def test():
# simulated image
mat_contents = sio.loadmat('data/sim_3dmri.mat')
x0 = mat_contents["sim_3dmri"]
x = x0[:, :, 40:60]
nx, ny, nz = x.shape
mask = ut.mask3d(nx, ny, nz, [15, 15, 0])
FTm = opts.FFTnd_kmask(mask)
#ut.plotim3(np.absolute(mask[:,:,1:10]))#plot the mask
# undersampling in k-space
b = FTm.forward(x)
ut.plotim3(np.absolute(FTm.backward(b))) #undersampled imag
scale = scaling(FTm, b)
#b = b/scale
#do cs mri recon
Nite = 2 #number of iterations
step = 0.5 #step size
#th = 1 #threshold
#xopt = solvers.IST_2(FTm.forward,FTm.backward,b, Nite, step,1) #soft thresholding
xopt = solvers.ADMM_l2Afxnb_tvx(FTm.forward, FTm.backward, b, Nite, step,
10, 1)
#xopt = solvers.ADMM_l2Afxnb_l1x_2( FTm.forward, FTm.backward, b, Nite, step, 100, 1 )
ut.plotim3(np.absolute(xopt))
def test():
# simulated image
mat_contents = sio.loadmat('data/kellman_data/PKdata3.mat',
struct_as_record=False,
squeeze_me=True)
xdata = mat_contents["data"]
im = xdata.images
field = xdata.FieldStrength
b0_gain = 100.0
TE = b0_gain * xdata.TE
fat_freq_arr = (1.0 / b0_gain) * 42.58 * field * np.array(
[-3.80, -3.40, -2.60, -1.94, -0.39, 0.60])
fat_rel_amp = np.array([0.087, 0.693, 0.128, 0.004, 0.039, 0.048])
#ut.plotim3(np.real(im[:,:,:]))
nx, ny, nte = im.shape
#undersampling
mask = ut.mask3d(nx, ny, nte, [15, 15, 0], 0.8)
#FTm = opts.FFT2d_kmask(mask)
FTm = opts.FFTW2d_kmask(mask)
#FTm = opts.FFT2d()
b = FTm.forward(im)
scaling = ut.optscaling(FTm, b)
b = b / scaling
#ut.plotim3(mask)
#ut.plotim3(np.absolute(FTm.backward(b))) #undersampled imag
#parameters
xpar = np.zeros((nx, ny, 3), np.complex128)
#xpar[:,:,0] = 10*np.ones((nx,ny))
#ut.plotim3(np.absolute(xpar),[3,-1])
# IDEAL and FFT jointly
IDEAL = idealc.IDEAL_opt2(TE, fat_freq_arr,
fat_rel_amp) #fat_freq_arr , fat_rel_amp
Aideal_ftm = opts.joint2operators(IDEAL, FTm) #(FTm,IDEAL)#
IDEAL.set_x(xpar) #this set the size of data
dwt = opts.DWT2d(wavelet='haar', level=4)
#do tv cs mri recon
Nite = 10 #number of iterations
step = 1 #step size
l1_r = 0.001
tv_r = 0.0001 # regularization term for tv term
rho = 1.0
xpar = ADMM_l2Agaussnewton_l1Tfx(IDEAL, FTm, dwt, b, Nite, step, l1_r, rho)
#xpar = ADMM_l2Aguassnewton_tvx(IDEAL, FTm, b, Nite, step, tv_r, rho )
#xpar = ADMM_l2Agaussnewton_l1Tfx_tvx( IDEAL, FTm, dwt, b, Nite, step, l1_r, tv_r, rho)
ut.plotim3(np.absolute(xpar)[..., 0:2], bar=1)
ut.plotim3(b0_gain * np.real(xpar)[..., 2], bar=1)
ut.plotim3(b0_gain * 2.0 * np.pi * np.imag(xpar)[..., 2], bar=1)
def test():
# read rf and tr arrays from mat file
mat_contents = sio.loadmat(pathdat +
'mrf_t1t2b0pd_mrf_randphasecyc_traintest.mat')
far = np.array(mat_contents["rf"].astype(np.complex128).squeeze())
trr = np.array(mat_contents["trr"].astype(np.float64).squeeze())
# input MRF time courses
mat_contents2 = sio.loadmat(pathdat + 'datax1.mat')
data_x = np.array(mat_contents2["datax1"]).astype(np.float64)
# prepare for sequence simulation, y->x_hat
Nk = far.shape[0]
Nexample = data_x.shape[0]
ti = 10 #ms
M0 = np.array([0.0, 0.0, 1.0]).astype(np.float64)
#image size
nx = 217
ny = 181
# mask in ksp
mask = ut.mask3d(nx, ny, Nk, [15, 15, 0], 0.4)
#FTm = opts.FFT2d_kmask(mask)
FTm = cuopts.FFT2d_cuda_kmask(mask)
#intial timing
timing = utc.timing()
# x real and imag parts should be combined
data_x_c = np.zeros((Nexample, Nk), np.complex128)
data_x_c = ssmrf.seqdata_realimag_2complex(data_x)
#could do mask M here, data_x_c = M*data_x_c
data_x_c = mask_ksp3d(nx, ny, Nk, FTm, data_x_c)
#data_x0_c = data_x_c
data_x = ssmrf.seqdata_complex_2realimag(data_x_c)
data_x0 = data_x
# intial tmp data
test_outy0 = np.zeros((Nexample, 4), dtype=np.float64)
test_outy = np.zeros((Nexample, 4), dtype=np.float64)
tmptest_outy = np.zeros((Nexample, 4), dtype=np.float64)
acctest_outy = np.zeros((Nexample, 4), dtype=np.float64)
# f'(y), cnn model
timing.start()
ssmrf.batch_apply_tf_cuda(Nexample, ny, sess.run, x, data_x, y_conv,
test_outy0, keep_prob)
timing.stop().display('CNN model first estimation ')
test_outy0 = constraints(test_outy0)
acctest_outy = test_outy0
# for loop start here
for _ in range(40):
#cuda simulate bloch for each pixel
T1r, T2r, dfr, PDr = ssmrf.set_par((test_outy))
timing.start()
data_x_c = ssmrf.bloch_sim_batch_cuda(Nexample, 7 * ny, Nk, PDr, T1r,
T2r, dfr, M0, trr, far, ti)
timing.stop().display('Bloch sim in loop ')
#0.5* d||M*f(p)-b||_2^2/dp = f'*(M*f(p)-b) = f'(M*f(p)) - f'(b)
# could do mask, M, here, e.g. data_x_c=M*data_x_c
data_x_c = mask_ksp3d(nx, ny, Nk, FTm, data_x_c)
#seperate real/imag parts or abs/angle parts
data_x = ssmrf.seqdata_complex_2realimag(data_x_c)
# apply CNN model, f'(i)
ssmrf.batch_apply_tf_cuda(Nexample, ny, sess.run, x, data_x, y_conv,
tmptest_outy, keep_prob)
tmptest_outy = constraints(tmptest_outy)
# gradient = f'(M * f(p)) - f'(b)
tmptest_outy = tmptest_outy - test_outy0 #+ 0.2*test_outy
# gradient descent for test_outy
acctest_outy = acctest_outy - tmptest_outy
print('gradient 0.5* d||f(p)-b||_2^2/dp: %g' %
np.linalg.norm(tmptest_outy))
#print('test_outy: %g' % np.linalg.norm(test_outy))
acctest_outy = constraints(acctest_outy)
#could do soft thresholding on test_outy here, i.e. test_outy = threshold(test_outy)
#test_outy = l0_par(nx, ny, acctest_outy) #
test_outy = tv_par(nx, ny, acctest_outy)
sio.savemat(pathdat + 'cnn_cs_testouty.mat', {'test_outy': test_outy})
def test():
# simulated image
mat_contents = sio.loadmat(pathdat, struct_as_record=False, squeeze_me=True)
xdata = mat_contents["data"]
im = xdata.images
field = xdata.FieldStrength
b0_gain = 100.0
TE = b0_gain * xdata.TE
fat_freq_arr = (1.0/b0_gain) * 42.58 * field * np.array([-3.80, -3.40, -2.60, -1.94, -0.39, 0.60])
fat_rel_amp = np.array([0.087, 0.693, 0.128, 0.004, 0.039, 0.048])
ut.plotim3(np.real(im[:,:,:]))
nx,ny,nte = im.shape
#undersampling
mask = ut.mask3d( nx, ny, nte, [15,15,0], 0.8)
#FTm = opts.FFT2d_kmask(mask)
#FTm = opts.FFTW2d_kmask(mask)
FTm = opts.FFT2d()
b = FTm.forward(im)
scaling = ut.optscaling(FTm,b)
b = b/scaling
#ut.plotim3(mask)
ut.plotim3(np.absolute(FTm.backward(b))) #undersampled imag
#parameters
xpar = np.zeros((nx,ny,3), np.complex128)
#xpar[:,:,0] = 10*np.ones((nx,ny))
#ut.plotim3(np.absolute(xpar),[3,-1])
# IDEAL and FFT jointly
IDEAL = idealc.IDEAL_opt2(TE, fat_freq_arr , fat_rel_amp )#fat_freq_arr , fat_rel_amp
Aideal_ftm = opts.joint2operators(IDEAL, FTm)#(FTm,IDEAL)#
IDEAL.set_x(xpar) #should update in each gauss newton iteration
residual = IDEAL.residual(b, FTm)
#ut.plotim3(np.absolute(FTm.backward(residual)))
# wavelet and x+d_x
addx_water = idealc.x_add_dx()
addx_fat = idealc.x_add_dx()
addx_df = idealc.x_add_dx()
#addx = idealc.x_add_dx()
#addx.set_w([1, 1, 0.0001])
addx_water.set_x(xpar[...,0]) #should update in each gauss newton iteration
addx_fat.set_x (xpar[...,1])
addx_df.set_x (xpar[...,2])
dwt = opts.DWT2d(wavelet = 'haar', level=4)
tvop = tvopc.TV2d()
Adwt_addx_w = opts.joint2operators(dwt, addx_water)
Adwt_addx_f = opts.joint2operators(dwt, addx_fat)
Adwt_addx_d = opts.joint2operators(tvop, addx_df)
#Adwt_addx = opts.joint2operators(dwt, addx)
#CGD
Nite = 80
l1_r1 = 0.01
l1_r2 = 0.01
def f(xi):
#return np.linalg.norm(Aideal_ftm.forward(xi)-residual)
return alg.obj_fidelity(Aideal_ftm, xi, residual) \
+ l1_r1 * alg.obj_sparsity(Adwt_addx_w, xi[...,0])\
+ l1_r1 * alg.obj_sparsity(Adwt_addx_f, xi[...,1])\
+ l1_r2 * alg.obj_sparsity(Adwt_addx_d, xi[...,2])
def df(xi):
#return 2*Aideal_ftm.backward(Aideal_ftm.forward(xi)-residual)
gradall = alg.grad_fidelity(Aideal_ftm, xi, residual)
gradall[...,0] += l1_r1 * alg.grad_sparsity(Adwt_addx_w, xi[...,0])
gradall[...,1] += l1_r1 * alg.grad_sparsity(Adwt_addx_f, xi[...,1])
gradall[...,2] += l1_r2 * alg.grad_sparsity(Adwt_addx_d, xi[...,2])
return gradall
#do soft thresholding
#Nite = 20 #number of iterations
#step = 0.1 #step size
#th = 1 # theshold level
#do tv cs mri recon
#Nite = 40 #number of iterations
#step = 1 #step size
#tv_r = 0.01 # regularization term for tv term
#rho = 1.0
ostep = 1.0#0.3
for i in range(20):
#wavelet L1 IST
# dxpar = solvers.IST_3( Aideal_ftm.forward, Aideal_ftm.backward,\
# Adwt_addx.backward, Adwt_addx.forward, residual, Nite, step, th )
#wavelet L1 ADMM
# dxpar = solvers.ADMM_l2Afxnb_l1Tfx( Aideal_ftm.forward, Aideal_ftm.backward, \
# Adwt_addx.backward, Adwt_addx.forward, residual, Nite, step, tv_r, rho,15 )
# TV ADMM
# dxpar = solvers.ADMM_l2Afxnb_tvx( Aideal_ftm.forward, Aideal_ftm.backward, residual\
# , Nite, step, tv_r, rho )
# L2 CGD
# dxpar = pf.prox_l2_Afxnb_CGD2( Aideal_ftm.forward, Aideal_ftm.backward, residual, rho, Nite )
# dxpar = pf.prox_l2_Afxnb_CGD2( Aideal_ftm.forward, Aideal_ftm.backward, residual, Nite )
# L1 CGD
dxpar = alg.conjugate_gradient(f, df, Aideal_ftm.backward(residual), Nite )
ostep,j = alg.BacktrackingLineSearch(f, df, xpar, dxpar)
if i%1 == 0:
ut.plotim3(np.absolute(xpar + ostep*dxpar)[...,0:2],bar=1)
ut.plotim3(np.real(xpar + ostep*dxpar)[...,2],bar=1)
ut.plotim3(np.imag(xpar + ostep*dxpar)[...,2],bar=1)
xpar = xpar + ostep*dxpar#.astype(np.float64)
#if i > 1: #unwrapping on frequence
# xpar[:,:,2] = np.real(unwrap_freq(np.real(xpar[:,:,2])))\
# +1j*(np.imag(xpar[:,:,2]))
IDEAL.set_x(xpar) #should update in each gauss newton iteration
residual = IDEAL.residual(b, FTm)
# addx.set_x(xpar) #should update in each gauss newton iteration
addx_water.set_x(xpar[...,0]) #should update in each gauss newton iteration
addx_fat.set_x (xpar[...,1])
addx_df.set_x (xpar[...,2])
ut.plotim3(np.absolute(xpar)[...,0:2],bar=1)
ut.plotim3(np.real(xpar + ostep*dxpar)[...,2],bar=1)
ut.plotim3(np.imag(xpar + ostep*dxpar)[...,2],bar=1)
sio.savemat(pathdat + 'IDEAL_CGD_result.mat', {'xpar':xpar, 'residual':residual})
def test():
# simulated image
mat_contents = sio.loadmat('data/kellman_data/PKdata3.mat',
struct_as_record=False,
squeeze_me=True)
xdata = mat_contents["data"]
im = xdata.images
field = xdata.FieldStrength
b0_gain = 100.0
TE = b0_gain * xdata.TE
fat_freq_arr = (1.0 / b0_gain) * 42.58 * field * np.array(
[-3.80, -3.40, -2.60, -1.94, -0.39, 0.60])
fat_rel_amp = np.array([0.087, 0.693, 0.128, 0.004, 0.039, 0.048])
ut.plotim3(np.real(im[:, :, :]))
nx, ny, nte = im.shape
#undersampling
mask = ut.mask3d(nx, ny, nte, [15, 15, 0], 0.8)
FTm = opts.FFT2d_kmask(mask)
#FTm = opts.FFTW2d_kmask(mask)
#FTm = opts.FFT2d()
b = FTm.forward(im)
scaling = ut.optscaling(FTm, b)
b = b / scaling
#ut.plotim3(mask)
ut.plotim3(np.absolute(FTm.backward(b))) #undersampled imag
#parameters
xpar = np.zeros((nx, ny, 3), np.complex128)
#xpar[:,:,0] = 10*np.ones((nx,ny))
#ut.plotim3(np.absolute(xpar),[3,-1])
# IDEAL and FFT jointly
IDEAL = idealc.IDEAL_opt2(TE, fat_freq_arr,
fat_rel_amp) #fat_freq_arr , fat_rel_amp
Aideal_ftm = opts.joint2operators(IDEAL, FTm) #(FTm,IDEAL)#
IDEAL.set_x(xpar) #should update in each gauss newton iteration
residual = IDEAL.residual(b, FTm)
#ut.plotim3(np.absolute(FTm.backward(residual)))
# wavelet and x+d_x
addx = idealc.x_add_dx()
addx.set_x(xpar)
#addx.set_w([1, 1, 0.0001])
dwt = opts.DWT2d(wavelet='haar', level=4)
Adwt_addx = opts.joint2operators(dwt, addx)
#do soft thresholding
#Nite = 200 #number of iterations
#step = 0.01 #step size
#th = 0.02 # theshold level
#do tv cs mri recon
Nite = 10 #number of iterations
step = 1 #step size
l1_r = 0.001
tv_r = 0.0001 # regularization term for tv term
rho = 1.0
ostep = 0.3
for i in range(20):
#wavelet L1 IST
# dxpar = solvers.IST_3( Aideal_ftm.forward, Aideal_ftm.backward,\
# Adwt_addx.backward, Adwt_addx.forward, residual, Nite, step, th )
#wavelet L1 ADMM
# dxpar = solvers.ADMM_l2Afxnb_l1Tfx( Aideal_ftm.forward, Aideal_ftm.backward, \
# Adwt_addx.backward, Adwt_addx.forward, residual, Nite, step, l1_r, rho, 200 )
# TV ADMM
# dxpar = solvers.ADMM_l2Afxnb_tvx( Aideal_ftm.forward, Aideal_ftm.backward, residual\
# , Nite, step, tv_r, rho, 15 )
dxpar = solvers.ADMM_l2Afxnb_tvTfx( Aideal_ftm.forward, Aideal_ftm.backward, \
addx.backward, addx.forward, residual, Nite, step, l1_r, rho, 200 )
# L2 CGD
# dxpar = pf.prox_l2_Afxnb_CGD2( Aideal_ftm.forward, Aideal_ftm.backward, residual, rho, Nite )
# dxpar = pf.prox_l2_Afxnb_CGD2( Aideal_ftm.forward, Aideal_ftm.backward, residual, Nite )
if i % 1 == 0:
ut.plotim3(np.absolute(xpar + ostep * dxpar)[..., 0:2], bar=1)
ut.plotim3(b0_gain * np.real(xpar + ostep * dxpar)[..., 2], bar=1)
ut.plotim3(np.imag(xpar + ostep * dxpar)[..., 2], bar=1)
xpar = xpar + ostep * dxpar #.astype(np.float64)
IDEAL.set_x(xpar) #should update in each gauss newton iteration
residual = IDEAL.residual(b, FTm)
addx.set_x(xpar) #should update in each gauss newton iteration
sio.savemat('data/kellman_data/xpar.mat', {'xpar': xpar})
ut.plotim3(np.absolute(xpar)[..., 0:2], bar=1)
def test():
ft = opts.FFTnd()
mat_contents = sio.loadmat(
'/working/larson/UTE_GRE_shuffling_recon/brain_mt_recon_20160919/brain_3dMRI_32ch.mat'
)
x = mat_contents["DATA"]
#ut.plotim3(np.absolute(x[:,:,:,0]))
im = ft.backward(x)
#ut.plotim3(np.absolute(im[:,:,im.shape[2]//2,:]))
#get shape
nx, ny, nz, nc = x.shape
#crop k-space
xcrop = ut.crop3d(x, 12)
if 1: #do espirit
Vim, sim = espirit_3d(xcrop, x.shape, 150, hkwin_shape = (12,12,12),\
pad_before_espirit = 0, pad_fact = 2)
#coil map
#ut.plotim3(np.absolute(Vim[:,:,im.shape[2]//2,:]),bar = 1)
#ut.plotim3(np.absolute(sim),bar = 1)
#create espirit operator
esp = opts.espirit(Vim)
#esp.save('../save_data/espirit_data_3d.mat')
esp.save(
'/working/larson/UTE_GRE_shuffling_recon/python_test/save_data/espirit_data_3d.mat'
)
else:
esp = opts.espirit()
#esp.restore('../save_data/espirit_data_3d.mat')
esp.restore(
'/working/larson/UTE_GRE_shuffling_recon/python_test/save_data/espirit_data_3d.mat'
)
#create mask
mask = ut.mask3d(nx, ny, nz, [15, 15, 0])
#FTm = opts.FFTnd_kmask(mask)
FTm = opts.FFTWnd_kmask(mask, threads=5)
#ut.plotim1(np.absolute(mask))#plot the mask
Aopt = opts.joint2operators(esp, FTm)
#create image
im = FTm.backward(x)
#ut.plotim3(np.absolute(im[:,:,:]))
#wavelet operator
dwt = opts.DWTnd(wavelet='haar', level=4)
# undersampling in k-space
b = FTm.forward(im)
scaling = ut.optscaling(FTm, b)
b = b / scaling
#ut.plotim3(np.absolute(Aopt.backward(b))) #undersampled imag
#do tv cs mri recon
Nite = 20 #number of iterations
step = 0.5 #step size
tv_r = 0.002 # regularization term for tv term
rho = 1.0
#xopt = solvers.ADMM_l2Afxnb_tvx( Aopt.forward, Aopt.backward, b, Nite, step, tv_r, rho )
xopt = solvers.ADMM_l2Afxnb_l1Tfx(Aopt.forward, Aopt.backward,
dwt.backward, dwt.forward, b, Nite, step,
tv_r, rho)
#do wavelet l1 soft thresholding
#Nite = 50 #number of iterations
#step = 1 #step size
#th = 0.4 # theshold level
#xopt = solvers.FIST_3( Aopt.forward, Aopt.backward, dwt.backward, dwt.forward, b, Nite, step, th )
ut.plotim3(np.absolute(xopt[:, :, :]))
def test():
# read rf and tr arrays from mat file
mat_contents = sio.loadmat(pathdat +
'mrf_t1t2b0pd_mrf_randphasecyc_traintest.mat')
far = np.array(mat_contents["rf"].astype(np.complex128).squeeze())
trr = np.array(mat_contents["trr"].astype(np.float64).squeeze())
# input MRF time courses
mat_contents2 = sio.loadmat(pathdat + 'datax1.mat')
data_x = np.array(mat_contents2["datax1"]).astype(np.float64)
# prepare for sequence simulation, y->x_hat
Nk = far.shape[0]
Nexample = data_x.shape[0]
ti = 10 #ms
M0 = np.array([0.0, 0.0, 1.0]).astype(np.float64)
#image size
nx = 217
ny = 181
# mask in ksp
mask = ut.mask3d(nx, ny, Nk, [15, 15, 0], 0.4)
#FTm = opts.FFT2d_kmask(mask)
FTm = cuopts.FFT2d_cuda_kmask(mask)
#intial timing
timing = utc.timing()
# x real and imag parts should be combined
data_x_c = np.zeros((Nexample, Nk), np.complex128)
data_x_c = ssmrf.seqdata_realimag_2complex(data_x)
#could do mask M here, data_x_c = M*data_x_c
data_x_c = mask_ksp3d(nx, ny, Nk, FTm, data_x_c)
data_x0_c = data_x_c
data_x_c_pre = data_x_c
data_x_acc = ssmrf.seqdata_complex_2realimag(data_x_c)
# intial tmp data
test_outy0 = np.zeros((Nexample, 4), dtype=np.float64)
test_outy = np.zeros((Nexample, 4), dtype=np.float64)
step = 1.0
# for loop start here
for _ in range(40):
for _ in range(10):
ssmrf.batch_apply_tf_cuda(Nexample, ny, sess.run, x, data_x_acc,
y_conv, test_outy, keep_prob)
#cuda bloch simulation for each pixel, parameters-->x
T1r, T2r, dfr, PDr = ssmrf.set_par(test_outy)
#timing.start()
data_x_c = ssmrf.bloch_sim_batch_cuda(Nexample, 7 * ny, Nk, PDr,
T1r, T2r, dfr, M0, trr, far,
ti)
print(
np.linalg.norm(data_x_c -
ssmrf.seqdata_realimag_2complex(data_x_acc)))
print(np.linalg.norm(data_x_c))
data_x_acc = ssmrf.seqdata_complex_2realimag(data_x_c)
#timing.stop().display('Bloch sim in loop ')
#0.5* d||M*FT(x)-b||_2^2/dx = 2 * FT^-1(M*(FT(x) - b))
# lamda * d(x^2)/dx = 2*lambda*x
data_x_c = mask_ksp3d(nx, ny, Nk, FTm,
data_x_c - data_x0_c) #data_x_c-data_x0_c#
# gradient descent/update
data_x_acc = data_x_acc - step * ssmrf.seqdata_complex_2realimag(
data_x_c)
#data_x_c = data_x_c - mask_ksp3d(nx, ny, Nk, FTm, data_x_c) + data_x0_c
#data_x_acc = ssmrf.seqdata_complex_2realimag(data_x_c)
#data_x_c = (wavel1_data(nx, ny, ssmrf.seqdata_realimag_2complex(data_x_acc)))
# apply CNN model, x-->parameters
print('gradient 0.5* d||f(p)-b||_2^2/dp: %g' %
np.linalg.norm(data_x0_c - data_x_c))
#could do soft thresholding on test_outy here, i.e. test_outy = threshold(test_outy)
#test_outy = wavel1_par(nx, ny, test_outy) #
#test_outy = constraints( test_outy )
#test_outy = tv_par(nx, ny, test_outy)
#test_outy = constraints( test_outy )
sio.savemat(pathdat + 'cnn_cs_testouty.mat', {'test_outy': test_outy})