def pack_all(self, output_dict: Dict):
r"""Pack the prediction results ``output_dict`` back to the
corresponding packs.
"""
start = 0
for i in range(len(self.batcher.data_pack_pool)):
output_dict_i = slice_batch(output_dict, start,
self.batcher.current_batch_sources[i])
self.pack(self.batcher.data_pack_pool[i], output_dict_i)
start += self.batcher.current_batch_sources[i]
def pack_all(
self,
packs: List[PackType],
contexts: List[Optional[Annotation]],
output_dict: Dict[str, List[Any]],
):
r"""
Pack the prediction results contained in the `output_dict` back to the
corresponding packs.
Args:
packs: The list of data packs corresponding to the output batches.
contexts: The list of contexts corresponding to the output batches.
output_dict: Stores the output in a specific format. The keys
are string names that specify data. The value is a list of
data in the shape of (batch_size, Any). There might be
additional structures inside `Any` as specific
implementation choices.
"""
# Group the same pack and context into the same segments.
# The list of (pack, context) tuple.
pack_context_pool = []
# Store the segments of the pack context, the len of this list should
# be the number of segments, and the value indicates the length of
# the segment. These will be used to slice the data batch.
segment_lengths = []
# Note that this will work if the elements in `contexts` are all
# None, which means they are all the same, and will have the correct
# behavior.
prev_pack_context = None
for pack_context_i in zip(packs, contexts):
if pack_context_i != prev_pack_context:
segment_lengths.append(1)
prev_pack_context = pack_context_i
pack_context_pool.append(pack_context_i)
else:
segment_lengths[-1] += 1
start = 0
for i, (pack_i, context) in enumerate(pack_context_pool):
# The slice should correspond to the portion of the data batch
# that should be assigned to these pack and context.
output_dict_i = slice_batch(output_dict, start, segment_lengths[i])
self.pack(pack_i, output_dict_i, context)
start += segment_lengths[i]
pack_i.add_all_remaining_entries()
def pack_all(self, packs: List[PackType], output_dict: Dict):
r"""Pack the prediction results ``output_dict`` back to the
corresponding packs.
"""
data_pack_pool = []
current_batch_sources = []
prev_pack = None
for pack_i in packs:
if pack_i != prev_pack:
current_batch_sources.append(1)
prev_pack = pack_i
data_pack_pool.append(pack_i)
else:
current_batch_sources[-1] += 1
start = 0
for i, pack_i in enumerate(data_pack_pool):
output_dict_i = slice_batch(output_dict, start,
current_batch_sources[i])
self.pack(pack_i, output_dict_i)
start += current_batch_sources[i]
pack_i.add_all_remaining_entries()