def main():
# I force using singularity for kilosort2 because on my computer, when docker tries to use gpu it messes up nvidia-container-cli and I need to restart the computer
import os
os.environ['HITHER_USE_SINGULARITY'] = 'TRUE'
jh = hither.SlurmJobHandler(
working_dir='tmp_slurm',
num_workers_per_batch=2,
num_cores_per_job=2,
use_slurm=True,
additional_srun_opts=['-p gpu', '--gres=gpu:v100-32gb:1'])
test_sort_tetrode(sorter_name='kilosort2',
min_avg_accuracy=0.15,
num_jobs=2,
job_handler=jh)
def main():
from spikeforest2 import sorters
from spikeforest2 import processing
parser = argparse.ArgumentParser(
description='Run the SpikeForest2 main analysis')
# parser.add_argument('analysis_file', help='Path to the analysis specification file (.json format).')
# parser.add_argument('--config', help='Configuration file', required=True)
# parser.add_argument('--output', help='Analysis output file (.json format)', required=True)
# parser.add_argument('--slurm', help='Optional SLURM configuration file (.json format)', required=False, default=None)
# parser.add_argument('--verbose', help='Provide some additional verbose output.', action='store_true')
parser.add_argument(
'spec',
help='Path to the .json file containing the analysis specification')
parser.add_argument('--output',
'-o',
help='The output .json file',
required=True)
parser.add_argument('--force-run',
help='Force rerunning of all spike sorting',
action='store_true')
parser.add_argument(
'--force-run-all',
help='Force rerunning of all spike sorting and other processing',
action='store_true')
parser.add_argument('--parallel',
help='Optional number of parallel jobs',
required=False,
default='0')
parser.add_argument('--slurm',
help='Path to slurm config file',
required=False,
default=None)
parser.add_argument('--cache',
help='The cache database to use',
required=False,
default=None)
parser.add_argument('--rerun-failing',
help='Rerun sorting jobs that previously failed',
action='store_true')
parser.add_argument('--test', help='Only run a few.', action='store_true')
parser.add_argument('--job-timeout',
help='Timeout for sorting jobs',
required=False,
default=600)
parser.add_argument('--log-file',
help='Log file for analysis progress',
required=False,
default=None)
args = parser.parse_args()
force_run_all = args.force_run_all
# the following apply to sorting jobs only
force_run = args.force_run or args.force_run_all
job_timeout = float(args.job_timeout)
cache_failing = True
rerun_failing = args.rerun_failing
with open(args.spec, 'r') as f:
spec = json.load(f)
# clear the log file
if args.log_file is not None:
with open(args.log_file, 'w'):
pass
studysets_path = spec['studysets']
studyset_names = spec['studyset_names']
spike_sorters = spec['spike_sorters']
ka.set_config(fr='default_readonly')
print(f'Loading study sets object from: {studysets_path}')
studysets_obj = ka.load_object(studysets_path)
if not studysets_obj:
raise Exception(f'Unable to load: {studysets_path}')
all_study_sets = studysets_obj['StudySets']
study_sets = []
for studyset in all_study_sets:
if studyset['name'] in studyset_names:
study_sets.append(studyset)
if int(args.parallel) > 0:
job_handler = hither.ParallelJobHandler(int(args.parallel))
job_handler_gpu = job_handler
job_handler_ks = job_handler
elif args.slurm:
with open(args.slurm, 'r') as f:
slurm_config = json.load(f)
job_handler = hither.SlurmJobHandler(working_dir='tmp_slurm',
**slurm_config['cpu'])
job_handler_gpu = hither.SlurmJobHandler(working_dir='tmp_slurm',
**slurm_config['gpu'])
job_handler_ks = hither.SlurmJobHandler(working_dir='tmp_slurm',
**slurm_config['ks'])
else:
job_handler = None
job_handler_gpu = None
job_handler_ks = None
with hither.config(container='default',
cache=args.cache,
force_run=force_run_all,
job_handler=job_handler,
log_path=args.log_file), hither.job_queue():
studies = []
recordings = []
for studyset in study_sets:
studyset_name = studyset['name']
print(f'================ STUDY SET: {studyset_name}')
studies0 = studyset['studies']
if args.test:
studies0 = studies0[:1]
studyset['studies'] = studies0
for study in studies0:
study['study_set'] = studyset_name
study_name = study['name']
print(f'======== STUDY: {study_name}')
recordings0 = study['recordings']
if args.test:
recordings0 = recordings0[:2]
study['recordings'] = recordings0
for recording in recordings0:
recording['study'] = study_name
recording['study_set'] = studyset_name
recording['firings_true'] = recording['firingsTrue']
recordings.append(recording)
studies.append(study)
# Download recordings
for recording in recordings:
ka.load_file(recording['directory'] + '/raw.mda')
ka.load_file(recording['directory'] + '/firings_true.mda')
# Attach results objects
for recording in recordings:
recording['results'] = dict()
# Summarize recordings
for recording in recordings:
recording_path = recording['directory']
sorting_true_path = recording['firingsTrue']
recording['results'][
'computed-info'] = processing.compute_recording_info.run(
_label=
f'compute-recording-info:{recording["study"]}/{recording["name"]}',
recording_path=recording_path,
json_out=hither.File())
recording['results'][
'true-units-info'] = processing.compute_units_info.run(
_label=
f'compute-units-info:{recording["study"]}/{recording["name"]}',
recording_path=recording_path,
sorting_path=sorting_true_path,
json_out=hither.File())
# Spike sorting
for sorter in spike_sorters:
for recording in recordings:
if recording['study_set'] in sorter['studysets']:
recording_path = recording['directory']
sorting_true_path = recording['firingsTrue']
algorithm = sorter['processor_name']
if not hasattr(sorters, algorithm):
raise Exception(
f'No such sorting algorithm: {algorithm}')
Sorter = getattr(sorters, algorithm)
if algorithm in ['ironclust']:
gpu = True
jh = job_handler_gpu
elif algorithm in ['kilosort', 'kilosort2']:
gpu = True
jh = job_handler_ks
else:
gpu = False
jh = job_handler
with hither.config(gpu=gpu,
force_run=force_run,
exception_on_fail=False,
cache_failing=cache_failing,
rerun_failing=rerun_failing,
job_handler=jh,
job_timeout=job_timeout):
sorting_result = Sorter.run(
_label=
f'{algorithm}:{recording["study"]}/{recording["name"]}',
recording_path=recording['directory'],
sorting_out=hither.File())
recording['results']['sorting-' +
sorter['name']] = sorting_result
recording['results'][
'comparison-with-truth-' +
sorter['name']] = processing.compare_with_truth.run(
_label=
f'comparison-with-truth:{algorithm}:{recording["study"]}/{recording["name"]}',
sorting_path=sorting_result.outputs.sorting_out,
sorting_true_path=sorting_true_path,
json_out=hither.File())
recording['results'][
'units-info-' +
sorter['name']] = processing.compute_units_info.run(
_label=
f'units-info:{algorithm}:{recording["study"]}/{recording["name"]}',
recording_path=recording_path,
sorting_path=sorting_result.outputs.sorting_out,
json_out=hither.File())
# Assemble all of the results
print('')
print('=======================================================')
print('Assembling results...')
for recording in recordings:
print(
f'Assembling recording: {recording["study"]}/{recording["name"]}')
recording['summary'] = dict(
plots=dict(),
computed_info=ka.load_object(
recording['results']['computed-info'].outputs.json_out._path),
true_units_info=ka.store_file(
recording['results']
['true-units-info'].outputs.json_out._path))
sorting_results = []
for sorter in spike_sorters:
for recording in recordings:
if recording['study_set'] in sorter['studysets']:
print(
f'Assembling sorting: {sorter["processor_name"]} {recording["study"]}/{recording["name"]}'
)
sorting_result = recording['results']['sorting-' +
sorter['name']]
comparison_result = recording['results'][
'comparison-with-truth-' + sorter['name']]
units_info_result = recording['results']['units-info-' +
sorter['name']]
console_out_str = _console_out_to_str(
sorting_result.runtime_info['console_out'])
console_out_path = ka.store_text(console_out_str)
sr = dict(
recording=recording,
sorter=sorter,
firings_true=recording['directory'] + '/firings_true.mda',
processor_name=sorter['processor_name'],
processor_version=sorting_result.version,
execution_stats=dict(
start_time=sorting_result.runtime_info['start_time'],
end_time=sorting_result.runtime_info['end_time'],
elapsed_sec=sorting_result.runtime_info['end_time'] -
sorting_result.runtime_info['start_time'],
retcode=0 if sorting_result.success else -1,
timed_out=sorting_result.runtime_info.get(
'timed_out', False)),
container=sorting_result.container,
console_out=console_out_path)
if sorting_result.success:
sr['firings'] = ka.store_file(
sorting_result.outputs.sorting_out._path)
sr['comparison_with_truth'] = dict(json=ka.store_file(
comparison_result.outputs.json_out._path))
sr['sorted_units_info'] = ka.store_file(
units_info_result.outputs.json_out._path)
else:
sr['firings'] = None
sr['comparison_with_truth'] = None
sr['sorted_units_info'] = None
sorting_results.append(sr)
# Delete results from recordings
for recording in recordings:
del recording['results']
# Aggregate sorting results
print('')
print('=======================================================')
print('Aggregating sorting results...')
aggregated_sorting_results = aggregate_sorting_results(
studies, recordings, sorting_results)
# Show output summary
for sr in aggregated_sorting_results['study_sorting_results']:
study_name = sr['study']
sorter_name = sr['sorter']
n1 = np.array(sr['num_matches'])
n2 = np.array(sr['num_false_positives'])
n3 = np.array(sr['num_false_negatives'])
accuracies = n1 / (n1 + n2 + n3)
avg_accuracy = np.mean(accuracies)
txt = 'STUDY: {}, SORTER: {}, AVG ACCURACY: {}'.format(
study_name, sorter_name, avg_accuracy)
print(txt)
output_object = dict(studies=studies,
recordings=recordings,
study_sets=study_sets,
sorting_results=sorting_results,
aggregated_sorting_results=ka.store_object(
aggregated_sorting_results,
basename='aggregated_sorting_results.json'))
print(f'Writing output to {args.output}...')
with open(args.output, 'w') as f:
json.dump(output_object, f, indent=4)
print('Done.')
def main():
from mountaintools import client as mt
parser = argparse.ArgumentParser(
description=
'Generate unit detail data (including spikesprays) for website')
parser.add_argument('analysis_path',
help='assembled analysis file (output.json)')
parser.add_argument(
'--studysets',
help='Comma-separated list of study set names to include',
required=False,
default=None)
parser.add_argument('--force-run',
help='Force rerunning of processing',
action='store_true')
parser.add_argument(
'--force-run-all',
help='Force rerunning of processing including filtering',
action='store_true')
parser.add_argument('--parallel',
help='Optional number of parallel jobs',
required=False,
default='0')
parser.add_argument('--slurm',
help='Path to slurm config file',
required=False,
default=None)
parser.add_argument('--cache',
help='The cache database to use',
required=False,
default=None)
parser.add_argument('--job-timeout',
help='Timeout for processing jobs',
required=False,
default=600)
parser.add_argument('--log-file',
help='Log file for analysis progress',
required=False,
default=None)
parser.add_argument(
'--force-regenerate',
help=
'Whether to force regenerating spike sprays (for when code has changed)',
action='store_true')
parser.add_argument('--test',
help='Whether to just test by running only 1',
action='store_true')
args = parser.parse_args()
mt.configDownloadFrom(['spikeforest.kbucket'])
with open(args.analysis_path, 'r') as f:
analysis = json.load(f)
if args.studysets is not None:
studyset_names = args.studysets.split(',')
print('Using study sets: ', studyset_names)
else:
studyset_names = None
study_sets = analysis['StudySets']
sorting_results = analysis['SortingResults']
studies_to_include = []
for ss in study_sets:
if (studyset_names is None) or (ss['name'] in studyset_names):
for study in ss['studies']:
studies_to_include.append(study['name'])
print('Including studies:', studies_to_include)
print('Determining sorting results to process ({} total)...'.format(
len(sorting_results)))
sorting_results_to_process = []
sorting_results_to_consider = []
for sr in sorting_results:
study_name = sr['studyName']
if study_name in studies_to_include:
if 'firings' in sr:
if sr.get('comparisonWithTruth', None) is not None:
sorting_results_to_consider.append(sr)
key = dict(name='unit-details-v0.1.0',
recording_directory=sr['recordingDirectory'],
firings_true=sr['firingsTrue'],
firings=sr['firings'])
val = mt.getValue(key=key, collection='spikeforest')
if (not val) or (args.force_regenerate):
sr['key'] = key
sorting_results_to_process.append(sr)
if args.test and len(sorting_results_to_process) > 0:
sorting_results_to_process = [sorting_results_to_process[0]]
print('Need to process {} of {} sorting results'.format(
len(sorting_results_to_process), len(sorting_results_to_consider)))
recording_directories_to_process = sorted(
list(
set([
sr['recordingDirectory'] for sr in sorting_results_to_process
])))
print('{} recording directories to process'.format(
len(recording_directories_to_process)))
if int(args.parallel) > 0:
job_handler = hither.ParallelJobHandler(int(args.parallel))
elif args.slurm:
with open(args.slurm, 'r') as f:
slurm_config = json.load(f)
job_handler = hither.SlurmJobHandler(working_dir='tmp_slurm',
**slurm_config['cpu'])
else:
job_handler = None
print('Filtering recordings...')
filter_results = []
with hither.config(container='default',
cache=args.cache,
force_run=args.force_run_all,
job_handler=job_handler,
log_path=args.log_file,
exception_on_fail=True,
cache_failing=False,
rerun_failing=True,
job_timeout=args.job_timeout), hither.job_queue():
for recdir in recording_directories_to_process:
result = filter_recording.run(recording_directory=recdir,
timeseries_out=hither.File())
filter_results.append(result)
filtered_timeseries_by_recdir = dict()
for i, recdir in enumerate(recording_directories_to_process):
result0 = filter_results[i]
if not result0.success:
raise Exception(
'Problem computing filtered timeseries for recording: {}'.
format(recdir))
filtered_timeseries_by_recdir[
recdir] = result0.outputs.timeseries_out._path
print('Creating spike sprays...')
with hither.config(container='default',
cache=args.cache,
force_run=args.force_run or args.force_run_all,
job_handler=job_handler,
log_path=args.log_file,
exception_on_fail=True,
cache_failing=False,
rerun_failing=True,
job_timeout=args.job_timeout), hither.job_queue():
for sr in sorting_results_to_process:
recdir = sr['recordingDirectory']
study_name = sr['studyName']
rec_name = sr['recordingName']
sorter_name = sr['sorterName']
print('====== COMPUTING {}/{}/{}'.format(study_name, rec_name,
sorter_name))
cwt = ka.load_object(path=sr['comparisonWithTruth']['json'])
filtered_timeseries = filtered_timeseries_by_recdir[recdir]
spike_spray_results = []
list0 = list(cwt.values())
for _, unit in enumerate(list0):
result = create_spike_sprays.run(
recording_directory=recdir,
filtered_timeseries=filtered_timeseries,
firings_true=os.path.join(recdir, 'firings_true.mda'),
firings_sorted=sr['firings'],
unit_id_true=unit['unit_id'],
unit_id_sorted=unit['best_unit'],
json_out=hither.File())
setattr(result, 'unit', unit)
spike_spray_results.append(result)
sr['spike_spray_results'] = spike_spray_results
for sr in sorting_results_to_process:
recdir = sr['recordingDirectory']
study_name = sr['studyName']
rec_name = sr['recordingName']
sorter_name = sr['sorterName']
print('====== SAVING {}/{}/{}'.format(study_name, rec_name,
sorter_name))
spike_spray_results = sr['spike_spray_results']
key = sr['key']
unit_details = []
ok = True
for i, result in enumerate(spike_spray_results):
if not result.success:
print(
'WARNING: Error creating spike sprays for {}/{}/{}'.format(
study_name, rec_name, sorter_name))
ok = False
break
ssobj = ka.load_object(result.outputs.json_out._path)
if ssobj is None:
raise Exception('Problem loading spikespray object output.')
address = mt.saveObject(object=ssobj,
upload_to='spikeforest.kbucket')
unit = getattr(result, 'unit')
unit_details.append(
dict(studyName=study_name,
recordingName=rec_name,
sorterName=sorter_name,
trueUnitId=unit['unit_id'],
sortedUnitId=unit['best_unit'],
spikeSprayUrl=mt.findFile(
path=address,
remote_only=True,
download_from='spikeforest.kbucket')))
if ok:
mt.saveObject(collection='spikeforest',
key=key,
object=unit_details,
upload_to='spikeforest.public')