_size(x, 3)
shape = tf.shape(x) shape = tf.shape(x)
self.assertAllEqual([7, 3, 4, 2, 5], shape) self.assertAllEqual([7, 3, 4, 2, 5], shape)
...@@ -36,7 +36,7 @@ class BeamSearchTests(tf.test.TestCase, parameterized.TestCase): ...@@ -36,7 +36,7 @@ class BeamSearchTests(tf.test.TestCase, parameterized.TestCase):
def test_flatten_beam_dim(self): def test_flatten_beam_dim(self):
x = tf.ones([7, 4, 2, 5]) x = tf.ones([7, 4, 2, 5])
x = beam_search._flatten_beam_dim(x) x = beam_search.flatten_beam_dim(x)
self.assertAllEqual([28, 2, 5], tf.shape(x)) self.assertAllEqual([28, 2, 5], tf.shape(x))
def test_unflatten_beam_dim(self): def test_unflatten_beam_dim(self):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment