def reshape_netwise(data_scale):
# Reshape with the following dim: nSubjects, nfeatures, nfeatures
for i in range(0,data_scale.shape[0]):
if i==0:
all_subjmat = ts.vec2mat(data_scale[i,:])[np.newaxis,...]
else:
all_subjmat = np.vstack((all_subjmat,ts.vec2mat(data_scale[i,:])[np.newaxis,...]))
#print all_subjmat.shape
return all_subjmat
def reshape_netwise(data_scale):
# Reshape with the following dim: nSubjects, nfeatures, nfeatures
for i in range(0, data_scale.shape[0]):
if i == 0:
all_subjmat = ts.vec2mat(data_scale[i, :])[np.newaxis, ...]
else:
all_subjmat = np.vstack(
(all_subjmat, ts.vec2mat(data_scale[i, :])[np.newaxis, ...]))
#print all_subjmat.shape
return all_subjmat
def itStability(x,y,ind,k=1,samp_ratio=0.5,nsample=100):
'''
A random iterative resampling of the subject to compute the stability of the selected features
'''
subj_idx = range(0,x.shape[0])
stability_hr_mat = np.zeros((len(ind),len(ind)))
for i in range(0,nsample):
sample_idx = np.random.permutation(subj_idx)[:np.int(len(subj_idx)*samp_ratio)]
bestidx = getkBest(x[sample_idx,:],y[sample_idx],k)
it_vote = np.zeros(x.shape[1])
it_vote[bestidx] = 1 # create the vector of selected features
lr_mat = ts.vec2mat(it_vote,include_diag=True) # convert to the low resolution matrix
hr_mat = cls.projectmat(lr_mat,ind) # remap in HR
# Add the matrix to the main HR stability matrix
stability_hr_mat += hr_mat
stability_hr_mat /= nsample
return stability_hr_mat
