def test_infomax(filename='./data/01-001.fmri.rest.1.volreg.band.pca_reduced.nii.gz',filename_mask='./data/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'):
from Image import Image
from sklearn.decomposition import FastICA
import scipy.io as sio
img = Image(img_filename=filename, mask_filename=filename_mask, fwhm=6.0)
xmat = img.image_mat_in_mask_normalised.T
from InfoMax import InfoMax
infomax = InfoMax(30).fit(xmat)
img.write_decomposition(maps=infomax.components_, filename='test_ica_infomax.nii.gz', normalise=True)
def test_ae(
filename='/Volumes/VAULT/YOAD/all/01-001.fmri.rest.1.volreg.band.nii.gz',
filename_mask='/Volumes/VAULT/YOAD/all/registrations/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'
):
img = Image(img_filename=filename, mask_filename=filename_mask)
xmat = img.image_mat_in_mask_normalised.T
ae = Autoencoder(30, max_iter=200, second_nonlinear=True,
bias=True).fit(xmat)
img.write_decomposition(maps=ae.components_,
filename='test_ae.nii.gz',
normalise=False)
def test_infomax(
filename='./data/01-001.fmri.rest.1.volreg.band.pca_reduced.nii.gz',
filename_mask='./data/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'):
from Image import Image
from sklearn.decomposition import FastICA
import scipy.io as sio
img = Image(img_filename=filename, mask_filename=filename_mask, fwhm=6.0)
xmat = img.image_mat_in_mask_normalised.T
from InfoMax import InfoMax
infomax = InfoMax(30).fit(xmat)
img.write_decomposition(maps=infomax.components_,
filename='test_ica_infomax.nii.gz',
normalise=True)
def test_rica(filename='./data/01-001.fmri.rest.1.volreg.nii.gz', filename_mask='./data/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'):
from Image import Image
import scipy.io as sio
img = Image(img_filename=filename, mask_filename=filename_mask)
xmat = img.image_mat_in_mask_normalised.T
from sklearn.decomposition import PCA, FastICA
pca = PCA(30)
xmat_red = pca.fit_transform(xmat.T)
import matplotlib.pyplot as plt
plt.plot(pca.components_.T)
plt.show()
import pdb; pdb.set_trace()
rica = RICA(30).fit(xmat)
img.write_decomposition(maps=rica.components_, filename='test_rica.nii.gz', normalise=False)
def test_rica(
filename='./data/01-001.fmri.rest.1.volreg.nii.gz',
filename_mask='./data/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'):
from Image import Image
import scipy.io as sio
img = Image(img_filename=filename, mask_filename=filename_mask)
xmat = img.image_mat_in_mask_normalised.T
from sklearn.decomposition import PCA, FastICA
pca = PCA(30)
xmat_red = pca.fit_transform(xmat.T)
import matplotlib.pyplot as plt
plt.plot(pca.components_.T)
plt.show()
import pdb
pdb.set_trace()
rica = RICA(30).fit(xmat)
img.write_decomposition(maps=rica.components_,
filename='test_rica.nii.gz',
normalise=False)
out_autoencoder_group = "{prefix}.autoencoder.group.nii.gz".format(prefix=in_nii__2__mat[:-7])
individual = sio.loadmat(out_autoencoder)
mat = sio.loadmat(out_nii__2__mat)
mat = mat['mat']
W1g = np.squeeze(W1group[:,i,:])
W2g = np.squeeze(W2group[i,:,:])
b2g = np.squeeze(b2group[i,:])
W1i = individual['W1']
W2i = individual['W2']
b1i = individual['b1']
#hi = individual['h']
# individual decomposition recomputed for debugging purposes
d = out_autoencoder
hi = sigmoid( W1i.dot(mat)+np.tile(b1i, (mat.shape[1], 1)).transpose() )
xi_tilde = W2i.dot(hi)
img.write_decomposition(maps=xi_tilde, filename="{prefix}.tilde.nii.gz".format(prefix=out_nii__2__mat[0:out_nii__2__mat.find('.mat')]), normalise=False)
sio.savemat("{prefix}.tilde.mat".format(prefix=out_nii__2__mat[0:out_nii__2__mat.find('.mat')]), {'mat_tilde':xi_tilde})
# compute group hidden layer representation for an individual subject
# hack
h = sigmoid( W1g.dot(hi) + np.tile(b1group, (hi.shape[1], 1)).transpose() )
img.write_decomposition(maps=h, filename="{prefix}.groupind.nii.gz".format(prefix=d[0:d.find('.mat')]), normalise=False)
h = hgroup
# compute individual hidden layer representation for an individual subject
# hack
hi_tilde = sigmoid( W2g.dot(h) + np.tile(b2g, (h.shape[1], 1)).transpose() )
xi_tilde = W2i.dot(hi_tilde)
#h = hgroup
suffix = args.gae[args.gae.find('/')+1:len(args.gae)]
img = Image(img_filename=out_autoencoder_group, mask_filename=args.gmask)
img.write_decomposition(maps=hi_tilde, filename="{prefix}.tilde.{suffix}.h.nii.gz".format(prefix=d[0:d.find('.mat')],suffix=suffix), normalise=False)
img.write_decomposition(maps=xi_tilde, filename="{prefix}.tilde.{suffix}.x.nii.gz".format(prefix=d[0:d.find('.mat')],suffix=suffix), normalise=True)
parser = argparse.ArgumentParser(description='build nifti file from mat file')
parser.add_argument('-img', metavar='img', nargs='+', type=str, required=True)
parser.add_argument('-mask', metavar='mask', nargs='+', type=str, required=True)
parser.add_argument('-mat', metavar='mat', nargs='+', type=str, required=True)
parser.add_argument('-normalise', dest='normalise', action='store_true')
parser.set_defaults(normalise=False)
args = parser.parse_args()
if len(args.img)>1 and len(args.mask)>1:
for i, mat in enumerate(args.mat):
print args.img[i], args.mask[i], mat
img = Image(img_filename=args.img[i], mask_filename=args.mask[i])
f = sio.loadmat(mat)
hiddenlayer = f['h']
img.write_decomposition(maps=hiddenlayer, filename="{prefix}.nii.gz".format(prefix=mat[0:mat.find('.mat')]), normalise=args.normalise)
else:
img = Image(img_filename=args.img[0], mask_filename=args.mask[0])
mat = args.mat[0]
f = sio.loadmat(mat)
if 'h' in f.keys():
hiddenlayer = f['h']
if 'mat' in f.keys():
hiddenlayer = f['mat']
if 'x_tilde' in f.keys():
hiddenlayer = f['x_tilde']
img.write_decomposition(maps=hiddenlayer, filename="{prefix}.nii.gz".format(prefix=mat[0:mat.find('.mat')]), normalise=args.normalise)
number_individual_hiddenlayer = W1group.shape[1]/number_subjects
number_group_hiddenlayer = W1group.shape[0]
W1group=np.reshape(W1group,[number_subjects, number_group_hiddenlayer, number_individual_hiddenlayer])
W2group=np.reshape(W1group,[number_individual_hiddenlayer,number_subjects,number_group_hiddenlayer])
#W1group=np.reshape(W1group,[number_group_hiddenlayer, number_individual_hiddenlayer, number_subjects])
#bgroup=np.reshape(b1group, [number_subjects, number_individual_hiddenlayer])
W1group[np.abs(W1group)<args.thr]=0
for i,d in enumerate(args.ind_decomp):
img = Image(img_filename=args.ind_fmri_nii[i], mask_filename=args.ind_mask[i])
individual = sio.loadmat(d)
mat = sio.loadmat(args.ind_fmri_mat[i])
mat = mat['mat']
W1g = np.squeeze(W1group[i,:,:])
W2g = np.squeeze(W2group[:,i,:])
Wi = individual['W1']
bi = individual['b1']
#hi = individual['h']
pdb.set_trace()
hi = Wi.dot(mat)+np.tile(bi, (mat.shape[1], 1)).transpose()
hi=sigmoid(hi)
img.write_decomposition(maps=hi, filename="{prefix}.hdummy.nii.gz".format(prefix=d[0:d.find('.mat')]), normalise=False)
#pdb.set_trace()
h = W1g.dot(hi)+np.tile(b1group, (hi.shape[1], 1)).transpose()
h = sigmoid(h)
hi_tilde = W2g.dot(h)
img.write_decomposition(maps=h, filename="{prefix}.groupind.nii.gz".format(prefix=d[0:d.find('.mat')]), normalise=False)
img.write_decomposition(maps=hi_tilde, filename="{prefix}.tilde.nii.gz".format(prefix=d[0:d.find('.mat')]), normalise=False)
print d
#pdb.set_trace()
from niftybold.img.Image import Image
import scipy.io as sio
import argparse
import numpy as np
parser = argparse.ArgumentParser(description='build nifti file from mat file')
parser.add_argument('-img', metavar='img', type=str, required=True)
parser.add_argument('-mask', metavar='mask', type=str, required=True)
parser.add_argument('-mat', metavar='mat', type=str, required=True)
parser.add_argument('-normalise',
metavar='normalises',
type=bool,
default=False)
args = parser.parse_args()
f = sio.loadmat(args.mat)
hiddenlayer = f['h']
img = Image(img_filename=args.img, mask_filename=args.mask)
img.write_decomposition(maps=hiddenlayer,
filename="{prefix}.nii.gz".format(
prefix=args.mat[0:args.mat.find('.mat')]),
normalise=args.normalise)
def test_ae(filename='/Volumes/VAULT/YOAD/all/01-001.fmri.rest.1.volreg.band.nii.gz', filename_mask='/Volumes/VAULT/YOAD/all/registrations/01-001.fmri.rest.1.volreg.cortical_mask.nii.gz'):
img = Image(img_filename=filename, mask_filename=filename_mask)
xmat = img.image_mat_in_mask_normalised.T
ae = Autoencoder(30, max_iter=200, second_nonlinear=True, bias=True).fit(xmat)
img.write_decomposition(maps=ae.components_, filename='test_ae.nii.gz', normalise=False)
