import time
import h5py_wrapper.wrapper
try:
job_parameter = int(os.environ['SLURM_ARRAY_TASK_ID'])
save_res = True
except:
job_parameter = 1
save_res = False
# =============================================================================
# Load Spinnaker data
# =============================================================================
if job_parameter == 0:
filename = '../../data/Spinnaker_Data/results/spikes_L5E.h5'
session = neo.NeoHdf5IO(filename=filename)
block = session.read_block()
sts = block.list_children_by_class(neo.SpikeTrain)[:100]
print("Number of spinnaker spike trains: " + str(len(sts)))
# =============================================================================
# Load Nest data
# =============================================================================
if job_parameter == 1:
filename = '../../data/Nest_Data/example_output_10500ms_nrec_100/spikes_L5E.h5'
session = neo.NeoHdf5IO(filename=filename)
sts = []
for k in range(100):
def crosscorrelogram_task(inputdata, number_of_jobs, job_id):
'''
Task Manifest Version: 1
Full Name: crosscorrelogram_task
Caption: cross-correlogram
Author: Elephant-Developers
Description: |
This task calculates all pair-wise cross-correlograms between all
combinations of spike trains in the input file. Significance of
the correlation is evaluated based on spike-dither surrogates.
Categories:
- FDAT
Compatible_queues: ['cscs_viz']
Accepts:
inputdata: application/unknown
number_of_jobs: long
job_id: long
Returns:
res: application/unknown
'''
import quantities as pq
import neo
import elephant
if job_id > number_of_jobs:
print "Input data is invalid, exiting"
return
# =========================================================================
# Load data
# =========================================================================
# stage the input file
original_path = crosscorrelogram_task.task.uri.get_file(inputdata)
session = neo.NeoHdf5IO(filename=original_path)
block = session.read_block()
# select spike trains
sts = block.filter(use_st=True)
# print("Number of spike trains: " + str(len(sts)))
# =========================================================================
# Cross-correlograms
# =========================================================================
max_lag_bins = 200
lag_res = 1 * pq.ms
max_lag = max_lag_bins * lag_res
smoothing = 10 * pq.ms
num_neurons = len(sts)
cc = {}
cc['unit_i'] = {}
cc['unit_j'] = {}
cc['times_ms'] = {}
cc['original'] = {}
cc['surr'] = {}
cc['original_measure'] = {}
cc['surr_measure'] = {}
cc['pvalue'] = {}
# create all combinations of tasks
num_total_pairs = 0
all_combos_unit_i = []
all_combos_unit_j = []
for ni in range(num_neurons):
for nj in range(ni, num_neurons):
all_combos_unit_i.append(ni)
all_combos_unit_j.append(nj)
num_total_pairs += 1
# calculate indices in cc['unit_i'] list which to calculate for each task
idx = np.linspace(0, num_total_pairs, number_of_jobs + 1, dtype=int)
task_starts_idx = idx[:-1]
task_stop_idx = idx[1:]
# Loop over all pairs of neurons
for calc_i in range(task_starts_idx[job_id], task_stop_idx[job_id]):
# save neuron i,j index
ni = all_combos_unit_i[calc_i]
nj = all_combos_unit_j[calc_i]
cc['unit_i'][calc_i] = ni
cc['unit_j'][calc_i] = nj
print("Cross-correlating %i and %i" % (ni, nj))
# original CCH
cco = elephant.spikecorr.cch(sts[ni],
sts[nj],
w=lag_res,
lag=max_lag,
smooth=smoothing)
cc['original'][calc_i] = cco.magnitude
cc['times_ms'][calc_i] = cco.times.rescale(pq.ms).magnitude
# extract measure
ind = np.argmin(np.abs(cco.times))
ccom = cch_measure(cco, ind)
cc['original_measure'][calc_i] = ccom
surr_i = elephant.surrogates.spike_dithering(sts[ni],
dither=50. * pq.ms,
n=num_surrs)
surr_j = elephant.surrogates.spike_dithering(sts[nj],
dither=50. * pq.ms,
n=num_surrs)
ccs = []
ccsm = []
# cross-correlogram of each surrogate pair
for surrogate in range(num_surrs):
scc = elephant.spikecorr.cch(surr_i[surrogate],
surr_j[surrogate],
w=lag_res,
lag=max_lag,
smooth=smoothing)
ccs.append(scc.magnitude)
ccsm.append(cch_measure(scc, ind))
cc['surr'][calc_i] = np.array(ccs)
cc['surr_measure'][calc_i] = np.sort(ccsm)
cc['pvalue'][calc_i] = np.count_nonzero(np.array(ccsm) >= ccom)
# save result to hdf5
outputname = 'cc_result' + str(number_of_jobs) + '_' + str(job_id) + '.h5'
export_hdf5(cc, outputname)
return crosscorrelogram_task.task.uri.save_file(mime_type='\
application/unknown',
src_path=outputname,
dst_path=outputname)