def Wt_calc(stc, epochs_eeg, epochs_meg, meas_dims):
total_batch_size = len(
stc) #number of events. we'll consider each event an example.
n_steps = stc[0]._data.shape[1]
n_eeg = epochs_eeg.get_data().shape[1]
eeg_xyz = np.squeeze(
np.array([
epochs_eeg.info['chs'][i]['loc'][:3].reshape([1, 3])
for i in range(0, n_eeg)
]))
n_meg = epochs_meg.get_data().shape[1]
meg_xyz = np.squeeze(
np.array([
epochs_meg.info['chs'][i]['loc'][:3].reshape([1, 3])
for i in range(0, n_meg)
]))
eeg_data = np.array(epochs_eeg.get_data()) #batch_sizexmxn_steps
meg_data = np.array(epochs_meg.get_data()) #batch_sizexmxn_steps
tf_meas = meas_class.meas(meg_data, meg_xyz, eeg_data, eeg_xyz, meas_dims,
n_steps, total_batch_size)
Wt = tf_meas.pca()
return Wt
def cnn_xjustdims(self):
if self.selection is 'all':
self.total_batch_size = len(self.stc)#number of events. we'll consider each event an example.
else:
self.total_batch_size = len(self.selection)#number of events. we'll consider each event an example.
n_eeg = self.epochs_eeg.get_data().shape[1]
eeg_xyz=np.squeeze(np.array([self.epochs_eeg.info['chs'][i]['loc'][:3].reshape([1,3]) for i in range(0,n_eeg)]))
n_meg = self.epochs_meg.get_data().shape[1]
meg_xyz=np.squeeze(np.array([self.epochs_meg.info['chs'][i]['loc'][:3].reshape([1,3]) for i in range(0,n_meg)]))
if self.selection is 'all':
eeg_data = np.array(self.epochs_eeg.get_data())#batch_sizexmxn_steps
meg_data = np.array(self.epochs_meg.get_data())#batch_sizexmxn_steps
else:
eeg_data = np.array(self.epochs_eeg.get_data())[self.selection,:,:]#batch_sizexmxn_steps
meg_data = np.array(self.epochs_meg.get_data())[self.selection,:,:]#batch_sizexmxn_steps
self.n_steps=meg_data.shape[2]
self.meas_dims=[11,11]
print "Image grid dimensions: ", self.meas_dims
tf_meas = meas_class.meas(meg_data,meg_xyz, eeg_data,eeg_xyz, self.meas_dims, self.n_steps, self.total_batch_size)
if self.pca is True:
self.Wt=tf_meas.pca(Wt=self.Wt)
elif self.pca is False:
tf_meas.scale()
else:
pass
tf_meas.interp()
tf_meas.reshape()
#for b in range(0,n_steps*total_batch_size):
# tf_meas.plot(b)
self.meas_img_all = tf_meas.meas_img
self.m =tf_meas.m
if self.selection is 'all':
dipole=np.array([self.stc[i]._data for i in range(0,len(self.stc))]).transpose((1,2,0))
else:
dipole=np.array([self.stc[i]._data for i in self.selection]).transpose((1,2,0))
if self.locate is not False:
if self.rnn is True or self.cnn is 'fft':
self.locationXYZT()
else:
self.location()
else:
if self.rnn is True or self.cnn is 'fft':
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipoleXYZT_OH(dipole,subsample=self.subsample)
#bxnxp
else:
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipole(dipole,subsample=self.subsample)
def rat_prepro(self):
tf_meas = meas_class.meas(self.meg_data,self.meg_xyz, self.eeg_data,self.eeg_xyz, self.meas_dims, self.n_steps, self.batch_size)
if self.pca is True:
self.Wt=tf_meas.pca(Wt = self.Wt)
elif self.pca is False:
tf_meas.scale()
else:
pass
if self.n_chan_in is 2:
if self.cnn is True:
print "Image grid dimensions: ", self.meas_dims
tf_meas.interp()
tf_meas.reshape()
self.meas_img_all = tf_meas.meas_img
self.m =tf_meas.m
else:
self.meas_img_all = np.vstack((tf_meas.meas_in[0],tf_meas.meas_in[1]))
self.m =tf_meas.m0+tf_meas.m1
elif self.n_chan_in is 1:
if self.cnn is True:
print "Image grid dimensions: ", self.meas_dims
tf_meas.interp()
tf_meas.reshape()
self.meas_img_all = np.expand_dims(tf_meas.meas_img[:,:,:,:,0],axis=-1)
self.m =tf_meas.m
else:
self.meas_img_all = np.vstack((tf_meas.meas_in[0]))
self.m =tf_meas.m0
if self.locate is not False:
if self.rnn is True or self.cnn is 'fft':
self.location_rat_XYZT()
else:
self.location_rat()
else:
if self.rnn is True or self.cnn is 'fft':
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipoleXYZT_OH(self.dipole,subsample=self.subsample)
#bxnxp
else:
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipole(self.dipole,subsample=self.subsample)
lambda2,
method=method,
pick_ori=None)
stc.save('sample_audvis-source-epochs')
batch_size = len(stc) #number of events. we'll consider each event an example.
n_steps = meg_data.shape[2]
dipole = stc[0]._data
for i in range(1, batch_size):
dipole = np.dstack((dipole, stc[i]._data))
#pxn_stepsxbatchsize
qtrue, p = meas_class.scale_dipole(dipole)
#bxnxp
#meas_meg_in n_stepsxmxbatchsize
#meas_eeg_in n_stepsxmxbatchsize
meas_dims = [int(np.sqrt(n_eeg + n_meg)),
int(np.sqrt(n_eeg + n_meg))
] #remember, MUST have image pixels<#sensors
print "Image grid dimensions: ", meas_dims
tf_meas = meas_class.meas(meg_data, meg_xyz, eeg_data, eeg_xyz, meas_dims,
n_steps, batch_size)
tf_meas.scale()
tf_meas.interp()
tf_meas.reshape()
tf_meas.plot(1)
tf_meas.plot(10)
tf_meas.plot(100)
def fftcnn_xjustdims(self):
if self.selection is 'all':
self.total_batch_size = len(self.stc)#number of evens. we'll consider each event an example.
else:
self.total_batch_size = len(self.selection)#number of events. we'll consider each event an example.
n_eeg = self.epochs_eeg.get_data().shape[1]
eeg_xyz=np.squeeze(np.array([self.epochs_eeg.info['chs'][i]['loc'][:3].reshape([1,3]) for i in range(0,n_eeg)]))
n_meg = self.epochs_meg.get_data().shape[1]
meg_xyz=np.squeeze(np.array([self.epochs_meg.info['chs'][i]['loc'][:3].reshape([1,3]) for i in range(0,n_meg)]))
if self.selection is 'all':
eeg_data = np.array(self.epochs_eeg.get_data())#batch_sizexmxself.n_steps
meg_data = np.array(self.epochs_meg.get_data())#batch_sizexmxn_steps
else:
eeg_data = np.array(self.epochs_eeg.get_data())[self.selection,:,:]#batch_sizexmxn_steps
meg_data = np.array(self.epochs_meg.get_data())[self.selection,:,:]#batch_sizexmxn_steps
self.n_steps=meg_data.shape[2]
self.meas_dims=[self.n_steps, n_eeg+n_meg]
print "Meas dims in: ", self.meas_dims
tf_meas = meas_class.meas(meg_data,meg_xyz, eeg_data,eeg_xyz, self.meas_dims, self.n_steps, self.total_batch_size)
if self.pca is True:
self.Wt=tf_meas.pca(Wt=self.Wt)
elif self.pca is False:
tf_meas.scale()
else:
pass
tf_meas.stack_reshape()
#ff=np.fft.fft(tf_meas.meas_stack,axis=1)
ff=tf_meas.meas_stack
#print ff.shape
#meas_img_all = np.expand_dims(np.abs(ff)*np.abs(ff),-1)
self.meas_img_all = np.expand_dims(ff,-1)
#print meas_img_all.shape
self.m =tf_meas.m
if self.selection is 'all':
dipole=np.array([self.stc[i]._data for i in range(0,len(self.stc))]).transpose((1,2,0))
else:
dipole=np.array([self.stc[i]._data for i in self.selection]).transpose((1,2,0))
if self.locate is not False:
if self.rnn is True or self.cnn is 'fft':
self.locationXYZT()
else:
self.location()
else:
if self.rnn is True or self.cnn is 'fft':
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipoleXYZT_OH(dipole,subsample=self.subsample)
#bxnxp
else:
#pxn_stepsxbatchsize
self.qtrue_all,self.p=meas_class.scale_dipole(dipole,subsample=self.subsample)
def xcnn_xjustdims(stc,
epochs,
epochs_eeg,
epochs_meg,
subject,
selection='all',
pca=False,
subsample=1,
justdims=True,
cnn=True,
locate=True,
rnn=True,
Wt=None):
if selection is 'all':
total_batch_size = len(
stc) #number of events. we'll consider each event an example.
else:
total_batch_size = len(
selection
) #number of events. we'll consider each event an example.
n_eeg = epochs_eeg.get_data().shape[1]
eeg_xyz = np.squeeze(
np.array([
epochs_eeg.info['chs'][i]['loc'][:3].reshape([1, 3])
for i in range(0, n_eeg)
]))
n_meg = epochs_meg.get_data().shape[1]
meg_xyz = np.squeeze(
np.array([
epochs_meg.info['chs'][i]['loc'][:3].reshape([1, 3])
for i in range(0, n_meg)
]))
if selection is 'all':
eeg_data = np.array(epochs_eeg.get_data()) #batch_sizexmxn_steps
meg_data = np.array(epochs_meg.get_data()) #batch_sizexmxn_steps
else:
eeg_data = np.array(
epochs_eeg.get_data())[selection, :, :] #batch_sizexmxn_steps
meg_data = np.array(
epochs_meg.get_data())[selection, :, :] #batch_sizexmxn_steps
n_steps = meg_data.shape[2]
meas_dims = n_eeg + n_meg
print "Meas dims in: ", meas_dims
tf_meas = meas_class.meas(meg_data, meg_xyz, eeg_data, eeg_xyz, meas_dims,
n_steps, total_batch_size)
if pca is True:
Wt = tf_meas.pca(Wt=Wt)
elif pca is False:
tf_meas.scale()
else:
pass
tf_meas.stack_reshape()
meas_img_all = tf_meas.meas_stack
m = tf_meas.m
if selection is 'all':
dipole = np.array([stc[i]._data
for i in range(0, len(stc))]).transpose((1, 2, 0))
else:
dipole = np.array([stc[i]._data for i in selection]).transpose(
(1, 2, 0))
if locate is not False:
if rnn is True or cnn is 'fft':
qtrue_all, p = locationXYZT(stc,
subject,
selection=selection,
locate=locate)
else:
qtrue_all, p = location(stc,
subject,
selection=selection,
locate=locate)
else:
if rnn is True or cnn is 'fft':
#pxn_stepsxbatchsize
qtrue_all, p = meas_class.scale_dipoleXYZT_OH(dipole,
subsample=subsample)
#bxnxp
else:
#pxn_stepsxbatchsize
qtrue_all, p = meas_class.scale_dipole(dipole, subsample=subsample)
#bxnxp
del stc, epochs, epochs_eeg, epochs_meg
return meas_img_all, qtrue_all, meas_dims, m, p, n_steps, total_batch_size, Wt
def rat_real(stim='Tones',
selection='all',
pca=True,
subsample=1,
justdims=True,
cnn=False,
locate=True,
treat=None,
rnn=False,
Wt=None):
#print 'selection ',selection
ecog_thresh = 1e-5
if stim == 'Tones':
name = '/home/jcasa/meld/code/python/rattest/processed/ECOG_MEG_Tones.grouped.pickle'
elif stim == 'P1':
name = '/home/jcasa/meld/code/python/rattest/processed/ECOG_MEG_P1.grouped.pickle'
elif stim == 'P0':
name = '/home/jcasa/meld/code/python/rattest/processed/MEG_ECOG.grouped.pickle'
with open(name, 'r') as f:
b = pickle.load(f)
ecog_data = np.transpose(np.array(b["ECoG_average"]),
(1, 2, 0)) #pxnxb - for dipole scaling
eeg_data = np.transpose(np.array(b["ECoG_average"]), (0, 1, 2))
ecog_data[abs(ecog_data) > ecog_thresh] = 100e-9 #A
#eeg_data[abs(eeg_data)>ecog_thresh]=100e-6#V
meg_data = np.transpose(np.array(b["MEG_average"]),
(0, 1, 2)) #bxmxn - for meas_class pca formatting
fs_MEG = b["fs_MEG"]
fs_ECoG = b["fs_ECoG"]
flag = b["flag"]
n_treat = b["n_treat"]
treatments = b["treatments"]
meg_xyz = b["meg_xyz"] / 1000. #mx3
ecog_xyz = b["ecog_xyz"] / 1000. #in meters
n_chan_in = 1
n_steps = meg_data.shape[2]
total_batch_size = ecog_data.shape[2]
m = meg_data.shape[1]
p = ecog_data.shape[0]
meas_dims = m
#print 'n_steps', n_steps, 'total_batch_size', total_batch_size, 'm', m, 'p', p
#print 'MEG array: ',meg_data.shape
#print 'ECOG array: ',ecog_data.shape
#print 'fake EEG array: ',eeg_data.shape
if pca:
tf_meas = meas_class.meas(meg_data, meg_xyz, np.array([]),
np.array([]), meas_dims, n_steps,
total_batch_size)
Wt = tf_meas.pca()
tf_meas.stack_reshape(
n_chan_in=n_chan_in) #ignore ecog - just a placeholder
meas_img_all = tf_meas.meas_stack
else:
meas_img_all = np.transpose(meg_data, (0, 2, 1))
#scale dipoles ~ ecog
#print 'Locate: ',locate
if rnn:
if locate is False:
qtrue_all, p = meas_class.scale_dipoleXYZT_OH(ecog_data,
subsample=subsample)
else:
p = 3
qtrue_all = location_rat_XYZT(locate, total_batch_size, n_steps, p,
ecog_data, ecog_xyz)
else:
if locate is False:
qtrue_all, p = meas_class.scale_dipole(ecog_data,
subsample=subsample)
else:
p = 3
qtrue_all = location_rat(locate, total_batch_size, n_steps, p,
ecog_data, ecog_xyz)
if justdims is True:
return meas_dims, m, p, n_steps, total_batch_size, Wt
else:
#print 'meas_img_all ',meas_img_all.shape
#print 'qtrue_all ',qtrue_all.shape
if selection is 'all':
return meas_img_all, qtrue_all, meas_dims, m, p, n_steps, total_batch_size, Wt
else:
return meas_img_all[selection, :, :], qtrue_all[
selection, :, :], meas_dims, m, p, n_steps, np.size(
selection), Wt
def rat_prepro(n_chan_in,
dipole,
dipole_xyz,
meg_data,
meg_xyz,
eeg_data,
eeg_xyz,
meas_dims,
n_steps,
batch_size,
subject,
selection='all',
pca=True,
subsample=1,
justdims=False,
cnn=True,
locate=True,
rnn=False,
Wt=None):
print n_chan_in
if locate is True:
p = 3
elif locate > 0:
p = 3 * locate
else:
p = dipole.shape[0]
tf_meas = meas_class.meas(meg_data, meg_xyz, eeg_data, eeg_xyz, meas_dims,
n_steps, batch_size)
if pca is True:
Wt = tf_meas.pca(Wt=Wt)
elif pca is False:
tf_meas.scale()
else:
pass
if n_chan_in is 2:
if cnn is True:
print "Image grid dimensions: ", meas_dims
tf_meas.interp()
tf_meas.reshape()
meas_img_all = tf_meas.meas_img
m = tf_meas.m
else:
tf_meas.stack_reshape(n_chan_in=n_chan_in)
meas_img_all = tf_meas.meas_stack
m = tf_meas.m0 + tf_meas.m1
meas_dims = m
elif n_chan_in is 1:
if cnn is True:
print "Image grid dimensions: ", meas_dims
tf_meas.interp()
tf_meas.reshape()
meas_img_all = np.expand_dims(tf_meas.meas_img[:, :, :, :, 0],
axis=-1)
m = tf_meas.m
else:
tf_meas.stack_reshape(n_chan_in=n_chan_in)
meas_img_all = tf_meas.meas_stack
m = tf_meas.m0
meas_dims = m
if locate is not False:
if rnn is True or cnn is 'fft':
qtrue_all = location_rat_XYZT(locate, batch_size, n_steps, p,
dipole, dipole_xyz)
else:
qtrue_all = location_rat(locate, batch_size, n_steps, p, dipole,
dipole_xyz)
else:
#print p, dipole.shape, "Dipoles (rat_prepro, before scaling)"
if rnn is True or cnn is 'fft':
#pxn_stepsxbatchsize
qtrue_all, p = meas_class.scale_dipoleXYZT_OH(dipole,
subsample=subsample)
#bxnxp
else:
#pxn_stepsxbatchsize
qtrue_all, p = meas_class.scale_dipole(dipole, subsample=subsample)
#bxnxp
#print p, qtrue_all.shape, "Dipoles (rat_prepro, after scaling)"
assert qtrue_all.shape == (batch_size, n_steps,
p), str(qtrue_all.shape) + ' ' + str(
(batch_size, n_steps, p))
if cnn is True:
assert meas_img_all.shape == (
batch_size, n_steps, meas_dims[0], meas_dims[1],
1), str(meas_img_all.shape) + ' ' + str(
(batch_size, n_steps, meas_dims[0], meas_dims[1], 1))
else:
assert meas_img_all.shape == (batch_size, n_steps,
m), str(meas_img_all.shape) + ' ' + str(
(batch_size, n_steps, m))
return meas_img_all, qtrue_all, meas_dims, m, p, n_steps, batch_size, Wt