def _get_frametimings(self):
filter_length, frametimings = asc.ft_nblinks(self.exp, self.stimnr,
self.pars.nblinks,
self.refresh_rate)
if self.maxframes is None:
frametimings = frametimings[:-1]
self.frametimings = frametimings
self.filter_length = filter_length
self.frame_duration = np.ediff1d(frametimings).mean()
def get_frametimings(self):
try:
filter_length, frametimings = asc.ft_nblinks(
self.exp, self.stimnr, self.param_file.get('Nblinks', None),
self.refresh_rate)
except ValueError as e:
if str(e).startswith('Unexpected value for nblinks'):
frametimings = asc.readframetimes(self.exp, self.stimnr)
filter_length = None
frametimings = frametimings[:self.maxframes]
self.filter_length = filter_length
self.frametimings = frametimings
self.frame_duration = np.ediff1d(frametimings).mean()
def _get_frametimings(self):
filter_length, frametimings = asc.ft_nblinks(self.exp, self.stimnr,
self.pars.nblinks,
self.refresh_rate)
self.frametimings = frametimings
self.filter_length = filter_length
def checkerflickerplusanalyzer(exp_name,
stimulusnr,
clusterstoanalyze=None,
frametimingsfraction=None,
cutoff=4):
"""
Analyzes checkerflicker-like data, typically interspersed
stimuli in between chunks of checkerflicker.
e.g. checkerflickerplusmovie, frozennoise
Parameters:
----------
exp_name:
Experiment name.
stimulusnr:
Number of the stimulus to be analyzed.
clusterstoanalyze:
Number of clusters should be analyzed. Default is None.
First N cells will be analyzed if this parameter is given.
In case of long recordings it might make sense to first
look at a subset of cells before starting to analyze
the whole dataset.
frametimingsfraction:
Fraction of the recording to analyze. Should be a number
between 0 and 1. e.g. 0.3 will analyze the first 30% of
the whole recording.
cutoff:
Worst rating that is wanted for the analysis. Default
is 4. The source of this value is manual rating of each
cluster.
"""
exp_dir = iof.exp_dir_fixer(exp_name)
stimname = iof.getstimname(exp_dir, stimulusnr)
exp_name = os.path.split(exp_dir)[-1]
clusters, metadata = asc.read_spikesheet(exp_dir, cutoff=cutoff)
# Check that the inputs are as expected.
if clusterstoanalyze:
if clusterstoanalyze > len(clusters[:, 0]):
warnings.warn('clusterstoanalyze is larger '
'than number of clusters in dataset. '
'All cells will be included.')
clusterstoanalyze = None
if frametimingsfraction:
if not 0 < frametimingsfraction < 1:
raise ValueError('Invalid input for frametimingsfraction: {}. '
'It should be a number between 0 and 1'
''.format(frametimingsfraction))
scr_width = metadata['screen_width']
scr_height = metadata['screen_height']
refresh_rate = metadata['refresh_rate']
parameters = asc.read_parameters(exp_dir, stimulusnr)
stx_h = parameters['stixelheight']
stx_w = parameters['stixelwidth']
# Check whether any parameters are given for margins, calculate
# screen dimensions.
marginkeys = ['tmargin', 'bmargin', 'rmargin', 'lmargin']
margins = []
for key in marginkeys:
margins.append(parameters.get(key, 0))
# Subtract bottom and top from vertical dimension; left and right
# from horizontal dimension
scr_width = scr_width - sum(margins[2:])
scr_height = scr_height - sum(margins[:2])
nblinks = parameters['Nblinks']
bw = parameters.get('blackwhite', False)
# Gaussian stimuli are not supported yet, we need to ensure we
# have a black and white stimulus
if bw is not True:
raise ValueError('Gaussian stimuli are not supported yet!')
seed = parameters.get('seed', -1000)
sx, sy = scr_height / stx_h, scr_width / stx_w
# Make sure that the number of stimulus pixels are integers
# Rounding down is also possible but might require
# other considerations.
if sx % 1 == 0 and sy % 1 == 0:
sx, sy = int(sx), int(sy)
else:
raise ValueError('sx and sy must be integers')
filter_length, frametimings = asc.ft_nblinks(exp_dir, stimulusnr)
if parameters['stimulus_type'] in [
'FrozenNoise', 'checkerflickerplusmovie'
]:
runfr = parameters['RunningFrames']
frofr = parameters['FrozenFrames']
# To generate the frozen noise, a second seed is used.
# The default value of this is -10000 as per StimulateOpenGL
secondseed = parameters.get('secondseed', -10000)
if parameters['stimulus_type'] == 'checkerflickerplusmovie':
mblinks = parameters['Nblinksmovie']
# Retrivee the number of frames (files) from parameters['path']
ipath = PureWindowsPath(parameters['path']).as_posix()
repldict = iof.config('stimuli_path_replace')
for needle, repl in repldict.items():
ipath = ipath.replace(needle, repl)
ipath = os.path.normpath(ipath) # Windows compatiblity
moviefr = len([
name for name in os.listdir(ipath)
if os.path.isfile(os.path.join(ipath, name))
and name.lower().endswith('.raw')
])
noiselen = (runfr + frofr) * nblinks
movielen = moviefr * mblinks
triallen = noiselen + movielen
ft_on, ft_off = asc.readframetimes(exp_dir,
stimulusnr,
returnoffsets=True)
frametimings = np.empty(ft_on.shape[0] * 2, dtype=float)
frametimings[::2] = ft_on
frametimings[1::2] = ft_off
import math
ntrials = math.floor(frametimings.size / triallen)
trials = np.zeros((ntrials, runfr + frofr + moviefr))
for t in range(ntrials):
frange = frametimings[t * triallen:(t + 1) * triallen]
trials[t, :runfr + frofr] = frange[:noiselen][::nblinks]
trials[t, runfr + frofr:] = frange[noiselen:][::mblinks]
frametimings = trials.ravel()
filter_length = np.int(np.round(.666 * refresh_rate / nblinks))
# Add frozen movie to frozen noise (for masking)
frofr += moviefr
savefname = str(stimulusnr) + '_data'
if clusterstoanalyze:
clusters = clusters[:clusterstoanalyze, :]
print('Analyzing first %s cells' % clusterstoanalyze)
savefname += '_' + str(clusterstoanalyze) + 'cells'
if frametimingsfraction:
frametimingsindex = int(len(frametimings) * frametimingsfraction)
frametimings = frametimings[:frametimingsindex]
print('Analyzing first {}% of'
' the recording'.format(frametimingsfraction * 100))
savefname += '_' + str(frametimingsfraction).replace('.',
'') + 'fraction'
frame_duration = np.average(np.ediff1d(frametimings))
total_frames = frametimings.shape[0]
all_spiketimes = []
# Store spike triggered averages in a list containing correct shaped
# arrays
stas = []
for i in range(len(clusters[:, 0])):
spiketimes = asc.read_raster(exp_dir, stimulusnr, clusters[i, 0],
clusters[i, 1])
spikes = asc.binspikes(spiketimes, frametimings)
all_spiketimes.append(spikes)
stas.append(np.zeros((sx, sy, filter_length)))
# Separate out the repeated parts
all_spiketimes = np.array(all_spiketimes)
mask = runfreezemask(total_frames, runfr, frofr, refresh_rate)
repeated_spiketimes = all_spiketimes[:, ~mask]
run_spiketimes = all_spiketimes[:, mask]
# We need to cut down the total_frames by the same amount
# as spiketimes
total_run_frames = run_spiketimes.shape[1]
# To be able to use the same code as checkerflicker analyzer,
# convert to list again.
run_spiketimes = list(run_spiketimes)
# Empirically determined to be best for 32GB RAM
desired_chunk_size = 21600000
# Length of the chunks (specified in number of frames)
chunklength = int(desired_chunk_size / (sx * sy))
chunksize = chunklength * sx * sy
nrofchunks = int(np.ceil(total_run_frames / chunklength))
print(f'\nAnalyzing {stimname}.\nTotal chunks: {nrofchunks}')
time = startime = datetime.datetime.now()
timedeltas = []
quals = np.zeros(len(stas))
frame_counter = 0
for i in range(nrofchunks):
randnrs, seed = randpy.ranb(seed, chunksize)
# Reshape and change 0's to -1's
stimulus = np.reshape(randnrs,
(sx, sy, chunklength), order='F') * 2 - 1
del randnrs
# Range of indices we are interested in for the current chunk
if (i + 1) * chunklength < total_run_frames:
chunkind = slice(i * chunklength, (i + 1) * chunklength)
chunkend = chunklength
else:
chunkind = slice(i * chunklength, None)
chunkend = total_run_frames - i * chunklength
for k in range(filter_length, chunkend - filter_length + 1):
stim_small = stimulus[:, :,
k - filter_length + 1:k + 1][:, :, ::-1]
for j in range(clusters.shape[0]):
spikes = run_spiketimes[j][chunkind]
if spikes[k] != 0:
stas[j] += spikes[k] * stim_small
qual = np.array([])
for c in range(clusters.shape[0]):
qual = np.append(qual, asc.staquality(stas[c]))
quals = np.vstack((quals, qual))
# Draw progress bar
width = 50 # Number of characters
prog = i / (nrofchunks - 1)
bar_complete = int(prog * width)
bar_noncomplete = width - bar_complete
timedeltas.append(msc.timediff(time)) # Calculate running avg
avgelapsed = np.mean(timedeltas)
elapsed = np.sum(timedeltas)
etc = startime + elapsed + avgelapsed * (nrofchunks - i)
sys.stdout.flush()
sys.stdout.write('\r{}{} |{:4.1f}% ETC: {}'.format(
'█' * bar_complete, '-' * bar_noncomplete, prog * 100,
etc.strftime("%a %X")))
time = datetime.datetime.now()
sys.stdout.write('\n')
# Remove the first row which is full of random nrs.
quals = quals[1:, :]
max_inds = []
spikenrs = np.array([spikearr.sum() for spikearr in run_spiketimes])
for i in range(clusters.shape[0]):
with warnings.catch_warnings():
warnings.filterwarnings('ignore', '.*true_divide*.')
stas[i] = stas[i] / spikenrs[i]
# Find the pixel with largest absolute value
max_i = np.squeeze(
np.where(np.abs(stas[i]) == np.max(np.abs(stas[i]))))
# If there are multiple pixels with largest value,
# take the first one.
if max_i.shape != (3, ):
try:
max_i = max_i[:, 0]
# If max_i cannot be found just set it to zeros.
except IndexError:
max_i = np.array([0, 0, 0])
max_inds.append(max_i)
print(f'Completed. Total elapsed time: {msc.timediff(startime)}\n' +
f'Finished on {datetime.datetime.now().strftime("%A %X")}')
savepath = os.path.join(exp_dir, 'data_analysis', stimname)
if not os.path.isdir(savepath):
os.makedirs(savepath, exist_ok=True)
savepath = os.path.join(savepath, savefname)
keystosave = [
'clusters', 'frametimings', 'mask', 'repeated_spiketimes',
'run_spiketimes', 'frame_duration', 'max_inds', 'nblinks', 'stas',
'stx_h', 'stx_w', 'total_run_frames', 'sx', 'sy', 'filter_length',
'stimname', 'exp_name', 'spikenrs', 'clusterstoanalyze',
'frametimingsfraction', 'cutoff', 'quals', 'nrofchunks', 'chunklength'
]
datadict = {}
for key in keystosave:
datadict[key] = locals()[key]
np.savez(savepath, **datadict)
t = (np.arange(nrofchunks) * chunklength * frame_duration) / refresh_rate
qmax = np.max(quals, axis=0)
qualsn = quals / qmax[np.newaxis, :]
ax = plt.subplot(111)
ax.plot(t, qualsn, alpha=0.3)
plt.ylabel('Z-score of center pixel (normalized)')
plt.xlabel('Minutes of stimulus analyzed')
plt.ylim([0, 1])
plf.spineless(ax, 'tr')
plt.title(f'Recording duration optimization\n{exp_name}\n {savefname}')
plt.savefig(savepath + '.svg', format='svg')
plt.close()
b = np.abs(stas).max(axis=1)
stas_normalized = stas / b.repeat(stas.shape[1]).reshape(stas.shape)
return stas_normalized
#%%
exp_name = '20180802'
stim_nr = 1
data = iof.load(exp_name, stim_nr)
stimulus = glm.loadstim(exp_name, stim_nr)
clusters = data['clusters']
#%%
#stas = np.array(data['stas'])
#stas_normalized = np.abs(stas).max(axis=1)
#stas_normalized = a / stas_normalized.repeat(stas.shape[1]).reshape(stas.shape)
frametimes = asc.ft_nblinks(exp_name, stim_nr)[1]
#stas = normalizestas(data['stas'])
stas = np.array(data['stas'])
predstas = np.zeros(stas.shape)
predmus = np.zeros(stas.shape[0])
start = dt.datetime.now()
allspikes = np.zeros((stas.shape[0], frametimes.shape[0]), dtype=np.int8)
for i, cluster in enumerate(clusters):
#cluster = data['clusters'][i]
sta = data['stas'][i]
def OMBanalyzer(exp_name, stimnr, plotall=False, nr_bins=20):
"""
Analyze responses to object moving background stimulus. STA and STC
are calculated.
Note that there are additional functions that make use of the
OMB class. This function was written before the OMB class existed
"""
# TODO
# Add iteration over multiple stimuli
exp_dir = iof.exp_dir_fixer(exp_name)
exp_name = os.path.split(exp_dir)[-1]
stimname = iof.getstimname(exp_dir, stimnr)
parameters = asc.read_parameters(exp_name, stimnr)
assert parameters['stimulus_type'] == 'objectsmovingbackground'
stimframes = parameters.get('stimFrames', 108000)
preframes = parameters.get('preFrames', 200)
nblinks = parameters.get('Nblinks', 2)
seed = parameters.get('seed', -10000)
seed2 = parameters.get('objseed', -1000)
stepsize = parameters.get('stepsize', 2)
ntotal = int(stimframes / nblinks)
clusters, metadata = asc.read_spikesheet(exp_name)
refresh_rate = metadata['refresh_rate']
filter_length, frametimings = asc.ft_nblinks(exp_name, stimnr, nblinks,
refresh_rate)
frame_duration = np.ediff1d(frametimings).mean()
frametimings = frametimings[:-1]
if ntotal != frametimings.shape[0]:
print(f'For {exp_name}\nstimulus {stimname} :\n'
f'Number of frames specified in the parameters file ({ntotal}'
f' frames) and frametimings ({frametimings.shape[0]}) do not'
' agree!'
' The stimulus was possibly interrupted during recording.'
' ntotal is changed to match actual frametimings.')
ntotal = frametimings.shape[0]
# Generate the numbers to be used for reconstructing the motion
# ObjectsMovingBackground.cpp line 174, steps are generated in an
# alternating fashion. We can generate all of the numbers at once
# (total lengths is defined by stimFrames) and then assign
# to x and y directions. Although there is more
# stuff around line 538
randnrs, seed = randpy.gasdev(seed, ntotal * 2)
randnrs = np.array(randnrs) * stepsize
xsteps = randnrs[::2]
ysteps = randnrs[1::2]
clusterids = plf.clusters_to_ids(clusters)
all_spikes = np.empty((clusters.shape[0], ntotal))
for i, (cluster, channel, _) in enumerate(clusters):
spiketimes = asc.read_raster(exp_name, stimnr, cluster, channel)
spikes = asc.binspikes(spiketimes, frametimings)
all_spikes[i, :] = spikes
# Collect STA for x and y movement in one array
stas = np.zeros((clusters.shape[0], 2, filter_length))
stc_x = np.zeros((clusters.shape[0], filter_length, filter_length))
stc_y = np.zeros((clusters.shape[0], filter_length, filter_length))
t = np.arange(filter_length) * 1000 / refresh_rate * nblinks
for k in range(filter_length, ntotal - filter_length + 1):
x_mini = xsteps[k - filter_length + 1:k + 1][::-1]
y_mini = ysteps[k - filter_length + 1:k + 1][::-1]
for i, (cluster, channel, _) in enumerate(clusters):
if all_spikes[i, k] != 0:
stas[i, 0, :] += all_spikes[i, k] * x_mini
stas[i, 1, :] += all_spikes[i, k] * y_mini
# Calculate non-centered STC (Cantrell et al., 2010)
stc_x[i, :, :] += all_spikes[i, k] * calc_covar(x_mini)
stc_y[i, :, :] += all_spikes[i, k] * calc_covar(y_mini)
eigvals_x = np.zeros((clusters.shape[0], filter_length))
eigvals_y = np.zeros((clusters.shape[0], filter_length))
eigvecs_x = np.zeros((clusters.shape[0], filter_length, filter_length))
eigvecs_y = np.zeros((clusters.shape[0], filter_length, filter_length))
bins_x = np.zeros((clusters.shape[0], nr_bins))
bins_y = np.zeros((clusters.shape[0], nr_bins))
spikecount_x = np.zeros(bins_x.shape)
spikecount_y = np.zeros(bins_x.shape)
generators_x = np.zeros(all_spikes.shape)
generators_y = np.zeros(all_spikes.shape)
# Normalize STAs and STCs with respect to spike numbers
for i in range(clusters.shape[0]):
totalspikes = all_spikes.sum(axis=1)[i]
stas[i, :, :] = stas[i, :, :] / totalspikes
stc_x[i, :, :] = stc_x[i, :, :] / totalspikes
stc_y[i, :, :] = stc_y[i, :, :] / totalspikes
try:
eigvals_x[i, :], eigvecs_x[i, :, :] = np.linalg.eigh(
stc_x[i, :, :])
eigvals_y[i, :], eigvecs_y[i, :, :] = np.linalg.eigh(
stc_y[i, :, :])
except np.linalg.LinAlgError:
continue
# Calculate the generator signals and nonlinearities
generators_x[i, :] = np.convolve(eigvecs_x[i, :, -1],
xsteps,
mode='full')[:-filter_length + 1]
generators_y[i, :] = np.convolve(eigvecs_y[i, :, -1],
ysteps,
mode='full')[:-filter_length + 1]
spikecount_x[i, :], bins_x[i, :] = nlt.calc_nonlin(
all_spikes[i, :], generators_x[i, :], nr_bins)
spikecount_y[i, :], bins_y[i, :] = nlt.calc_nonlin(
all_spikes[i, :], generators_y[i, :], nr_bins)
savepath = os.path.join(exp_dir, 'data_analysis', stimname)
if not os.path.isdir(savepath):
os.makedirs(savepath, exist_ok=True)
# Calculated based on last eigenvector
magx = eigvecs_x[:, :, -1].sum(axis=1)
magy = eigvecs_y[:, :, -1].sum(axis=1)
r_ = np.sqrt(magx**2 + magy**2)
theta_ = np.arctan2(magy, magx)
# To draw the vectors starting from origin, insert zeros every other element
r = np.zeros(r_.shape[0] * 2)
theta = np.zeros(theta_.shape[0] * 2)
r[1::2] = r_
theta[1::2] = theta_
plt.polar(theta, r)
plt.gca().set_xticks(np.pi / 180 * np.array([0, 90, 180, 270]))
plt.title(f'Population plot for motion STAs\n{exp_name}')
plt.savefig(os.path.join(savepath, 'population.svg'))
if plotall:
plt.show()
plt.close()
for i in range(stas.shape[0]):
stax = stas[i, 0, :]
stay = stas[i, 1, :]
ax1 = plt.subplot(211)
ax1.plot(t, stax, label=r'STA$_{\rm X}$')
ax1.plot(t, stay, label=r'STA$_{\rm Y}$')
ax1.plot(t, eigvecs_x[i, :, -1], label='Eigenvector_X 0')
ax1.plot(t, eigvecs_y[i, :, -1], label='Eigenvector_Y 0')
plt.legend(fontsize='x-small')
ax2 = plt.subplot(4, 4, 9)
ax3 = plt.subplot(4, 4, 13)
ax2.set_yticks([])
ax2.set_xticklabels([])
ax3.set_yticks([])
ax2.set_title('Eigenvalues', size='small')
ax2.plot(eigvals_x[i, :],
'o',
markerfacecolor='C0',
markersize=4,
markeredgewidth=0)
ax3.plot(eigvals_y[i, :],
'o',
markerfacecolor='C1',
markersize=4,
markeredgewidth=0)
ax4 = plt.subplot(2, 3, 5)
ax4.plot(bins_x[i, :], spikecount_x[i, :] / frame_duration)
ax4.plot(bins_y[i, :], spikecount_y[i, :] / frame_duration)
ax4.set_ylabel('Firing rate [Hz]')
ax4.set_title('Nonlinearities', size='small')
plf.spineless([ax1, ax2, ax3, ax4], 'tr')
ax5 = plt.subplot(2, 3, 6, projection='polar')
ax5.plot(theta, r, color='k', alpha=.3)
ax5.plot(theta[2 * i:2 * i + 2], r[2 * i:2 * i + 2], lw=3)
ax5.set_xticklabels(['0', '', '', '', '180', '', '270', ''])
ax5.set_title('Vector sum of X and Y STCs', size='small')
plt.suptitle(f'{exp_name}\n{stimname}\n{clusterids[i]}')
plt.subplots_adjust(hspace=.4)
plt.savefig(os.path.join(savepath, clusterids[i] + '.svg'),
bbox_inches='tight')
if plotall:
plt.show()
plt.close()
keystosave = [
'nblinks', 'all_spikes', 'clusters', 'frame_duration', 'eigvals_x',
'eigvals_y', 'eigvecs_x', 'eigvecs_y', 'filter_length', 'magx', 'magy',
'ntotal', 'r', 'theta', 'stas', 'stc_x', 'stc_y', 'bins_x', 'bins_y',
'nr_bins', 'spikecount_x', 'spikecount_y', 'generators_x',
'generators_y', 't'
]
datadict = {}
for key in keystosave:
datadict[key] = locals()[key]
npzfpath = os.path.join(savepath, str(stimnr) + '_data')
np.savez(npzfpath, **datadict)
def loadstim(exp, stim_nr, maxframenr=10000):
"""
Recreate the stimulus based on the seed for a given stimulus type.
Each type of stimulus requires a different way of handling the
random numbers from the PRNG.
"""
sortedstim = asc.stimulisorter(exp)
clusters, metadata = asc.read_spikesheet(exp)
pars = asc.read_parameters(exp, stim_nr)
for key, val in sortedstim.items():
if stim_nr in val:
stimtype = key
if stimtype in ['fff', 'stripeflicker', 'checkerflicker', 'frozennoise']:
seed = pars.get('seed', -10000)
bw = pars.get('blackwhite', False)
filter_length, frametimings = asc.ft_nblinks(exp, stim_nr)
total_frames = frametimings.shape[0]
if stimtype == 'fff':
if bw:
randnrs, seed = randpy.ranb(seed, total_frames)
# Since ranb returns zeros and ones, we need to convert
# the zeros into -1s.
stimulus = np.array(randnrs) * 2 - 1
else:
randnrs, seed = randpy.gasdev(seed, total_frames)
stimulus = np.array(randnrs)
elif stimtype in ['checkerflicker', 'frozennoise']:
scr_width = metadata['screen_width']
scr_height = metadata['screen_height']
stx_h = pars['stixelheight']
stx_w = pars['stixelwidth']
# Check whether any parameters are given for margins, calculate
# screen dimensions.
marginkeys = ['tmargin', 'bmargin', 'rmargin', 'lmargin']
margins = []
for key in marginkeys:
margins.append(pars.get(key, 0))
# Subtract bottom and top from vertical dimension; left and right
# from horizontal dimension
scr_width = scr_width - sum(margins[2:])
scr_height = scr_height - sum(margins[:2])
sx, sy = scr_height / stx_h, scr_width / stx_w
# Make sure that the number of stimulus pixels are integers
# Rounding down is also possible but might require
# other considerations.
if sx % 1 == 0 and sy % 1 == 0:
sx, sy = int(sx), int(sy)
else:
raise ValueError('sx and sy must be integers')
# HINT: fixing stimulus length for now because of memory
# capacity
total_frames = maxframenr
randnrs, seed = randpy.ranb(seed, sx * sy * total_frames)
# Reshape and change 0's to -1's
stimulus = np.reshape(randnrs,
(sx, sy, total_frames), order='F') * 2 - 1
return stimulus
if stimtype == 'OMB':
stimframes = pars.get('stimFrames', 108000)
preframes = pars.get('preFrames', 200)
nblinks = pars.get('Nblinks', 2)
seed = pars.get('seed', -10000)
seed2 = pars.get('objseed', -1000)
stepsize = pars.get('stepsize', 2)
ntotal = int(stimframes / nblinks)
clusters, metadata = asc.read_spikesheet(exp)
refresh_rate = metadata['refresh_rate']
filter_length, frametimings = asc.ft_nblinks(exp, stim_nr, nblinks,
refresh_rate)
frame_duration = np.ediff1d(frametimings).mean()
frametimings = frametimings[:-1]
if ntotal != frametimings.shape[0]:
print(
f'For {exp}\nstimulus {iof.getstimname(exp, stim_nr)} :\n'
f'Number of frames specified in the parameters file ({ntotal}'
f' frames) and frametimings ({frametimings.shape[0]}) do not'
' agree!'
' The stimulus was possibly interrupted during recording.'
' ntotal is changed to match actual frametimings.')
ntotal = frametimings.shape[0]
randnrs, seed = randpy.gasdev(seed, ntotal * 2)
randnrs = np.array(randnrs) * stepsize
xsteps = randnrs[::2]
ysteps = randnrs[1::2]
return np.vstack((xsteps, ysteps))
return None
data = iof.load(exp_name, stim_nr)
stimulus = glm.loadstim(exp_name, stim_nr)
cell_lim = slice(None)
clusters = data['clusters'][cell_lim]
stas = np.array(data['stas'])
#stas = glm.normalizestas(data['stas'][cell_lim])
#frame_dur = data['frame_duration']
predstas = stas.copy()
predmus = np.zeros((stas.shape[0], stas.shape[-1]))
parameters = asc.read_parameters(exp_name, stim_nr)
_, frametimes = asc.ft_nblinks(exp_name, stim_nr, parameters.get('Nblinks', 2))
frametimes = frametimes[:-1]
frame_dur = np.ediff1d(frametimes).mean()
stashape = stas[:, 0, :].shape
#%%
start = dt.datetime.now()
for i, cluster in enumerate(clusters):
for j, direction in enumerate(['x', 'y']):
spikes = asc.read_raster(exp_name, stim_nr, cluster[0], cluster[1])
spikes = asc.binspikes(spikes, frametimes)
res = glm.minimize_loglhd(stas[i, j, :],
0,
stimulus[j, :],
frame_dur,
def fffanalyzer(exp_name, stimnrs):
"""
Analyzes and plots data from full field flicker
stimulus.
"""
exp_dir = iof.exp_dir_fixer(exp_name)
exp_name = os.path.split(exp_dir)[-1]
if isinstance(stimnrs, int):
stimnrs = [stimnrs]
for stimnr in stimnrs:
stimnr = str(stimnr)
stimname = iof.getstimname(exp_name, stimnr)
clusters, metadata = asc.read_spikesheet(exp_dir)
parameters = asc.read_parameters(exp_dir, stimnr)
clusterids = plf.clusters_to_ids(clusters)
refresh_rate = metadata['refresh_rate']
if parameters['stixelheight'] < 600 or parameters['stixelwidth'] < 800:
raise ValueError('Make sure the stimulus is full field flicker.')
nblinks = parameters['Nblinks']
bw = parameters.get('blackwhite', False)
seed = parameters.get('seed', -10000)
filter_length, frametimings = asc.ft_nblinks(exp_dir, stimnr)
frame_duration = np.average(np.ediff1d(frametimings))
total_frames = frametimings.shape[0]
all_spiketimes = []
# Store spike triggered averages in a list containing correct shaped
# arrays
stas = []
# Make a list for covariances of the spike triggered ensemble
covars = []
for i in range(len(clusters[:, 0])):
spiketimes = asc.read_raster(exp_dir, stimnr,
clusters[i, 0], clusters[i, 1])
spikes = asc.binspikes(spiketimes, frametimings)
all_spiketimes.append(spikes)
stas.append(np.zeros(filter_length))
covars.append(np.zeros((filter_length, filter_length)))
if bw:
randnrs, seed = randpy.ranb(seed, total_frames)
# Since ranb returns zeros and ones, we need to convert the zeros
# into -1s.
stimulus = np.array(randnrs) * 2 - 1
else:
randnrs, seed = randpy.gasdev(seed, total_frames)
stimulus = np.array(randnrs)
for k in range(filter_length, total_frames-filter_length+1):
stim_small = stimulus[k-filter_length+1:k+1][::-1]
for j in range(clusters.shape[0]):
spikes = all_spiketimes[j]
if spikes[k] != 0:
stas[j] += spikes[k]*stim_small
# This trick is needed to use .T for tranposing
stim_small_n = stim_small[np.newaxis, :]
# Calculate the covariance as the weighted outer product
# of small stimulus(i.e. snippet) with itself
# This is non-centered STC (a la Cantrell et al., 2010)
covars[j] += spikes[k]*(np.dot(stim_small_n.T,
stim_small_n))
spikenrs = np.array([spikearr.sum() for spikearr in all_spiketimes])
plotpath = os.path.join(exp_dir, 'data_analysis',
stimname, 'filters')
if not os.path.isdir(plotpath):
os.makedirs(plotpath, exist_ok=True)
t = np.arange(filter_length)*frame_duration*1000
eigvals = [np.zeros((filter_length)) for i in range(clusters.shape[0])]
eigvecs = [np.zeros((filter_length,
filter_length)) for i in range(clusters.shape[0])]
for i in range(clusters.shape[0]):
stas[i] = stas[i]/spikenrs[i]
covars[i] = covars[i]/spikenrs[i]
try:
eigvals[i], eigvecs[i] = np.linalg.eigh(covars[i])
except np.linalg.LinAlgError:
eigvals[i] = np.full((filter_length), np.nan)
eigvecs[i] = np.full((filter_length, filter_length), np.nan)
fig = plt.figure(figsize=(9, 6))
ax = plt.subplot(111)
ax.plot(t, stas[i], label='STA')
ax.plot(t, eigvecs[i][:, 0], label='STC component 1', alpha=.5)
ax.plot(t, eigvecs[i][:, -1], label='STC component 2', alpha=.5)
# Add eigenvalues as inset
ax2 = fig.add_axes([.65, .15, .2, .2])
# Highlight the first and second components which are plotted
ax2.plot(0, eigvals[i][0], 'o',
markersize=7, markerfacecolor='C1', markeredgewidth=0)
ax2.plot(filter_length-1, eigvals[i][-1], 'o',
markersize=7, markerfacecolor='C2', markeredgewidth=0)
ax2.plot(eigvals[i], 'ko', alpha=.5, markersize=4,
markeredgewidth=0)
ax2.set_axis_off()
plf.spineless(ax)
ax.set_xlabel('Time[ms]')
ax.set_title(f'{exp_name}\n{stimname}\n{clusterids[i]} Rating:'
f' {clusters[i, 2]} {int(spikenrs[i])} spikes')
plt.savefig(os.path.join(plotpath, clusterids[i])+'.svg',
format='svg', dpi=300)
plt.close()
savepath = os.path.join(os.path.split(plotpath)[0], stimnr+'_data')
keystosave = ['stas', 'clusters', 'frame_duration', 'all_spiketimes',
'stimname', 'total_frames', 'spikenrs', 'bw', 'nblinks',
'filter_length', 'exp_name', 'covars', 'eigvals',
'eigvecs']
data_in_dict = {}
for key in keystosave:
data_in_dict[key] = locals()[key]
np.savez(savepath, **data_in_dict)
print(f'Analysis of {stimname} completed.')
if bgnoise != 4:
raise NotImplementedError('Only gaussian correlated binary '
'noise is implemented.')
bgcontrast = pars.get('bgcontrast', 0.3)
bggenerationseed = -10000
filterstd = pars.get('filterstdv', bgstixel)
meanintensity = pars.get('meanintensity', 0.5)
contrast = pars.get('contrast', 1)
squareheight, squarewidth = (800, 800)
ntotal = int(stimframes / nblinks)
refresh_rate = metadata['refresh_rate']
_, frametimings = asc.ft_nblinks(exp, stimnr, nblinks,
refresh_rate)
frame_duration = np.ediff1d(frametimings).mean()
frametimings = frametimings[:-1]
if ntotal != frametimings.shape[0]:
print(f'For {exp}\nstimulus {iof.getstimname(exp, stimnr)} :\n'
f'Number of frames specified in the parameters file ({ntotal}'
f' frames) and frametimings ({frametimings.shape[0]}) do not'
' agree!'
' The stimulus was possibly interrupted during recording.'
' ntotal is changed to match actual frametimings.')
ntotal = frametimings.shape[0]
randnrs, seed = randpy.gasdev(seed, ntotal*2)
randnrs = np.array(randnrs)*stepsize
xsteps = randnrs[::2]
def stripeflickeranalysis(exp_name, stim_nrs):
exp_dir = iof.exp_dir_fixer(exp_name)
if isinstance(stim_nrs, int):
stim_nrs = [stim_nrs]
elif len(stim_nrs) == 0:
return
for stim_nr in stim_nrs:
stimname = iof.getstimname(exp_name, stim_nr)
clusters, metadata = asc.read_spikesheet(exp_dir)
parameters = asc.read_parameters(exp_dir, stim_nr)
scr_width = metadata['screen_width']
px_size = metadata['pixel_size(um)']
refresh_rate = metadata['refresh_rate']
stx_w = parameters['stixelwidth']
stx_h = parameters['stixelheight']
if (stx_h / stx_w) < 2:
raise ValueError('Make sure the stimulus is stripeflicker.')
sy = scr_width / stx_w
if sy % 1 == 0:
sy = int(sy)
else:
raise ValueError('sy is not an integer')
nblinks = parameters['Nblinks']
bw = parameters.get('blackwhite', False)
seed = parameters.get('seed', -10000)
filter_length, frametimings = asc.ft_nblinks(exp_dir, stim_nr)
# Omit everything that happens before the first 10 seconds
cut_time = 10
frame_duration = np.average(np.ediff1d(frametimings))
total_frames = frametimings.shape[0]
all_spiketimes = []
# Store spike triggered averages in a list containing correct
# shaped arrays
stas = []
for i in range(len(clusters[:, 0])):
spiketimes = asc.read_raster(exp_dir, stim_nr, clusters[i, 0],
clusters[i, 1])
spikes = asc.binspikes(spiketimes, frametimings)
all_spiketimes.append(spikes)
stas.append(np.zeros((sy, filter_length)))
# Add one more element to correct for random noise
clusters = np.vstack((clusters, [0, 0, 0]))
all_spiketimes.append(np.ones(frametimings.shape, dtype=int))
stas.append(np.zeros((sy, filter_length)))
if bw:
randnrs, seed = randpy.ranb(seed, sy * total_frames)
else:
randnrs, seed = randpy.gasdev(seed, sy * total_frames)
stimulus = np.reshape(randnrs, (sy, total_frames), order='F')
if bw:
# Since ranb returns zeros and ones, we need to convert the zeros
# into -1s.
stimulus = stimulus * 2 - 1
del randnrs
for k in range(filter_length, total_frames - filter_length + 1):
stim_small = stimulus[:, k - filter_length + 1:k + 1][:, ::-1]
for j in range(clusters.shape[0]):
spikes = all_spiketimes[j]
if spikes[k] != 0 and frametimings[k] > cut_time:
stas[j] += spikes[k] * stim_small
max_inds = []
spikenrs = np.array([spikearr.sum() for spikearr in all_spiketimes])
quals = np.array([])
# Remove the random noise correction element from clusters
correction = stas.pop() / spikenrs[-1]
clusters = clusters[:-1, :]
all_spiketimes.pop()
spikenrs = spikenrs[:-1]
for i in range(clusters.shape[0]):
stas[i] = stas[i] / spikenrs[i]
stas[i] = stas[i] - correction
# Find the pixel with largest absolute value
max_i = np.squeeze(
np.where(np.abs(stas[i]) == np.max(np.abs(stas[i]))))
# If there are multiple pixels with largest value,
# take the first one.
if max_i.shape != (2, ):
try:
max_i = max_i[:, 0]
# If max_i cannot be found just set it to zeros.
except IndexError:
max_i = np.array([0, 0])
# In case of spike numbers being zero, all elements are NaN
# imshow and savefig do not play nice with NaN so set all to zero
if np.all(np.isnan(stas[i])):
stas[i] = np.zeros(stas[i].shape)
max_inds.append(max_i)
quals = np.append(quals, asc.staquality(stas[i]))
savefname = str(stim_nr) + '_data'
savepath = pjoin(exp_dir, 'data_analysis', stimname)
exp_name = os.path.split(exp_dir)[-1]
if not os.path.isdir(savepath):
os.makedirs(savepath, exist_ok=True)
savepath = os.path.join(savepath, savefname)
keystosave = [
'stas', 'max_inds', 'clusters', 'sy', 'correction',
'frame_duration', 'all_spiketimes', 'stimname', 'total_frames',
'stx_w', 'spikenrs', 'bw', 'quals', 'nblinks', 'filter_length',
'exp_name'
]
data_in_dict = {}
for key in keystosave:
data_in_dict[key] = locals()[key]
np.savez(savepath, **data_in_dict)
print(f'Analysis of {stimname} completed.')
