def make_examples_runner(options):
"""Runs examples creation stage of deepvariant."""
resource_monitor = resources.ResourceMonitor().start()
logging.info('Preparing inputs')
regions = processing_regions_from_options(options)
# Create a processor to create candidates and examples for each region.
region_processor = RegionProcessor(options)
logging.info('Writing examples to %s', options.examples_filename)
if options.candidates_filename:
logging.info('Writing candidates to %s', options.candidates_filename)
if options.gvcf_filename:
logging.info('Writing gvcf records to %s', options.gvcf_filename)
n_regions, n_candidates, n_examples = 0, 0, 0
with OutputsWriter(options) as writer:
for region in regions:
candidates, examples, gvcfs = region_processor.process(region)
n_candidates += len(candidates)
n_examples += len(examples)
n_regions += 1
writer.write_candidates(*candidates)
# If we have any gvcf records, write them out. This if also serves to
# protect us from trying to write to the gvcfs output of writer when gvcf
# generation is turned off. In that case, gvcfs will always be empty and
# we'll never execute the write.
if gvcfs:
writer.write_gvcfs(*gvcfs)
writer.write_examples(*examples)
# Construct and then write out our MakeExamplesRunInfo proto.
if options.run_info_filename:
run_info = deepvariant_pb2.MakeExamplesRunInfo(
options=options, resource_metrics=resource_monitor.metrics())
if in_training_mode(options):
if region_processor.labeler.metrics is not None:
run_info.labeling_metrics.CopyFrom(
region_processor.labeler.metrics)
else:
logging.warning(
'Labeling metrics requested but the selected labeling '
'algorithm %s does not collect metrics; skipping.',
options.labeler_algorithm)
logging.info('Writing MakeExamplesRunInfo to %s',
options.run_info_filename)
write_make_examples_run_info(run_info, path=options.run_info_filename)
logging.info('Found %s candidate variants', n_candidates)
logging.info('Created %s examples', n_examples)
def read_make_examples_run_info(path):
"""Reads a MakeExamplesRunInfo proto in text_format from path."""
with tf.gfile.GFile(path) as f:
return text_format.Parse(f.read(),
deepvariant_pb2.MakeExamplesRunInfo())