def glm(image_ts, regressors):
'''
image_ts should be a matrix of the form (t x v) where t is the no. of timepoints
and v is the no. of voxels
regressors should be a matrix of the form (t x n) where t is the no. of timepoints
and n is the number of regressors
------------------
beta output is a serie of beta vector of the form (n x v).
'''
Y = image_ts
if len(regressors.shape) == 1:
X = np.expand_dims(regressors, axis=1)
else:
X = regressors
glm_dist = GeneralLinearModel(X)
glm_dist.transform(Y)
beta = glm_dist.get_beta()
return beta
def glm(image_ts, regressors):
'''
image_ts should be a matrix of the form (t x v) where t is the no. of timepoints
and v is the no. of voxels
regressors should be a matrix of the form (t x n) where t is the no. of timepoints
and n is the number of regressors
------------------
beta output is a serie of beta vector of the form (n x v).
'''
Y = image_ts
if len(regressors.shape) == 1:
X = np.expand_dims(regressors, axis=1)
else:
X = regressors
glm_dist = GeneralLinearModel(X)
glm_dist.transform(Y)
beta = glm_dist.get_beta()
return beta
def global_signal_regression(timeserie, regressor):
#Get timeseries data
Y = timeserie.data.T
X = np.expand_dims(regressor, axis=1)
glm_dist = GeneralLinearModel(X)
glm_dist.transform(Y)
beta_dist = glm_dist.get_beta()
r_signal = np.dot(X, beta_dist)
regressed_s = Y - r_signal
return regressed_s
def global_signal_regression(timeserie, regressor):
#Get timeseries data
Y = timeserie.data.T
X = np.expand_dims(regressor, axis=1)
glm_dist = GeneralLinearModel(X)
glm_dist.transform(Y)
beta_dist = glm_dist.get_beta()
r_signal = np.dot(X, beta_dist)
regressed_s = Y - r_signal
return regressed_s
]
my_roi.set_feature('contrast', contrast_feature)
my_roi.set_roi_feature('contrast_avg',
my_roi.representative_feature('contrast'))
########################################
# GLM analysis on the ROI average time courses
########################################
n_reg = len(names)
roi_tc = my_roi.get_roi_feature('signal_avg')
glm.fit(roi_tc.T)
plt.figure()
plt.subplot(1, 2, 1)
betas = glm.get_beta()
b1 = plt.bar(np.arange(n_reg - 1),
betas[:-1, 0],
width=.4,
color='blue',
label='region 1')
b2 = plt.bar(np.arange(n_reg - 1) + 0.3,
betas[:-1, 1],
width=.4,
color='red',
label='region 2')
plt.xticks(np.arange(n_reg - 1), names[:-1], fontsize=10)
plt.legend()
plt.title('Parameter estimates \n for the roi time courses')
bx = plt.subplot(1, 2, 2)
def KK_visit_Vermeer_GLM(stats_path):
stimulus_txt = np.loadtxt('/Users/davidzipser/Google_Drive/Data/stimuli/2015/6/Vermeer_1024x768/natural_stimuli_10Hz_Fri_Jun_19_21-56-59_2015.medium_letters/stimulus.txt')
n_Secs = len(stimulus_txt) # DANGER
n_TRs_all = 300
n_images = 24
stimulus_REGRESSORS = np.zeros((n_Secs, n_images))
for s in range(n_Secs):
if stimulus_txt[s] > 0:
stimulus_REGRESSORS[s,stimulus_txt[s]-1] = 1
hrf = KK_getcanonicalhrf()
hrf_1Hz = hrf[::5]
#PP[FF]=3,3
#plt.figure('hrf_1Hz')
#plt.plot(hrf_1Hz,'.-')
#mi(stimulus_REGRESSORS,'stimulus_REGRESSORS')
HRF_REG = np.zeros((n_TRs_all,n_images))
stimulus_times = np.arange(0,n_Secs)
TR_times = 0.9 * np.arange(0,n_TRs_all)
for i in range(n_images):
r = stimulus_REGRESSORS[:,i]
hp = np.convolve(r, hrf_1Hz)[:(np.shape(r)[0])]
hp_interp = np.interp(TR_times,stimulus_times,hp)
HRF_REG[:,i]=hp_interp
#PP[FF] = 5,8
#mi(HRF_REG,'HRF_REG')
nii = nib.load(opj(stats_path,'all_average.nii.gz'))
data = nii.get_data()
header = nii.get_header()
affine = nii.get_affine()
Voxels,Vox_xyzs = vol_to_voxels(data)
# - http://nipy.org/nipy/api/generated/nipy.modalities.fmri.glm.html
from nipy.modalities.fmri.glm import GeneralLinearModel
n_TRs = 294
n_voxels = np.shape(Voxels)[1]
n_images = 24
DATA = Voxels
REGRESSORS = HRF_REG.copy()[6:] # this deals with DELETE volumes
model = GeneralLinearModel(REGRESSORS)
model.fit(DATA)
#mi(DATA,1,[1,1,1],'DATA ' + str(np.shape(DATA)))
#mi(REGRESSORS,2,[1,1,1],'REGRESSORS ' + str(np.shape(REGRESSORS)))
betas = model.get_beta()
#mi(betas,3,[1,1,1],'betas ' + str(np.shape(betas)))
mse = model.get_mse()
#plt.figure('mse ' + str(np.shape(mse)))
#plt.plot(mse)
vol_betas = vox_list_to_vol(betas,Vox_xyzs,data[:,:,:,0].copy())
betas_nii = nib.Nifti1Image(vol_betas, affine, header)
nib.save(betas_nii, opj(stats_path,'betas.nipy_GLM.nii.gz'))
for id in my_roi.get_id()]
my_roi.set_feature('contrast', contrast_feature)
my_roi.set_roi_feature('contrast_avg',
my_roi.representative_feature('contrast'))
########################################
# GLM analysis on the ROI average time courses
########################################
n_reg = len(names)
roi_tc = my_roi.get_roi_feature('signal_avg')
glm.fit(roi_tc.T)
plt.figure()
plt.subplot(1, 2, 1)
betas = glm.get_beta()
b1 = plt.bar(np.arange(n_reg - 1), betas[:-1, 0], width=.4, color='blue',
label='region 1')
b2 = plt.bar(np.arange(n_reg - 1) + 0.3, betas[:- 1, 1], width=.4,
color='red', label='region 2')
plt.xticks(np.arange(n_reg - 1), names[:-1], fontsize=10)
plt.legend()
plt.title('Parameter estimates \n for the roi time courses')
bx = plt.subplot(1, 2, 2)
my_roi.plot_feature('contrast', bx)
########################################
# fitted and adjusted response
########################################
def _fit_model(self, ds, X, reg_names):
glm = GeneralLinearModel(X)
glm.fit(ds.samples, **self.glmfit_kwargs)
out = Dataset(glm.get_beta(),
sa={self.get_space(): reg_names})
return glm, out
def _fit_model(self, ds, X, reg_names):
glm = GeneralLinearModel(X)
glm.fit(ds.samples, **self.glmfit_kwargs)
out = Dataset(glm.get_beta(), sa={self.get_space(): reg_names})
return glm, out
