def extract_features (subjs, exclusubjs, mask, maskf, scale, scale_min, scale_max):
#population features
nsubjs = len(subjs)
s = nib.load(subjs[0])
subjsiz = np.prod (s.shape)
stype = s.get_data_dtype()
#loading subject data
data = np.empty([nsubjs, subjsiz], dtype=stype)
#number of voxels > 0 in mask
mask = mask.flatten()
nfeats = np.sum(mask > 0)
#reading each subject and saving the features in a vector
feats = np.empty([nsubjs, nfeats], dtype=stype)
#extracting features from non-excluded subjects
c = 0
for s in subjs:
au.log.debug("Reading " + s)
#check geometries
au.check_has_same_geometry (s, maskf)
#load subject
subj = nib.load(s).get_data().flatten()
#mask data and save it
feats[c,:] = subj[mask > 0]
c += 1
#scaling if asked
dmin = scale_min
dmax = scale_max
if scale:
au.log.info("Scaling data.")
[feats, dmin, dmax] = rescale(feats, scale_min, scale_max)
#extracting features from excluded subjects
exclfeats = []
if exclusubjs:
au.log.info("Processing excluded subjects.")
nexcl = len(exclusubjs)
exclfeats = np.empty([nexcl, nfeats], dtype=stype)
c = 0
for s in exclusubjs:
au.log.debug("Reading " + s)
#check geometries
au.check_has_same_geometry (s, maskf)
#load subject
subj = nib.load(s).get_data().flatten()
#mask data and save it
exclfeats[c,:] = subj[mask > 0]
c += 1
if scale:
[exclfeats, emin, emax] = rescale(exclfeats, scale_min, scale_max, dmin, dmax)
return [feats, exclfeats, dmin, dmax]
def check_data (subsfname, labelsfname, outdir, maskfname, outbase):
au.log.debug('Checking data')
nsubs = au.file_len(subsfname)
nlabs = au.file_len(labelsfname)
if (nsubs != nlabs):
err = 'Not same number of lines in input files\n'
err += labelsfname + ': ' + str(nlabs) + '\n'
err += (subsfname + ': ' + str(nsubs)) + '\n'
raise IOError(err)
if not outdir:
outdir = os.getcwd()
slicesdir = outdir + os.path.sep + au.slices_str()
if not os.path.exists (slicesdir):
os.mkdir (slicesdir)
tempdir = outdir + os.path.sep + au.temp_str()
if not os.path.exists (tempdir):
os.mkdir (tempdir)
au.log.info (slicesdir)
#check all files
subsfile = open(subsfname, 'r')
if maskfname:
for line in subsfile:
fpath = line.strip()
au.check_has_same_geometry (maskfname, fpath)
else:
path1 = ''
path2 = ''
for line in subsfile:
path1 = line.strip()
if not path2:
path2 = line.strip()
continue
else:
au.check_has_same_geometry (path1, path2)
subsfile.close()
mask = nib.load(maskf).get_data()
if globalmask:
mask = mask[gm > 0]
#number of voxels > 0 in mask
mask = mask.flatten()
nfeats = np.sum(mask > 0)
#reading each subject and saving the features in a vector
feats = np.empty([nsubjs, nfeats], dtype=float)
#extracting features from non-excluded subjects
c = 0
for s in subjs:
#check geometries
au.check_has_same_geometry(s, maskf)
#mask data and save it
feats[c, :] = data[c, mask > 0]
c += 1
#scaling if asked
if scale:
[feats, dmin, dmax] = rescale(feats, scale_min, scale_max)
#writing in a text file the scaling values of this training set
f = open(outdir + os.path.sep + ofname + '.scaling_range', 'w')
f.write('#data_min, data_max, range_min, range_max')
f.write('\n')
f.write(
str(dmin) + ',' + str(dmax) + ',' + str(scale_min) + ',' +
str(scale_max))
f.close()
mask = nib.load(maskf).get_data()
if globalmask:
mask = mask[gm > 0]
# number of voxels > 0 in mask
mask = mask.flatten()
nfeats = np.sum(mask > 0)
# reading each subject and saving the features in a vector
feats = np.empty([nsubjs, nfeats], dtype=float)
# extracting features from non-excluded subjects
c = 0
for s in subjs:
# check geometries
au.check_has_same_geometry(s, maskf)
# mask data and save it
feats[c, :] = data[c, mask > 0]
c += 1
# scaling if asked
if scale:
[feats, dmin, dmax] = rescale(feats, scale_min, scale_max)
# writing in a text file the scaling values of this training set
f = open(outdir + os.path.sep + ofname + ".scaling_range", "w")
f.write("#data_min, data_max, range_min, range_max")
f.write("\n")
f.write(str(dmin) + "," + str(dmax) + "," + str(scale_min) + "," + str(scale_max))
f.close()
# extracting features from excluded subjects
