def retrieve_plot_data(self):
'''retrieve default data from spec data file'''
'''
data parser for 2-D mesh and hklmesh
'''
label1, start1, end1, intervals1, label2, start2, end2, intervals2, time = self.scan.scanCmd.split()[1:]
if label1 not in self.scan.data:
label1 = self.scan.L[0] # mnemonic v. name
if label2 not in self.scan.data:
label2 = self.scan.L[1] # mnemonic v. name
axis1 = self.scan.data.get(label1)
axis2 = self.scan.data.get(label2)
intervals1, intervals2 = map(int, (intervals1, intervals2))
start1, end1, start2, end2, time = map(float, (start1, end1, start2, end2, time))
if len(axis1) < intervals1 and min(axis2) == max(axis2):
# stopped scan before second row started, 1-D plot is better (issue #82)
self.axes = [label1,]
self.signal = self.scan.column_last
self.data[label1] = self.scan.data[label1]
self.data[self.signal] = self.scan.data[self.signal]
return
axis1 = axis1[0:intervals1+1]
self.data[label1] = axis1 # 1-D array
axis2 = [axis2[row] for row in range(len(axis2)) if row % (intervals1+1) == 0]
self.data[label2] = axis2 # 1-D array
column_labels = self.scan.L
column_labels.remove(label1) # special handling
column_labels.remove(label2) # special handling
if self.scan.scanCmd.startswith('hkl'):
# find the reciprocal space axis held constant
label3 = [key for key in ('H', 'K', 'L') if key in column_labels][0]
self.data[label3] = self.scan.data.get(label3)[0] # constant
# build 2-D data objects (do not build label1, label2, [or label3] as 2-D objects)
data_shape = [len(axis2), len(axis1)]
for label in column_labels:
if label not in self.data:
axis = numpy.array( self.scan.data.get(label) )
self.data[label] = utils.reshape_data(axis, data_shape)
else:
pass
self.signal = utils.clean_name(self.scan.column_last)
self.axes = [label1, label2]
if spec.MCA_DATA_KEY in self.scan.data: # 3-D array(s)
# save each spectrum
for key, spectrum in sorted(self.scan.data[spec.MCA_DATA_KEY].items()):
num_channels = len(spectrum[0])
data_shape.append(num_channels)
mca = numpy.array(spectrum)
data = utils.reshape_data(mca, data_shape)
channels = range(1, num_channels+1)
ds_name = '_' + key + '_'
self.data[ds_name] = data
self.data[ds_name+'channel_'] = channels
def mesh(self, nxdata, scan):
'''*internal*: data parser for 2-D mesh and hklmesh'''
# 2-D parser: http://www.certif.com/spec_help/mesh.html
# mesh motor1 start1 end1 intervals1 motor2 start2 end2 intervals2 time
# 2-D parser: http://www.certif.com/spec_help/hklmesh.html
# hklmesh Q1 start1 end1 intervals1 Q2 start2 end2 intervals2 time
# mesh: data/33id_spec.dat scan 22
# hklmesh: data/33bm_spec.dat scan 17
signal, axes = '', ['',]
label1, start1, end1, intervals1, label2, start2, end2, intervals2, time = scan.scanCmd.split()[1:]
if label1 not in scan.data:
label1 = scan.L[0] # mnemonic v. name
if label2 not in scan.data:
label2 = scan.L[1] # mnemonic v. name
axis1 = scan.data.get(label1)
axis2 = scan.data.get(label2)
intervals1, intervals2 = map(int, (intervals1, intervals2))
start1, end1, start2, end2, time = map(float, (start1, end1, start2, end2, time))
if len(axis1) < intervals1: # stopped scan before second row started
signal, axes = self.oneD(nxdata, scan) # fallback support
else:
axis1 = axis1[0:intervals1+1]
axis2 = [axis2[row] for row in range(len(axis2)) if row % (intervals1+1) == 0]
column_labels = scan.L
column_labels.remove(label1) # special handling
column_labels.remove(label2) # special handling
if scan.scanCmd.startswith('hkl'):
# find the reciprocal space axis held constant
label3 = [key for key in ('H', 'K', 'L') if key not in (label1, label2)][0]
axis3 = scan.data.get(label3)[0]
self.write_ds(nxdata, label3, axis3)
self.write_ds(nxdata, label1, axis1) # 1-D array
self.write_ds(nxdata, label2, axis2) # 1-D array
# build 2-D data objects (do not build label1, label2, [or label3] as 2-D objects)
data_shape = [len(axis1), len(axis2)]
for label in column_labels:
if label not in nxdata:
axis = np.array( scan.data.get(label) )
self.write_ds(nxdata, label, utils.reshape_data(axis, data_shape))
else:
pass
signal = utils.clean_name(scan.column_last)
axes = ':'.join([label1, label2])
if '_mca_' in scan.data: # 3-D array(s)
# save each spectrum
for key, spectrum in sorted(scan.data['_mca_'].items()):
num_channels = len(spectrum[0])
data_shape.append(num_channels)
mca = np.array(spectrum)
data = utils.reshape_data(mca, data_shape)
channels = range(1, num_channels+1)
ds_name = '_' + key + '_'
self.write_ds(nxdata, ds_name, data, axes=axes+':'+ds_name+'channel_', units='counts')
self.write_ds(nxdata, ds_name+'channel_', channels, units='channel')
return signal, axes
def __init__(self, imagename, psfname=None, sourcefinder_name='pybdsm',
makeplots=True, do_psf_corr=True, do_local_var=True,
psf_corr_region=2, local_var_region=10, rel_excl_src=None,
pos_smooth=1.6, neg_smooth=1.6, loglevel=0, thresh_pix=5,
thresh_isl=3, neg_thresh_isl=3, neg_thresh_pix=5,
prefix=None, do_nearsources=False, **kw):
""" Takes in image and extracts sources and makes
reliability estimations..
imagename: Fits image
psfname: PSF fits image, optional.
sourcefinder_name: str, optional. Default 'pybdsm'.
Uses source finder specified by the users.
makeplots: bool, optional. Default is True.
Make reliability plots.
do_psf_corr : bool, optional. Default True.
If True, correlation of sources with PSF will be added
as an extra source parameter in reliability estimation.
But the PSF fits image must be provided.
do_local_var : bool, optional. Default is True.
Adds local variance as an extra source parameter,
similar to do_psf_corr but independent of the PSF image.
psf_corr_region : int, optional. Default value is 2.
Data size to correlate around a source in beam sizes.
local_var_region: int, optional. Default 10.
Data size to compute the local variance in beam sizes.
rel_excl_src : float numbers, optional. Default is None.
Excludes sources in this region from the reliability
estimations, e.g ra, dec, radius in degrees. For
many regions: ra1, dec1, radius1: ra2, dec2, radius2.
pos_smooth : float, optional. Default 1.6
Data smoothing threshold in the positive side of an image.
For default value 1.6, data peaks < 1.6 * image noise
will be averaged out.
neg_smooth : float, optional. Default 1.6.
Similar to pos_smooth but applied to the negative side of
an image.
loglevel : int, optional. Default is 0.
Provides Pythonlogging options, 0, 1, 2 and 3 for info, debug,
error and critial respectively.
thresh_isl : float, optional. Default is 3.
Threshold for the island boundary in number of sigma above
the mean. Determines extent of island used for fitting
[pybdsm]. For positive pixels.
thresh_pix : float, optional. Default is 5.
Source detection threshold: threshold for the island
peak in number of sigma above the mean. For positive pixels.
neg_thresh_isl : float, optional. Default is 3.
Simialr to thresh_isl but applied to negative side
of the image.
neg_thresh_pix : float, optional. Default is 5.
Similar to thresh_pix but applied to the negative
side of an image.
do_nearsources: boolean. Default is False.
If true it adds number of nearest neighnours as an extra
parameter. It looks for sources around 5 beam sizes.
kw : kward for source extractions. Should be a mapping e.g
kw['thresh_isl'] = 2.0 or kw['do_polarization'] = True
"""
# image, psf image
self.imagename = imagename
self.psfname = psfname
# setting output file names
self.prefix = prefix
self.poslsm = self.prefix + "_positive.lsm.html"
self.neglsm = self.prefix + "_negative.lsm.html"
# log level
self.loglevel = loglevel
self.log = utils.logger(self.loglevel, prefix=self.prefix)
self.log.info("Loading Image data")
# reading imagename data
self.imagedata, self.wcs, self.header, self.pixelsize =\
utils.reshape_data(self.imagename, prefix=self.prefix)
self.bmaj = numpy.deg2rad(self.header["BMAJ"])
self.do_psf_corr = do_psf_corr
if not self.psfname:
self.log.info("No psf provided, do_psf_corr = False.")
self.do_psf_corr = False
# computing negative noise
self.noise = utils.negative_noise(self.imagedata)
self.log.info("The negative noise is %e"%self.noise)
if self.noise == 0:
self.log.debug("The negative noise is 0, check image")
# source finder initialization
self.sourcefinder_name = sourcefinder_name
self.log.info("Using %s source finder to extract sources."%
self.sourcefinder_name)
# making negative image
self.negativeimage = utils.invert_image(
self.imagename, self.imagedata,
self.header, self.prefix)
# boolean optionals
self.makeplots = makeplots
self.do_local_var = do_local_var
self.nearsources = do_nearsources
# smoothing factors
self.pos_smooth = pos_smooth
self.neg_smooth = neg_smooth
# region to evaluate
self.psf_corr_region = psf_corr_region
self.local_var_region = local_var_region
self.rel_excl_src = rel_excl_src
# Pybdsm or source finder fitting thresholds
self.thresh_isl = thresh_isl
self.thresh_pix = thresh_pix
self.opts_pos = dict(thresh_pix=self.thresh_pix,
thresh_isl=self.thresh_isl)
self.opts_pos.update(kw)
self.opts_neg = {}
self.neg_thresh_isl = neg_thresh_isl
self.neg_thresh_pix = neg_thresh_pix
self.opts_neg["thresh_isl"] = self.neg_thresh_isl
self.opts_neg["thresh_pix"] = self.neg_thresh_pix
def _load_data(self):
"""Load input and output data from text files."""
df, area = load_forcing(self.camels_root, self.basin)
df['QObs(mm/d)'], cal_start = load_discharge(self.camels_root,
self.basin, area)
# determine dates of start and end of period
if self.period == 'n_cal':
end_year = cal_start.year + self.years[-1]
cal_end = pd.to_datetime(f"{end_year}/09/30", yearfirst=True)
if self.period == 'n_cal':
self.means = df[cal_start:cal_end].mean()
self.stds = df[cal_start:cal_end].std()
# Check which period of the time series is requested
if self.period == 'n_cal':
df = df[cal_start:cal_end]
else:
start_year = cal_start.year + self.years[0]
start_date = pd.to_datetime(f"{start_year}/10/01", yearfirst=True)
# check if end date is till end of time series
if self.years[1] < 0:
end_date = df.index[-1]
else:
end_year = cal_start.year + self.years[1]
end_date = pd.to_datetime(f"{end_year}/09/30", yearfirst=True)
df = df[start_date:end_date]
# store first and last date of the selected period
self.period_start = df.index[0]
self.period_end = df.index[-1]
# extract data matrix from data frame
x = np.array([
df['prcp(mm/day)'].values, df['srad(W/m2)'].values,
df['tmax(C)'].values, df['tmin(C)'].values, df['vp(Pa)'].values
]).T
y = np.array([df['QObs(mm/d)'].values]).T
# normalize data, reshape for LSTM training and remove invalid samples
x = self._local_normalization(x, variable='inputs')
x, y = reshape_data(x, y, 365)
if self.period != "n_test":
# Deletes all records, where no discharge was measured (-999)
x = np.delete(x, np.argwhere(y < 0)[:, 0], axis=0)
y = np.delete(y, np.argwhere(y < 0)[:, 0], axis=0)
# Delete all samples, where discharge is NaN
if np.sum(np.isnan(y)) > 0:
print(f"Deleted some records because of NaNs {self.basin}")
x = np.delete(x, np.argwhere(np.isnan(y)), axis=0)
y = np.delete(y, np.argwhere(np.isnan(y)), axis=0)
y = self._local_normalization(y, variable='output')
# convert arrays to torch tensors
x = torch.from_numpy(x.astype(np.float32))
y = torch.from_numpy(y.astype(np.float32))
return x, y
def retrieve_plot_data(self):
'''retrieve default data from spec data file'''
'''
data parser for 2-D mesh and hklmesh
'''
label1, start1, end1, intervals1, label2, start2, end2, intervals2, time = self.scan.scanCmd.split(
)[1:]
if label1 not in self.scan.data:
label1 = self.scan.L[0] # mnemonic v. name
if label2 not in self.scan.data:
label2 = self.scan.L[1] # mnemonic v. name
axis1 = self.scan.data.get(label1)
axis2 = self.scan.data.get(label2)
intervals1, intervals2 = map(int, (intervals1, intervals2))
start1, end1, start2, end2, time = map(
float, (start1, end1, start2, end2, time))
if len(axis1) < intervals1 and min(axis2) == max(axis2):
# stopped scan before second row started, 1-D plot is better (issue #82)
self.axes = [
label1,
]
self.signal = self.scan.column_last
self.data[label1] = self.scan.data[label1]
self.data[self.signal] = self.scan.data[self.signal]
return
axis1 = axis1[0:intervals1 + 1]
self.data[label1] = axis1 # 1-D array
axis2 = [
axis2[row] for row in range(len(axis2))
if row % (intervals1 + 1) == 0
]
self.data[label2] = axis2 # 1-D array
column_labels = self.scan.L
column_labels.remove(label1) # special handling
column_labels.remove(label2) # special handling
if self.scan.scanCmd.startswith('hkl'):
# find the reciprocal space axis held constant
label3 = [key for key in ('H', 'K', 'L')
if key in column_labels][0]
self.data[label3] = self.scan.data.get(label3)[0] # constant
# build 2-D data objects (do not build label1, label2, [or label3] as 2-D objects)
data_shape = [len(axis2), len(axis1)]
for label in column_labels:
if label not in self.data:
axis = numpy.array(self.scan.data.get(label))
self.data[label] = utils.reshape_data(axis, data_shape)
else:
pass
self.signal = utils.clean_name(self.scan.column_last)
self.axes = [label1, label2]
if spec.MCA_DATA_KEY in self.scan.data: # 3-D array(s)
# save each spectrum
for key, spectrum in sorted(
self.scan.data[spec.MCA_DATA_KEY].items()):
num_channels = len(spectrum[0])
data_shape.append(num_channels)
mca = numpy.array(spectrum)
data = utils.reshape_data(mca, data_shape)
channels = range(1, num_channels + 1)
ds_name = '_' + key + '_'
self.data[ds_name] = data
self.data[ds_name + 'channel_'] = channels
def __init__(self, imagename, psfname=None, sourcefinder_name='pybdsm',
makeplots=True, do_psf_corr=True, do_local_var=True,
psf_corr_region=5, local_var_region=10, rel_excl_src=None,
pos_smooth=2, neg_smooth=2, loglevel=0, thresh_pix=5,
thresh_isl=3, neg_thresh_isl=3, neg_thresh_pix=5, reset_rel=None,
prefix=None, do_nearsources=False, savefits=False,
increase_beam_cluster=False, savemask_pos=False, savemask_neg=False,
**kw):
""" Takes in image and extracts sources and makes
reliability estimations..
imagename: Fits image
psfname: PSF fits image, optional.
sourcefinder_name: str, optional. Default 'pybdsm'.
Uses source finder specified.
makeplots: bool, optional. Default is True.
Make reliability plots.
do_psf_corr : bool, optional. Default True.
If True, PSF correlation will be added
as an extra parameter for density estimations.
NB: the PSF fits image must be provided.
do_local_var : bool, optional. Default is True.
If True, adds local variance as an extra parameter,
for density estimations.
do_nearsources: boolean. Default is False.
If true it adds number of nearest neighnours as an extra
parameter. It looks for sources around 5 beam sizes.
psf_corr_region : int, optional. Default value is 5.
Data size to correlate around a source, in beam sizes.
local_var_region: int, optional. Default 10.
Data size to compute the local variance in beam sizes.
rel_excl_src : floats, optional. Default is None.
Excludes sources in a specified region
e.g ra, dec, radius in degrees. For
2 regions: ra1, dec1, radius1: ra2, dec2, radius2, etc.
pos_smooth : float, optional. Default 2.
Masking threshold for the positive image.
For default value 2, data peaks < 2 * image noise
are masked.
neg_smooth : float, optional. Default 2.
Similar to pos_smooth but applied to the negative image.
thresh_isl : float, optional. Default is 3.
Threshold for forming islands in the positive image
thresh_pix : float, optional. Default is 5.
Threshold for model fitting, in positive image.
neg_thresh_isl : float, optional. Default is 3.
Simialr to thresh_isl but for negative image.
neg_thresh_pix : float, optional. Default is 5.
Similar to thresh_pix but for negative image.
savefits: boolean. Default is False.
If True a negative image is saved.
reset_rel: boolean. Default is False. If true then
sources with correlation < 0.002 and rel >0.60
have their reliabilities set to 0.
increase_beam_cluster: boolean, optional. If True, sources
groupings will be increase by 20% the beam size. If False,
the actual beam size will be used. Default is False.
savemask_pos: boolean, optional. If true the mask applied on
the positive side of an image after smoothing is saved.
savemask_neg: Similar to savemask_pos but for the negative
side of an image.
loglevel : int, optional. Default is 0.
Provides Pythonlogging options, 0, 1, 2 and 3 are for info, debug,
error and critial respectively.
kw : kward for source extractions. Should be a mapping e.g
kw['thresh_isl'] = 2.0 or kw['do_polarization'] = True
"""
self.prefix = prefix
# log level
self.loglevel = loglevel
self.log = utils.logger(self.loglevel, prefix=self.prefix)
# image, psf image
self.imagename = imagename
self.psfname = psfname
# reading imagename data
imagedata, self.wcs, self.header, self.pixelsize =\
utils.reshape_data(self.imagename, prefix=self.prefix)
self.imagedata = numpy.array(imagedata, dtype=numpy.float32)
self.image2by2 = numpy.array(utils.image_data(imagedata, prefix),
dtype=numpy.float32)
self.bmaj = numpy.deg2rad(self.header["BMAJ"])
# boolean optionals
self.makeplots = makeplots
self.do_local_var = do_local_var
self.nearsources = do_nearsources
self.do_psf_corr = do_psf_corr
self.savemaskpos = savemask_pos
self.savemaskneg = savemask_neg
self.savefits = savefits
self.derel = reset_rel
if not self.psfname:
self.log.info(" No psf provided, do_psf_corr is set to False.")
self.do_psf_corr = False
if self.psfname:
psfdata, self.psfhdr = utils.open_psf_image(self.psfname)
self.psfdata = utils.image_data(psfdata, prefix)
# computing negative noise
self.noise, self.mean = utils.negative_noise(self.imagedata, self.prefix) #here is 2X2 data here
self.log.info(" The negative noise is %e Jy/beam"%self.noise)
if self.noise == 0:
self.log.debug(" The negative noise is 0, check image")
# source finder initialization
self.sourcefinder_name = sourcefinder_name
self.log.info(" Using %s source finder to extract the sources."%
self.sourcefinder_name)
self.negimage = self.prefix + "_negative.fits"
negativedata = utils.invert_image(
self.imagename, self.imagedata,
self.header, self.negimage, prefix)
self.negimage2by2 = numpy.array(utils.image_data(negativedata, prefix),
dtype=numpy.float32)
self.negativedata = numpy.array(negativedata, numpy.float32)
# smoothing factors
self.pos_smooth = pos_smooth
self.neg_smooth = neg_smooth
# region to evaluate
self.corrstep = psf_corr_region
self.localstep = local_var_region
self.radiusrm = rel_excl_src
self.do_beam = increase_beam_cluster
beam_pix = int(round(numpy.rad2deg(self.bmaj)/self.pixelsize))
self.locstep = self.localstep * beam_pix
self.cfstep = self.corrstep * beam_pix
self.bmin, self.bpa = self.header["BMIN"], self.header["BPA"]
self.opts_pos = {}
if self.do_beam:
bmaj = self.header["BMAJ"]
self.opts_pos["beam"] = (1.2*bmaj, 1.2*self.bmin, self.bpa)
# Pybdsm or source finder fitting thresholds
self.thresh_isl = thresh_isl
self.thresh_pix = thresh_pix
self.opts_pos = dict(thresh_pix=self.thresh_pix,
thresh_isl=self.thresh_isl)
self.opts_pos.update(kw)
self.opts_neg = {}
self.neg_thresh_isl = neg_thresh_isl
self.neg_thresh_pix = neg_thresh_pix
self.opts_neg["thresh_isl"] = self.neg_thresh_isl
self.opts_neg["thresh_pix"] = self.neg_thresh_pix
def __init__(self, imagename, psfname, poscatalog, negcatalog,
snr_thresh=100, local_thresh=0.6, local_region=10,
psfcorr_region=2, high_corr_thresh=0.5, negdetec_region=10,
negatives_thresh=10, phasecenter_excl_radius=None,
prefix=None, loglevel=0):
""" Determines sources that require direction-dependent (DD)
calibration solutions.
imagename: Fits data
psfname : PSF fits data
poscatalog : Catalog of positive detections.
negcatalog : Catalog of negative detections.
Sources extracted from the negative side
of an image.
snr_thresh : float, optional. Default is 100.
Any source with 100 times the minimum SNR is
considered a high SN source.
local_thresh : float, optional. Default is 0.6.
Sources with local variance greater than
0.6 * negative noise are considered as
sources of high local variance.
local_region : integer, optional. Default is 10.
A region to compute the local variance in
beam sizes.
psfcorr : integer, optional. Default is 2.
Data size to correlate. In beam sizes.
high_corr_thresh : float, optional. Default is 0.5.
Correlation threshold. Sources of high correlation
with the PSF have correlation > the specified.
negdetec_region : float, optional. Default is 10.
Region to lookup for negative detections around
a given source. In beam size.
negative_thresh : float, optional. Default is 6.
Number of negative detections, N, threshold. Sources
with number > N negatives around them are require direction
dependent (DD) calibration solutions.
phasecenter_excl_region : float (in degrees), optional.
A radius from the phase center (in beam sizes) to exclude
in making final DD source selection.
prefix : str, optional. Sets a prefix to the output directory.
loglevel : int, optional. Default 0. Python logging.
0, 1, 2, 3 for info, debug, error and critical respectively.
"""
# image, psf image, positive and negative catalogues
self.imagename = imagename
self.psfname = psfname
self.poscatalog = poscatalog
self.negcatalog = negcatalog
self.loglevel = loglevel
self.prefix = prefix
self.log = utils.logger(self.loglevel, prefix=self.prefix)
# reading the imagename data
self.imagedata, self.wcs, self.header, self.pixsize =\
utils.reshape_data(self.imagename, prefix=self.prefix)
self.log.info("Loading image data")
# computing the noise
self.noise = utils.negative_noise(self.imagedata)
self.log.info("The negative noise of an image is %e"%
self.noise)
# tags
self.snr_tag = "snr"
self.high_local_tag = "high_var"
self.high_corr_tag = "high_corr"
self.dd_tag = "dE"
# thresholds
self.snr_thresh = snr_thresh
self.local_thresh = local_thresh
self.high_corr_thresh = high_corr_thresh
self.negatives_thresh = negatives_thresh
#regions
self.psfcorr_region = psfcorr_region
self.local_region = local_region
self.phasecenter_excl_radius = phasecenter_excl_radius
self.negdetec_region = negdetec_region
# central ra, dec, beam major axes
self. ra0 = numpy.deg2rad(self.header["CRVAL1"])
self.dec0 = numpy.deg2rad(self.header["CRVAL2"])
self.bmaj_deg = self.header['BMAJ'] # in degrees
# Check dimensionality
print('subject %s: fusiform: voxel by TR matrix - shape: ' % sub, masked_fusiform_data.shape)
print('subject %s: fusiform: label list - shape: ' % sub, fusiform_TR_onsets_shifted.shape)
print()
print('subject %s: DFR: voxel by TR matrix - shape: ' % sub, masked_DFR_data.shape)
print('subject %s: DFR: label list - shape: ' % sub, DFR_TR_onsets_shifted.shape)
# feature select based on top voxels from Faces vs Objects contrast
spm_img_file = "spmT_0004.img"
#spm_img_file = "1005_FFA_FvO.img"
top_voxels = find_top_voxels(sub, spm_img_file, mask, 100)
# extract BOLD from non-zero labels (only really matters for the fusiform)
roi_masked_data = masked_fusiform_data[top_voxels]
fusiform_data_masked_reduced_nonzero, fusiform_TR_onsets_shifted_nonzero = reshape_data(fusiform_TR_onsets_shifted,
roi_masked_data, 0)
DFR_data_masked_reduced = masked_DFR_data[top_voxels]
# create average trial types for DFR
DFR_average_trials = create_trial_type_averages(DFR_data_masked_reduced, DFR_onsets)
# run classifier
run_ids = fusiform_TR_onsets_shifted_nonzero[:, 1] - 1
# set up collector arrays
sp = PredefinedSplit(run_ids)
clf_score = np.array([])
C_best = []
high_corr_score = np.array([])
high_incorr_score = np.array([])
