def find_nearest_neighbors(nside, pix_idx):
# TODO
# Determine mapping of pixel indices at highest resolutions to index
# in the array pix_idx
nside_max = np.max(nside)
pix_idx_highres = []
for n in np.unique(nside):
idx = (nside == n)
pix_idx
l = np.empty(pix_idx.size, dtype='f8')
b = np.empty(pix_idx.size, dtype='f8')
neighbor_idx = np.empty((8, pix_idx.size), dtype='i8')
neighbor_dist = np.empty((8, pix_idx.size), dtype='f8')
for n in np.unique(nside):
idx = (nside == n)
l[idx], b[idx] = hputils.pix2lb(n, pix_idx[idx], nest=True)
neighbor_idx[:, idx] = 1
def get_prior_ln_Delta_EBV(nside, pix_idx, n_regions=30):
# Find (l, b) for each pixel
l = np.empty(pix_idx.size, dtype='f8')
b = np.empty(pix_idx.size, dtype='f8')
for n in np.unique(nside):
idx = (nside == n)
l[idx], b[idx] = hputils.pix2lb(n, pix_idx[idx], nest=True)
# Determine priors in each pixel
gal_model = model.TGalacticModel()
ln_Delta_EBV = np.empty((pix_idx.size, n_regions+1), dtype='f8')
for i,(ll,bb) in enumerate(zip(l, b)):
ret = gal_model.EBV_prior(ll, bb, n_regions=n_regions)
ln_Delta_EBV[i, :] = ret[1]
return ln_Delta_EBV
def find_neighbors_naive(nside, pix_idx, n_neighbors):
'''
Find the neighbors of each pixel using a naive algorithm.
Each pixel is defined by a HEALPix nside and pixel index (in nested
order).
Returns two arrays:
neighbor_idx (n_pix, n_neighbors) Index of each neighbor in
the nside and pix_idx
arrays.
neighbor_dist (n_pix, n_neighbors) Distance to each neighbor.
'''
# Determine (l, b) of all pixels
l = np.empty(pix_idx.size, dtype='f8')
b = np.empty(pix_idx.size, dtype='f8')
nside_unique = np.unique(nside)
for n in nside_unique:
idx = (nside == n)
l[idx], b[idx] = hputils.pix2lb(n, pix_idx[idx], nest=True)
# Determine distances between all pixel pairs
dist = gc_dist_all(l, b)
# Determine closest neighbors
sort_idx = np.argsort(dist, axis=1)
neighbor_idx = sort_idx[:, 1:n_neighbors+1]
neighbor_dist = np.sort(dist, axis=1)[:, 1:n_neighbors+1]
return neighbor_idx, neighbor_dist
def load_output_file_unified(f, bounds=None,
max_samples=None,
load_stacked_pdfs=False):
# Pixel locations
dset = f['locations']
nside = dset['nside'][:]
pix_idx = dset['healpix_index'][:]
cloud_mask = dset['cloud_mask'][:].astype(np.bool)
los_mask = dset['piecewise_mask'][:].astype(np.bool)
n_stars = dset['n_stars'][:]
# Cloud model
tmp_samples, cloud_lnp, cloud_GR = unpack_dset(f['cloud'],
max_samples=max_samples)
n_clouds = tmp_samples.shape[2] / 2
cloud_mu = tmp_samples[:, :, :n_clouds]
cloud_delta_EBV = tmp_samples[:, :, n_clouds:]
# Piecewise-linear model
los_EBV, los_lnp, los_GR = unpack_dset(f['piecewise'],
max_samples=max_samples)
DM_min = float(f['piecewise'].attrs['DM_min'])
DM_max = float(f['piecewise'].attrs['DM_max'])
EBV_min, EBV_max = 0., 5.
# Stacked pdfs
star_stack = None
if load_stacked_pdfs:
dset = f['stacked_pdfs']
star_stack = dset[:]
EBV_min = float(dset.attrs['EBV_min'])
EBV_max = float(dset.attrs['EBV_max'])
# Filter out-of-bounds pixels
if bounds != None:
l = np.empty(nside.size, dtype='f8')
b = np.empty(nside.size, dtype='f8')
for n in np.unique(nside):
idx = (nside == n)
l[idx], b[idx] = hputils.pix2lb(n, pix_idx[idx], nest=True)
idx = ( (l >= bounds[0]) & (l <= bounds[1])
& (b >= bounds[2]) & (b <= bounds[3]) )
nside = nside[idx]
pix_idx = pix_idx[idx]
cloud_mask = cloud_mask[idx]
los_mask = los_mask[idx]
n_stars = n_stars[idx]
cloud_mu = cloud_mu[idx]
cloud_delta_EBV = cloud_delta_EBV[idx]
cloud_GR = cloud_GR[idx]
cloud_lnp = cloud_lnp[idx]
los_EBV = los_EBV[idx]
los_GR = los_GR[idx]
los_lnp = los_lnp[idx]
if load_stacked_pdfs:
star_stack = star_stack[idx]
# Return
if len(pix_idx) == 0:
return None
pix_info = (pix_idx, nside, cloud_mask, los_mask, n_stars)
# Cloud information
cloud_info = None
if np.sum(cloud_mask) > 0:
cloud_info = (cloud_mu, cloud_delta_EBV, cloud_lnp, cloud_GR)
# Piecewise-linear model information
los_info = None
if np.sum(los_mask) > 0:
los_info = (los_EBV, los_lnp, los_GR)
# Limits on DM and E(B-V) (for l.o.s. fits and stacked surfaces)
DM_EBV_lim = (DM_min, DM_max, EBV_min, EBV_max)
return pix_info, cloud_info, los_info, star_stack, DM_EBV_lim
def dust_priors():
img_fname = pan1 + 'paper-plots/dust_priors.svg'
dpi = 400
figsize = (9, 4)
img_shape = (4000, 2000)
nside = 16
n_regions = 30
sigma_bin = 1.4
gal_model = model.TGalacticModel()
n_pix = hp.pixelfunc.nside2npix(nside)
pix_idx = np.arange(n_pix)
l, b = hputils.pix2lb(nside, pix_idx)
log_Delta_EBV = np.empty((n_pix, n_regions+1), dtype='f8')
# Determine log(Delta EBV) in each pixel and bin
for i,ll,bb in zip(pix_idx, l, b):
tmp = gal_model.EBV_prior(ll, bb, n_regions=n_regions, sigma_bin=1.4)
log_Delta_EBV[i,:] = tmp[1][:]
scatter = sigma_bin * np.random.normal(size=log_Delta_EBV.shape)
EBV_rand = np.sum(np.exp(log_Delta_EBV + scatter), axis=1)
EBV_smooth = np.sum(np.exp(log_Delta_EBV), axis=1) * np.exp(0.5*sigma_bin**2.)
# Load SFD dust map
EBV_SFD = model.get_SFD_map(nside=nside)
# Rasterize maps
proj = hputils.Hammer_projection()
nside_arr = nside * np.ones(n_pix, dtype='i8')
rasterizer = hputils.MapRasterizer(nside_arr, pix_idx, img_shape,
proj=proj)
bounds = rasterizer.get_lb_bounds()
log_SFD = np.log10(EBV_SFD)
log_rand = np.log10(EBV_rand)
log_smooth = np.log10(EBV_smooth)
log_SFD_diff = log_SFD - log_smooth
log_rand_diff = log_rand - log_smooth
print np.percentile(log_SFD_diff, [5., 10., 50., 90., 95.])
vmin_top = min([np.min(log_SFD), np.min(log_rand)])
vmax_top = max([np.max(log_SFD), np.max(log_rand)])
vmax_bot = max([np.max(np.abs(log_SFD_diff)), np.max(np.abs(log_rand_diff))])
vmin_bot = -vmax_bot
img_SFD = rasterizer(log_SFD)
img_rand = rasterizer(log_rand)
img_SFD_diff = rasterizer(log_SFD_diff)
img_rand_diff = rasterizer(log_rand_diff)
# Rasterize regions to shade on maps
shade_fn = lambda l, b: hputils.lb_in_bounds(l, b, [-60., 60., -5., 5.]).astype('f8')
img_shade = rasterizer.rasterize_func(shade_fn)
nside_hidpi = 256
n_pix_hidpi = hp.pixelfunc.nside2npix(nside_hidpi)
pix_idx_hidpi = np.arange(n_pix_hidpi)
nside_arr_hidpi = nside_hidpi * np.ones(n_pix_hidpi, dtype='i8')
rasterizer_hidpi = hputils.MapRasterizer(nside_arr_hidpi, pix_idx_hidpi,
img_shape, proj=proj)
bounds_hidpi = rasterizer.get_lb_bounds()
l, b = hputils.pix2lb(nside_hidpi, pix_idx_hidpi)
clipped = np.empty(n_pix_hidpi, dtype='f8')
# Determine log(Delta EBV) in each pixel and bin
for i,(ll,bb) in enumerate(zip(l, b)):
tmp = gal_model.EBV_prior(ll, bb, n_regions=n_regions, sigma_bin=1.4)
clipped[i] = np.max(tmp[1][:]) >= -4. - 1.e-5
#clipped = (np.max(log_Delta_EBV, axis=1) >= -4. - 1.e-5).astype('f8')
img_clipped = rasterizer_hidpi(clipped)
# Plot in 2x2 grid
fig = plt.figure(figsize=figsize, dpi=dpi)
ax = [fig.add_subplot(2,2,i+1) for i in xrange(4)]
im_11 = ax[0].imshow(img_SFD.T, extent=bounds, aspect='auto',
origin='lower', interpolation='nearest',
vmin=vmin_top, vmax=vmax_top,
rasterized=True)
im_12 = ax[1].imshow(img_rand.T, extent=bounds, aspect='auto',
origin='lower', interpolation='nearest',
vmin=vmin_top, vmax=vmax_top,
rasterized=True)
im_21 = ax[2].imshow(img_SFD_diff.T, extent=bounds, aspect='auto',
origin='lower', interpolation='nearest',
vmin=vmin_bot, vmax=vmax_bot,
cmap='bwr',
rasterized=True)
im_22 = ax[3].imshow(img_rand_diff.T, extent=bounds, aspect='auto',
origin='lower', interpolation='nearest',
vmin=vmin_bot, vmax=vmax_bot,
cmap='bwr',
rasterized=True)
for i in xrange(4):
#ax[i].imshow(img_shade.T, extent=bounds, aspect='auto',
# origin='lower', interpolation='nearest',
# vmin=0., vmax=1., cmap='binary',
# alpha=0.25)
ax[i].imshow(img_clipped.T, extent=bounds_hidpi, aspect='auto',
origin='lower', interpolation='nearest',
vmin=0., vmax=1., cmap='binary',
alpha=0.25, rasterized=True)
fig.subplots_adjust(left=0.02, right=0.82,
bottom=0.02, top=0.92,
hspace=0.02, wspace=0.02)
# Remove axis frames
for a in ax:
a.axis('off')
# Color bars
x0,y0,w,h = ax[1].get_position().bounds
cax_top = fig.add_axes([x0 + w + 0.01, y0, 0.03, h])
x0,y0,w,h = ax[3].get_position().bounds
cax_bot = fig.add_axes([x0 + w + 0.01, y0, 0.03, h])
cbar_top = fig.colorbar(im_12, cax=cax_top)
cbar_bot = fig.colorbar(im_22, cax=cax_bot)
# Labels
x0,y0,w,h = ax[0].get_position().bounds
fig.text(x0+0.5*w, y0+h+0.01, r'$\mathrm{SFD}$',
fontsize=16, ha='center', va='bottom')
x0,y0,w,h = ax[1].get_position().bounds
fig.text(x0+0.5*w, y0+h+0.01, r'$\mathrm{Priors}$',
fontsize=16, ha='center', va='bottom')
cbar_top.set_label(r'$\log_{10} \mathrm{E} \left( B - V \right)$',
fontsize=14)
cbar_bot.set_label(r'$\log_{10} \left[ \frac{ \mathrm{E} \left( B - V \right) }{ \left< \mathrm{E} \left( B - V \right) \right> } \right]$',
fontsize=14)
cbar_top.locator = ticker.MaxNLocator(nbins=6)
cbar_top.update_ticks()
cbar_bot.locator = ticker.MaxNLocator(nbins=6)
cbar_bot.update_ticks()
cbar_bot.solids.set_edgecolor('face') # Fix for matplotlib svg-rendering bug
cbar_top.solids.set_edgecolor('face') # Fix for matplotlib svg-rendering bug
fig.savefig(img_fname, dpi=dpi, bbox_inches='tight')
plt.show()
def find_neighbors(nside, pix_idx, n_neighbors):
'''
Find the neighbors of each pixel.
Each pixel is defined by a HEALPix nside and pixel index (in nested
order).
Returns two arrays:
neighbor_idx (n_pix, n_neighbors) Index of each neighbor in
the nside and pix_idx
arrays.
neighbor_dist (n_pix, n_neighbors) Distance to each neighbor.
'''
# Determine (l, b) and which downsampled pixel
# each (nside, pix_idx) combo belongs to
nside_rough = np.min(nside)
if nside_rough != 1:
nside_rough /= 2
l = np.empty(pix_idx.size, dtype='f8')
b = np.empty(pix_idx.size, dtype='f8')
pix_idx_rough = np.empty(pix_idx.size, dtype='i8')
nside_unique = np.unique(nside)
for n in nside_unique:
idx = (nside == n)
factor = (n / nside_rough)**2
pix_idx_rough[idx] = pix_idx[idx] / factor
l[idx], b[idx] = hputils.pix2lb(n, pix_idx[idx], nest=True)
# For each downsampled pixel, determine nearest neighbors of all subpixels
neighbor_idx = -np.ones((pix_idx.size, n_neighbors), dtype='i8')
neighbor_dist = np.inf * np.ones((pix_idx.size, n_neighbors), dtype='f8')
for i_rough in np.unique(pix_idx_rough):
rough_neighbors = hp.get_all_neighbours(nside_rough, i_rough, nest=True)
idx_centers = np.argwhere(pix_idx_rough == i_rough)[:,0]
tmp = [np.argwhere(pix_idx_rough == i)[:,0] for i in rough_neighbors]
tmp.append(idx_centers)
idx_search = np.hstack(tmp)
dist = gc_dist(l[idx_centers], b[idx_centers],
l[idx_search], b[idx_search])
tmp = np.argsort(dist, axis=1)[:, 1:n_neighbors+1]
fill = idx_search[tmp]
neighbor_idx[idx_centers, :fill.shape[1]] = fill
fill = np.sort(dist, axis=1)[:, 1:n_neighbors+1]
neighbor_dist[idx_centers, :fill.shape[1]] = fill
return neighbor_idx, neighbor_dist
