cpa_space.calc_T(pat,pts = src, mysign=1,out=transformed_gpu,
**params_flow_int)
toc = time.clock()
print "time (keeps result in gpu)",(toc-tic)/M
if M==1:
print """
Note that for running it only once, we pay some overhead --
probably because of the addiitonal krnl for bdry checks.
This levels out when we make many runs"""
if not np.allclose(transformed_gpu.get(),transformed):
raise ValueError
transformed = transformed_gpu.get()
if TF.plot:
v_dense.gpu2cpu()
transformed.gpu2cpu()
Visualize.simple(x_dense,v_dense,interval,src,transformed,subplot_layout=[2,2])
if computer.has_good_gpu_card:
raw_input('raw_input:')
print "Bye now"
mysign=1,
out=transformed_gpu,
**params_flow_int)
toc = time.clock()
print "time (keeps result in gpu)", (toc - tic) / M
if M == 1:
print """
Note that for running it only once, we pay some overhead --
probably because of the addiitonal krnl for bdry checks.
This levels out when we make many runs"""
if not np.allclose(transformed_gpu.get(), transformed):
raise ValueError
transformed = transformed_gpu.get()
if TF.plot:
v_dense.gpu2cpu()
transformed.gpu2cpu()
Visualize.simple(x_dense,
v_dense,
interval,
src,
transformed,
subplot_layout=[2, 2])
if computer.has_good_gpu_card:
raw_input('raw_input:')
print "Bye now"
def example(base=[5],
scale_spatial=100,
nLevels=2,
zero_v_across_bdry=[1],
use_local_basis=True):
nPtsDense = 10000
tw = TransformWrapper(nCols=100,
nLevels=nLevels,
base=base,
scale_spatial=scale_spatial,
nPtsDense=nPtsDense,
zero_v_across_bdry=zero_v_across_bdry)
print_iterable(tw.ms.L_cpa_space)
seed=0
np.random.seed(seed)
for level in range(tw.ms.nLevels):
cpa_space = tw.ms.L_cpa_space[level]
Avees = cpa_space.Avees
velTess = cpa_space.zeros_velTess()
if use_local_basis:
if 0:
tw.sample_gaussian_velTess(level,Avees,velTess,mu=None)
Avees*=0.001
velTess*=0.001
else:
if not zero_v_across_bdry[0]:
velTess[:]=10*np.random.standard_normal(velTess.shape)
cpa_space.velTess2Avees(velTess=velTess,Avees=Avees)
else:
velTess[1:-1]=10*np.random.standard_normal(velTess[1:-1].shape)
cpa_space.velTess2Avees(velTess=velTess,Avees=Avees)
cpa_space.velTess2Avees(velTess=velTess,Avees=Avees)
else:
theta= cpa_space.get_zeros_theta()
tw.sample_gaussian(level,Avees,theta,mu=None)
# theta/=10
cpa_space.theta2Avees(theta=theta,Avees=Avees)
# This step is important and must be done
# before are trying to "use" the new values of
# the (vectorized) A's.
tw.update_pat_from_Avees(Avees,level)
pts_src = tw.x_dense
tw.calc_v(level=level,pts=pts_src,v=tw.v_dense)
tw.v_dense.gpu2cpu()
pts_fwd = CpuGpuArray.zeros_like(pts_src) # Create a buffer for the output
tw.calc_T_fwd(pts_src,pts_fwd,level=level)
pts_fwd.gpu2cpu()
pts_inv = CpuGpuArray.zeros_like(pts_src) # Create a buffer for the output
tw.calc_T_inv(pts_src,pts_inv,level=level)
pts_inv.gpu2cpu()
plt.figure(level)
plt.clf()
interval = pts_src.cpu # interval doesn't have to be pts_src.cpu
Visualize.simple(tw.x_dense,tw.v_dense,interval,pts_src,
transformed_fwd=pts_fwd,transformed_inv=pts_inv,
cpa_space=cpa_space)
return tw
def example(base=[5],
scale_spatial=100,
nLevels=2,
zero_v_across_bdry=[1],
use_local_basis=True):
nPtsDense = 10000
tw = TransformWrapper(nCols=100,
nLevels=nLevels,
base=base,
scale_spatial=scale_spatial,
nPtsDense=nPtsDense,
zero_v_across_bdry=zero_v_across_bdry)
print_iterable(tw.ms.L_cpa_space)
seed = 0
np.random.seed(seed)
for level in range(tw.ms.nLevels):
cpa_space = tw.ms.L_cpa_space[level]
Avees = cpa_space.Avees
velTess = cpa_space.zeros_velTess()
if use_local_basis:
if 0:
tw.sample_gaussian_velTess(level, Avees, velTess, mu=None)
Avees *= 0.001
velTess *= 0.001
else:
if not zero_v_across_bdry[0]:
velTess[:] = 10 * np.random.standard_normal(velTess.shape)
cpa_space.velTess2Avees(velTess=velTess, Avees=Avees)
else:
velTess[1:-1] = 10 * np.random.standard_normal(
velTess[1:-1].shape)
cpa_space.velTess2Avees(velTess=velTess, Avees=Avees)
cpa_space.velTess2Avees(velTess=velTess, Avees=Avees)
else:
theta = cpa_space.get_zeros_theta()
tw.sample_gaussian(level, Avees, theta, mu=None)
# theta/=10
cpa_space.theta2Avees(theta=theta, Avees=Avees)
# This step is important and must be done
# before are trying to "use" the new values of
# the (vectorized) A's.
tw.update_pat_from_Avees(Avees, level)
pts_src = tw.x_dense
tw.calc_v(level=level, pts=pts_src, v=tw.v_dense)
tw.v_dense.gpu2cpu()
pts_fwd = CpuGpuArray.zeros_like(
pts_src) # Create a buffer for the output
tw.calc_T_fwd(pts_src, pts_fwd, level=level)
pts_fwd.gpu2cpu()
pts_inv = CpuGpuArray.zeros_like(
pts_src) # Create a buffer for the output
tw.calc_T_inv(pts_src, pts_inv, level=level)
pts_inv.gpu2cpu()
plt.figure(level)
plt.clf()
interval = pts_src.cpu # interval doesn't have to be pts_src.cpu
Visualize.simple(tw.x_dense,
tw.v_dense,
interval,
pts_src,
transformed_fwd=pts_fwd,
transformed_inv=pts_inv,
cpa_space=cpa_space)
return tw
