def test_find_neighbors_adaptive_res():
n_neighbors = 36
n_disp = 5
processes = 4
bounds = [60., 80., -5., 5.]
import glob
fnames = glob.glob('/n/fink1/ggreen/bayestar/output/l70/l70.*.h5')
los_coll = maptools.los_collection(fnames, bounds=bounds,
processes=processes)
nside, pix_idx = los_coll.nside, los_coll.pix_idx
print 'Finding neighbors ...'
neighbor_idx, neighbor_dist = find_neighbors(nside, pix_idx,
n_neighbors)
print 'Done.'
# Highlight a couple of nearest-neighbor sections
pix_val = np.zeros(pix_idx.size)
l = 0.025 * np.pi / 180.
print l
for k in xrange(n_disp):
idx = np.random.randint(pix_idx.size)
pix_val[idx] = 2
pix_val[neighbor_idx[idx, :]] = 1
d = neighbor_dist[idx, :]
print 'Center pixel: %d' % idx
print d
print np.exp(-(d/l)**2.)
print ''
nside_max, pix_idx_exp, pix_val_exp = maptools.reduce_to_single_res(pix_idx,
nside,
pix_val)
size = (2000, 2000)
img, bounds, xy_bounds = hputils.rasterize_map(pix_idx_exp, pix_val_exp,
nside_max, size)
# Plot nearest neighbors
import matplotlib.pyplot as plt
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.imshow(img, extent=bounds, origin='lower', aspect='auto',
interpolation='nearest')
plt.show()
def vol_render_mw(fnames, method='median', bounds=None, processes=1, max_samples=None): los_coll = maptools.los_collection(fnames, bounds=bounds, processes=processes, max_samples=max_samples) x, y, z = los_coll.get_xyz(oversample=3, max_points=20) rho = los_coll.get_density(method=method, oversample=3, max_points=20) print '# of points: %d' % (rho.size) x += 0.1 * np.random.random(x.shape) y += 0.1 * np.random.random(y.shape) z += 0.1 * np.random.random(z.shape) vol = vol_render_pts(x, y, z, rho, show_pts=True) set_opacity_transfer(vol, np.max(rho), 0.0001) set_bw_colorscheme(vol, 0., 1.) mlab.view(focalpoint=(0., 0., 0.)) print 'showing...' mlab.show()
def __init__(self, fnames, bounds=None,
processes=1,
n_neighbors=32,
corr_length_core=0.25,
corr_length_tail=1.,
max_corr=1.,
tail_weight=0.,
dist_floor=0.1):
self.n_neighbors = n_neighbors
self.corr_length_core = corr_length_core
self.corr_length_tail = corr_length_tail
self.max_corr = max_corr
self.tail_weight = tail_weight
self.los_coll = maptools.los_collection(fnames, bounds=bounds,
processes=processes)
self.nside = self.los_coll.nside
self.pix_idx = self.los_coll.pix_idx
print 'Finding neighbors ...'
self.neighbor_idx, self.neighbor_ang_dist = find_neighbors(self.nside,
self.pix_idx,
self.n_neighbors)
# Determine difference from priors
self.delta = np.log(self.los_coll.los_delta_EBV)
self.n_pix, self.n_samples, self.n_slices = self.delta.shape
print 'Calculating priors ...'
ln_Delta_EBV_prior = get_prior_ln_Delta_EBV(self.nside,
self.pix_idx,
n_regions=self.n_slices-1)
#print ''
#print 'priors:'
#print ln_Delta_EBV_prior[0, :]
for n in xrange(self.n_samples):
self.delta[:, n, :] -= ln_Delta_EBV_prior
self.delta /= 1.5 # Standardize to units of the std. dev. on the priors
#print ''
#print 'delta:'
#print self.delta
# Distance in pc to each bin
slice_dist = np.power(10., self.los_coll.los_mu_anchor/5. + 1.)
slice_dist = np.hstack([[0.], slice_dist])
self.bin_dist = 0.5 * (slice_dist[:-1] + slice_dist[1:])
# Determine physical distance of each voxel to its neighbors
# shape = (pix, neighbor, slice)
print 'Determining neighbor correlation weights ...'
self.neighbor_dist = np.einsum('ij,k->ijk', self.neighbor_ang_dist, self.bin_dist)
self.neighbor_dist = np.sqrt(self.neighbor_dist * self.neighbor_dist + dist_floor * dist_floor)
self.neighbor_corr = self.corr_of_dist(self.neighbor_dist)
#self.neighbor_corr = self.hard_sphere_corr(self.neighbor_dist, self.corr_length_core)
#print ''
#print 'dist:'
#print self.bin_dist
#print ''
#print 'corr:'
#print self.neighbor_corr
#print ''
# Set initial state
print 'Randomizing initial state ...'
self.beta = 1.
self.chain = []
self.randomize()
self.log_state()
self.update_order = np.arange(self.n_pix)
