def generate_psth_from_resp(rec,
epoch_regex='^STIM_',
smooth_resp=False,
**context):
'''
generate PSTH prediction from rec['resp'] (before est/val split). Could
be considered "cheating" b/c predicted PSTH then is based on data in
val set, but this is because we're interested in testing state effects,
not sensory coding models. The appropriate control, however is to run
generate_psth_from_est_for_both_est_and_val_nfold on each nfold est/val
split.
'''
rec = preproc.generate_psth_from_resp(rec,
epoch_regex,
smooth_resp=smooth_resp)
return {'rec': rec}
state_signals=['pupil'],
permute_signals=[''],
new_signalname='state')
rec = preproc.make_state_signal(rec,
state_signals=['pupil'],
permute_signals=['pupil'],
new_signalname='state') #shuffled pupil
# mask out data from incorrect trials
#rec = preproc.mask_all_but_correct_references(rec)#not for allen?
# calculate a PSTH response for each stimulus, save to a new signal 'psth'
epoch_regex = "^STIM_"
#epoch_regex="^natural_scene" #only natural scene stimulus
rec = preproc.generate_psth_from_resp(rec,
epoch_regex=epoch_regex,
smooth_resp=False)
#rec=nwb_resp_psth(rec,epoch_regex) #should return similar results as above for allen data
# ----------------------------------------------------------------------------
# INSPECT THE DATA
resp = rec['resp'].rasterize()
epochs = resp.epochs
epoch_regex = "^STIM_"
#epoch_regex="^natural_scene" #only natural scene stimulus
epoch_list = ep.epoch_names_matching(epochs, epoch_regex)
# list all stimulus events
print(epochs[epochs['name'].isin(epoch_list)])
subwin = 0.25 # sub window size (mean rate across all / sd across all)
# CV = sd of spike counts across all subwindows divided by the mean across all sub windows
# If all neurons are Poisson and statistically independent, then the CV of the population rate will approach zero
site = 'TAR010c'
batch = 289
manager = BAPHYExperiment(cellid=site, batch=batch)
options = {'rasterfs': 4, 'resp': True, 'stim': False, 'pupil': True}
rec = manager.get_recording(**options)
rec['resp'] = rec['resp'].rasterize()
if batch == 331:
rec = nems_preproc.fix_cpn_epochs(rec)
else:
rec = nems_preproc.mask_high_repetion_stims(rec)
rec = generate_psth_from_resp(rec)
# extract continuous data (subtract psth?)
data = rec.apply_mask()['resp']._data #- rec.apply_mask()['psth_sp']._data
pupil = rec.apply_mask()['pupil']._data
# divide into bins
win_bin = int(rec['resp'].fs * win)
subwin_bin = int(rec['resp'].fs * subwin)
CV = []
bpupil = []
i = 0
while ((i * win_bin) <= data.shape[-1]):
mean = data[:, int(i * win_bin):int((i + 1) * win_bin)].mean()
sd = data[:, int(i * win_bin):int((i + 1) * win_bin)].std()
CV.append(sd / mean)
#manager = BAPHYExperiment(cellid='ARM033a', batch=331)
#manager = BAPHYExperiment(cellid='AMT026a', batch=331)
#manager = BAPHYExperiment(cellid='CRD018d', batch=331)
#manager = BAPHYExperiment(cellid='AMT020a', batch=331)
#manager = BAPHYExperiment(cellid='ARM031a', batch=331)
r = manager.get_recording(recache=True,
**{
'rasterfs': 4,
'resp': True,
'pupil': True,
'stim': False,
'pupil_variable_name': 'area'
})
r['resp'] = r['resp'].rasterize()
r = fix_cpn_epochs(r)
r = generate_psth_from_resp(r)
epochs = [e for e in r['resp'].epochs.name.unique() if e.startswith('STIM')]
r = r.and_mask(epochs)
binsizes = [0.5, 1, 2] #, 3, 4]
thresh = 1
significant_pairs = False
for binsize in binsizes:
rec = r.copy()
# before masking, take raw signals and compute variance of an eyelid over a sliding window
binsize = int(binsize * rec['resp'].fs)
signal = rec['pupil_extras'].extract_channels(['eyelid_top_y'])._data
signal2 = rec['pupil_extras'].extract_channels(['eyelid_bottom_y'])._data
varsig = np.zeros(signal.shape)
for i in range(varsig.shape[-1]):