def input_skymap(order1, d_order, fraction): """Construct a test multi-resolution sky map, with values that are proportional to the NESTED pixel index. To make the test more interesting by mixing together multiple resolutions, part of the sky map is refined to a higher order. Parameters ---------- order1 : int The HEALPix resolution order. d_order : int The increase in orer for part of the sky map. fraction : float The fraction of the original pixels to refine. """ order2 = order1 + d_order npix1 = hp.nside2npix(hp.order2nside(order1)) npix2 = hp.nside2npix(hp.order2nside(order2)) ipix1 = np.arange(npix1) ipix2 = np.arange(npix2) # Create a random sky map. area = hp.nside2pixarea(hp.order2nside(order1)) probdensity = np.random.uniform(0, 1, npix1) prob = probdensity * area normalization = prob.sum() prob /= normalization probdensity /= normalization distmean = np.random.uniform(100, 110, npix1) diststd = np.random.uniform(0, 1 / np.sqrt(3) - 0.1, npix1) * distmean distmu, distsigma, distnorm = moments_to_parameters(distmean, diststd) assert np.all(np.isfinite(distmu)) data1 = table.Table({ 'UNIQ': moc.nest2uniq(order1, ipix1), 'PROBDENSITY': probdensity, 'DISTMU': distmu, 'DISTSIGMA': distsigma, 'DISTNORM': distnorm }) # Add some upsampled pixels. data2 = table.Table(np.repeat(data1, npix2 // npix1)) data2['UNIQ'] = moc.nest2uniq(order2, ipix2) n = int(npix1 * (1 - fraction)) result = table.vstack((data1[:n], data2[n * npix2 // npix1:])) # Add marginal distance mean and standard deviation. rbar = (prob * distmean).sum() r2bar = (prob * (np.square(diststd) + np.square(distmean))).sum() result.meta['distmean'] = rbar result.meta['diststd'] = np.sqrt(r2bar - np.square(rbar)) return result
def test_rasterize_downsample(order_in, d_order_in, fraction_in, order_out): npix_in = hp.nside2npix(hp.order2nside(order_in)) npix_out = hp.nside2npix(hp.order2nside(order_out)) skymap_in = input_skymap(order_in, d_order_in, fraction_in) skymap_out = moc.rasterize(skymap_in, order_out) assert len(skymap_out) == npix_out reps = npix_in // npix_out expected = np.mean(np.arange(npix_in).reshape(-1, reps), axis=1) np.testing.assert_array_equal(skymap_out['VALUE'], expected)
def test_rasterize_upsample(order_in, d_order_in, fraction_in, order_out): npix_in = hp.nside2npix(hp.order2nside(order_in)) npix_out = hp.nside2npix(hp.order2nside(order_out)) skymap_in = input_skymap(order_in, d_order_in, fraction_in) skymap_out = moc.rasterize(skymap_in, order_out) assert len(skymap_out) == npix_out ipix = np.arange(npix_in) reps = npix_out // npix_in for i in range(reps): np.testing.assert_array_equal(skymap_out['VALUE'][i::reps], ipix)
def make_healpix_map_for_energy_band(energy_band, order): log.info(f'Making HEALPix map for energy band: {energy_band} and order: {order}') # Select events in energy band table = Table.read('input_data/fermi_hgps_events_selected.fits.gz', hdu=1) energy = table['ENERGY'].quantity.to('GeV').value mask = (energy_band['min'] <= energy) & (energy < energy_band['max']) table = table[mask] log.info(f'Number of events: {len(table)}') # Bin the events into a HEALPix counts map nside = hp.order2nside(order + shift_order) ipix = hp.ang2pix( nside=nside, nest=True, lonlat=True, theta=table['L'], phi=table['B'], ) npix = hp.nside2npix(nside) log.debug(f'Number of pixels: {npix}') resolution = np.rad2deg(hp.nside2resol(nside)) log.debug(f'Pixel resolution: {resolution} deg') image = np.bincount(ipix, minlength=npix) image = image.astype('float32') # TODO: smoothing the HEALPix map with default setting is very slow. # Maybe chunk the data into local WCS maps and then stitch back together? # For now: no smoothing # image = hp.smoothing(image, sigma=np.deg2rad(0.1)) path = DATA_DIR / 'maps' / 'Fermi10GeV_healpix_maps' / 'energy_{min}_{max}.fits.gz'.format_map(energy_band) path.parent.mkdir(exist_ok=True, parents=True) log.info(f'Writing {path}') hp.write_map(str(path), image, coord='G', nest=True)
def _reconstruct_nested_breadthfirst(m, extra): m = np.asarray(m) max_npix = len(m) max_nside = hp.npix2nside(max_npix) max_order = hp.nside2order(max_nside) seen = np.zeros(max_npix, dtype=bool) for order in range(max_order + 1): nside = hp.order2nside(order) npix = hp.nside2npix(nside) skip = max_npix // npix if skip > 1: b = m.reshape(-1, skip) a = b[:, 0].reshape(-1, 1) b = b[:, 1:] aseen = seen.reshape(-1, skip) eq = ((a == b) | ((a != a) & (b != b))).all(1) & (~aseen).all(1) else: eq = ~seen for ipix in np.flatnonzero(eq): ipix0 = ipix * skip ipix1 = (ipix + 1) * skip seen[ipix0:ipix1] = True if extra: yield _HEALPixTreeVisitExtra( nside, max_nside, ipix, ipix0, ipix1, m[ipix0]) else: yield _HEALPixTreeVisit(nside, ipix)
def testCcdVisitHpix8(self): filters = "ugrizy" num_ccd_visits = 10 cell_id = 2500 chunker = Chunker(0, num_stripes, num_substripes) ccdVisitGenerator = columns.CcdVisitGenerator(chunker, num_ccd_visits, filters=filters) hpix8_values = ccdVisitGenerator._find_hpix8_in_cell(cell_id) print(hpix8_values) self.assertTrue(len(hpix8_values) > 0) nside = healpy.order2nside(8) chunks = [ chunker.locate(healpy.pix2ang(nside, pixel, nest=True, lonlat=True)) for pixel in hpix8_values ] hpix_centers_in_chunk = np.array(chunks) == cell_id # Some of the hpix centers will be outside of the chunk area, and that seems ok. # The test is to confirm that we get enough of them with centers inside the # chunk to confirm that the code is working. print(chunks) self.assertGreaterEqual( np.sum(hpix_centers_in_chunk) / float(len(hpix8_values)), 0.5)
def _reconstruct_nested_breadthfirst(m, extra): max_npix = len(m) max_nside = hp.npix2nside(max_npix) max_order = hp.nside2order(max_nside) seen = np.zeros(max_npix, dtype=bool) for order in range(max_order + 1): nside = hp.order2nside(order) npix = hp.nside2npix(nside) skip = max_npix // npix if skip > 1: b = m.reshape(-1, skip) a = b[:, 0].reshape(-1, 1) b = b[:, 1:] aseen = seen.reshape(-1, skip) eq = ((a == b) | ((a != a) & (b != b))).all(1) & (~aseen).all(1) else: eq = ~seen for ipix in np.flatnonzero(eq): ipix0 = ipix * skip ipix1 = (ipix + 1) * skip seen[ipix0:ipix1] = True if extra: yield _HEALPixTreeVisitExtra( nside, max_nside, ipix, ipix0, ipix1, m[ipix0]) else: yield _HEALPixTreeVisit(nside, ipix)
def best_order_for(pixel_scale): d = pixel_scale best_o = 1. / 2. * math.log(math.pi / (3. * d * d)) / math.log(2) o = int(best_o + 1) logging.info('pixel scale (arcsec): {:.4f} -> {:.4f}'.format( rad2arcsec(pixel_scale), rad2arcsec(healpy.nside2resol(healpy.order2nside(o))))) return o
def mapped_tile(self, out_file, tile_order, tile_index, pixel_order): tile_size_order = pixel_order - tile_order tile_nside = healpy.order2nside(tile_size_order) pixels_per_tile = tile_nside * tile_nside pixel_nside = healpy.order2nside(pixel_order) pixel_index_start = tile_index << (2 * tile_size_order) pixel_indices = numpy.arange( pixel_index_start, pixel_index_start + pixels_per_tile)[hips.nest2ring(tile_nside)] THETA, PHI = healpy.pix2ang(pixel_nside, pixel_indices, nest=True) RA = numpy.rad2deg(PHI) DEC = numpy.rad2deg(numpy.pi / 2 - THETA) X, Y = self.wcs.wcs_world2pix(RA, DEC, 0) mapped = interpolate.linear(self.hdu.data, X, Y, dtype=numpy.float32).reshape( (tile_nside, tile_nside)) return mapped
def input_skymap(order1, d_order, fraction): """Construct a test multi-resolution sky map, with values that are proportional to the NESTED pixel index. To make the test more interesting by mixing together multiple resolutions, part of the sky map is refined to a higher order. Parameters ---------- order1 : int The HEALPix resolution order. d_order : int The increase in orer for part of the sky map. fraction : float The fraction of the original pixels to refine. """ order2 = order1 + d_order npix1 = hp.nside2npix(hp.order2nside(order1)) npix2 = hp.nside2npix(hp.order2nside(order2)) ipix1 = np.arange(npix1) ipix2 = np.arange(npix2) data1 = table.Table({ 'UNIQ': moc.nest2uniq(order1, ipix1), 'VALUE': ipix1.astype(float), 'VALUE2': np.pi * ipix1.astype(float) }) data2 = table.Table({ 'UNIQ': moc.nest2uniq(order2, ipix2), 'VALUE': np.repeat(ipix1, npix2 // npix1).astype(float), 'VALUE2': np.pi * np.repeat(ipix1, npix2 // npix1).astype(float) }) n = int(npix1 * (1 - fraction)) return table.vstack((data1[:n], data2[n * npix2 // npix1:]))
def apply_hpxmoc(self, lon=None, lat=None): try: moc = MOC() moc.read(self.filename, filetype="fits") except: raise Exception(f"Unable to find/open Healpix MOC file: {self.filename}") # get the moc nside at the max resolution of the MOC nside = hp.order2nside(moc.order) #get the healpix pixels indices of the targets at the max resolution idx = hp.ang2pix(nside, lon, lat, lonlat=True, nest=True ) m_mask = moc.contains(idx) return m_mask
def warp_source_files(fnames, work_dir, pixel_order, tile_order): tile_nside = healpy.order2nside(tile_order) for fname in fnames: with afits.open(fname) as hdul: src = Source(find_image_hdu(hdul)) tile_indices = healpy.query_polygon(tile_nside, ad2xyz(*src.polygon).T, nest=True) logging.info('{} tiles'.format(len(tile_indices))) def warp(tile_index): logging.info('warping {}:{}...'.format(fname, tile_index)) src.warp(work_dir, tile_order, tile_index, pixel_order) parallel.map(warp, tile_indices)
def get_pixlist_for_fields(fields, radius=2, level=12): import healpy pixlist = [] radius = np.radians(radius) nside = healpy.order2nside(level) for ra, dec in fields: vec = healpy.ang2vec(ra, dec, lonlat=True) pixlist.append( healpy.query_disc(nside, vec, radius, inclusive=True, fact=4, nest=True)) pixlist = np.hstack(pixlist) return np.unique(pixlist)
def get_pixlist(level=7, max_declination=30., plot=False): """ """ nside = hp.order2nside(level) npix = hp.nside2npix(nside) ipix = np.arange(npix) theta, phi = hp.pix2ang(nside, ipix, nest=True) # dec = -theta * 180. / np.pi + 90. dec = -np.degrees(theta - np.pi / 2.) idx = dec < 35. if plot: m = np.zeros(npix) m[idx] = 1. hp.mollview(m, nest=True) return ipix[idx]
def _find_hpix8_in_cell(self, chunk_id): chunk_box = self.chunker.getChunkBounds(chunk_id) grid_size = 9 lon_arr = np.linspace(chunk_box.getLon().getA().asDegrees(), chunk_box.getLon().getB().asDegrees(), grid_size) lat_arr = np.linspace(chunk_box.getLat().getA().asDegrees(), chunk_box.getLat().getB().asDegrees(), grid_size) xx, yy = np.meshgrid(lon_arr[1::-2], lat_arr[1::-2]) nside = healpy.order2nside(8) trial_healpix = healpy.ang2pix(nside, lon_arr, lat_arr, nest=True, lonlat=True) return list(set(trial_healpix))
def fetch(survey, url, geocentric_coords, order, **kwargs): """ Fetch tile from a HiPS server. Performs conversion as well. Parameters ---------- survey : str Suvery where HiPS is hosted url : str URL excluding survey geocentric_coords : list(int) Contains WCS coordinates order : int HiPS order """ # Rule followed: Tile N in order K -> NorderK / DirD / NpixN{.ext} nside = hp.order2nside(order) c = SkyCoord.from_name('crab') theta = (np.pi / 2) - c.dec.radian phi = c.ra.radian npixel = hp.ang2pix(nside, theta, phi) directory = np.around(int(npixel), decimals=-(len(str(npixel)) - 1)) base_url = url + survey \ + '/Norder' + str(order) + '/Dir' + str(directory) + \ '/Npix' + str(npixel) + kwargs['format'] # URL: Crab image # http://alasky.unistra.fr/DSS/DSSColor/Norder6/Dir20000/Npix24185.jpg # base_url = 'http://cade.irap.omp.eu/documents/Ancillary/4Aladin/AKARI_WideS/Norder3/Dir0/Npix346.' + \ # kwargs['format'] if (kwargs['package'] is 'urllib'): import urllib.request data = urllib.request.urlopen(base_url).read() elif (kwargs['package'] is 'requests'): import requests data = requests.get(base_url).content print('WCS : ', ang2WCS(theta, phi)) return data
def make_healpix_map_for_energy_band(energy_band, order): log.info( f'Making HEALPix map for energy band: {energy_band} and order: {order}' ) # Select events in energy band table = Table.read('input_data/fermi_hgps_events_selected.fits.gz', hdu=1) energy = table['ENERGY'].quantity.to('GeV').value mask = (energy_band['min'] <= energy) & (energy < energy_band['max']) table = table[mask] log.info(f'Number of events: {len(table)}') # Bin the events into a HEALPix counts map nside = hp.order2nside(order + shift_order) ipix = hp.ang2pix( nside=nside, nest=True, lonlat=True, theta=table['L'], phi=table['B'], ) npix = hp.nside2npix(nside) log.debug(f'Number of pixels: {npix}') resolution = np.rad2deg(hp.nside2resol(nside)) log.debug(f'Pixel resolution: {resolution} deg') image = np.bincount(ipix, minlength=npix) image = image.astype('float32') # TODO: smoothing the HEALPix map with default setting is very slow. # Maybe chunk the data into local WCS maps and then stitch back together? # For now: no smoothing # image = hp.smoothing(image, sigma=np.deg2rad(0.1)) path = DATA_DIR / 'maps' / 'Fermi10GeV_healpix_maps' / 'energy_{min}_{max}.fits.gz'.format_map( energy_band) path.parent.mkdir(exist_ok=True, parents=True) log.info(f'Writing {path}') hp.write_map(str(path), image, coord='G', nest=True)
def cone_search(self, center: SkyCoord, radius: Angle) -> Iterator[int]: """ cone_search retrieves the Candidate IDs for alerts that can be found in a region of the sky. """ if center.representation_type != CartesianRepresentation: center = center.replicate( representation_type=CartesianRepresentation) pixels = healpy.query_disc( nside=healpy.order2nside(self.order), vec=(center.x.value, center.y.value, center.z.value), radius=radius.degree, inclusive=True, nest=True, ) logger.info("found %d pixels which might match", len(pixels)) ranges = self._compact_pixel_ranges(pixels) logger.info("compacted range into %d elements", len(ranges)) for start, stop in ranges: logger.info("checking range %d to %d", start, stop) for pixel in self.healpixels.iterate(start, stop): for candidate_id in pixel: yield candidate_id
ranking_samples = hist_samples = samples # Place the histogram bins. theta = 0.5 * np.pi - hist_samples['dec'] phi = hist_samples['ra'] p = adaptive_healpix_histogram(theta, phi, opts.samples_per_bin, nside=opts.nside, max_nside=opts.max_nside, nest=True) # Evaluate per-pixel density. p = derasterize(Table([p], names=['PROB'])) order, ipix = moc.uniq2nest(p['UNIQ']) nside = hp.order2nside(order.astype(int)) max_order = order.max().astype(int) max_nside = hp.order2nside(max_order) theta = 0.5 * np.pi - ranking_samples['dec'] phi = ranking_samples['ra'] ranking_samples['ipix'] = hp.ang2pix(max_nside, theta, phi, nest=True) ranking_samples.sort('ipix') result = np.transpose( [pixstats(ranking_samples, max_nside, n, i) for n, i in zip(nside, ipix)]) # Add distance info if necessary. if 'dist' in ranking_samples.colnames: p['PROBDENSITY'], distmean, diststd = result p['DISTMU'], p['DISTSIGMA'], p['DISTNORM'] = \ distance.moments_to_parameters(distmean, diststd)
def test_rasterize_default(order): npix = hp.nside2npix(hp.order2nside(order)) skymap_in = input_skymap(order, 0, 0) skymap_out = moc.rasterize(skymap_in) assert len(skymap_out) == npix
def main(args): # tr = tracker.SummaryTracker() sdt, stars = load_hipparcos_cat() # These will be the stars we see in data source_star_table = sdt[sdt.vmag < 6.5] source_stars = stars[sdt.vmag < 6.5] # These are the stars we use to identify the patch of sky in # the FOV cvmag_lim = 6.1 # Define telescope frame location = EarthLocation.from_geodetic(lon=14.974609, lat=37.693267, height=1750) obstime = Time("2019-05-09T01:37:54.728026") altaz_frame = AltAz(location=location, obstime=obstime) # Get pixel coordinates from TargetCalib camera_config = CameraConfiguration("1.1.0") mapping = camera_config.GetMapping() pos = np.array( [np.array(mapping.GetXPixVector()), np.array(mapping.GetYPixVector())] ) focal_length = u.Quantity(2.15191, u.m) matcher = StarPatternMatch.from_location( altaz_frame=altaz_frame, stars=stars, sdt=sdt, fov=12, focal_length=focal_length, min_alt=-90, vmag_lim=cvmag_lim, pixsize=mapping.GetSize(), ) matcher.silence = True order = 10 nside = healpy.order2nside(order) npix = healpy.nside2npix(nside) npixs_above_horizon = np.where( healpy.pix2ang(nside, np.arange(npix))[0] > np.pi / 2 )[0] tstamp0 = 1557360406 np.random.seed(args.seed) context = zmq.Context() socket = context.socket(zmq.PUB) ip = "127.0.0.1" con_str = "tcp://%s:" % ip + str(args.port) socket.connect(con_str) obstime = Time(tstamp0 + np.random.uniform(-43200, 43200), format="unix") pixsize = mapping.GetSize() for i in range(args.n_iterations): print(f"Sample: {i}", flush=True) tmp_timestamp = tstamp0 + np.random.uniform(-43200, 43200) obstime = Time(tmp_timestamp, format="unix") altaz_frame = AltAz(location=location, obstime=obstime) pixid = int(np.random.uniform(0, npixs_above_horizon.shape[0])) ang = healpy.pix2ang(nside, pixid) alt = ang[0] az = ang[1] hotspots, tel_pointing, star_ind, hips_in_fov, all_hips = generate_hotspots( alt * u.rad, az * u.rad, altaz_frame, source_stars, source_star_table, pixsize, cvmag_lim, pos, ) true_hotspots = np.array(hotspots) frame = Frame() frame.add("hips_in_fov", np.array(hips_in_fov)) frame.add("hotspots", true_hotspots) tel_sky_pointing = tel_pointing.transform_to("icrs") frame.add( "tel_pointing", np.array([tel_sky_pointing.ra.rad, tel_sky_pointing.dec.rad]), ) hotspots = true_hotspots.copy() N_change = 1 # hotspots[N_change, :] = hotspots[N_change, :] + 0.003 matched_hs = matcher.identify_stars( hotspots, horizon_level=0, obstime=obstime, only_one_it=False ) if matched_hs is not None and len(matched_hs) > 0: ra, dec = matcher.determine_pointing(matched_hs) frame.add("matched_hs", np.array(matched_hs)) frame.add("est_pointing", np.array([ra, dec])) else: print(tmp_timestamp, alt, az) matched_hs = matcher.identify_stars( hotspots, horizon_level=0, obstime=obstime, only_one_it=True ) match = np.array(matched_hs) match_quantity = match[:, 2] * match[:, 3] * match[:, 1] index = np.where(match[:, 0] == all_hips[0][1])[0] matched_match = np.argmax(match_quantity) frame.add('true_match', match[index]) frame.add('matched_match', match[matched_match]) frame.add('match_quantity_list', match) socket.send(frame.serialize())
ranking_samples = Table(ranking_samples) hist_samples = Table(samples) else: ranking_samples = hist_samples = samples # Place the histogram bins. theta = 0.5*np.pi - hist_samples['dec'] phi = hist_samples['ra'] p = adaptive_healpix_histogram( theta, phi, opts.samples_per_bin, nside=opts.nside, max_nside=opts.max_nside, nest=True) # Evaluate per-pixel density. p = derasterize(Table([p], names=['PROB'])) order, ipix = moc.uniq2nest(p['UNIQ']) nside = hp.order2nside(order.astype(int)) max_order = order.max().astype(int) max_nside = hp.order2nside(max_order) theta = 0.5*np.pi - ranking_samples['dec'] phi = ranking_samples['ra'] ranking_samples['ipix'] = hp.ang2pix(max_nside, theta, phi, nest=True) ranking_samples.sort('ipix') result = np.transpose( [pixstats(ranking_samples, max_nside, n, i) for n, i in zip(nside, ipix)]) # Add distance info if necessary. if 'dist' in ranking_samples.colnames: p['PROBDENSITY'], distmean, diststd = result p['DISTMU'], p['DISTSIGMA'], p['DISTNORM'] = \ distance.moments_to_parameters(distmean, diststd)
def find_injection_moc(sky_map, true_ra=None, true_dec=None, true_dist=None, contours=(), areas=(), modes=False, nest=False): """ Given a sky map and the true right ascension and declination (in radians), find the smallest area in deg^2 that would have to be searched to find the source, the smallest posterior mass, and the angular offset in degrees from the true location to the maximum (mode) of the posterior. Optionally, also compute the areas of and numbers of modes within the smallest contours containing a given total probability. """ if (true_ra is None) ^ (true_dec is None): raise ValueError('Both true_ra and true_dec must be provided or None') contours = np.asarray(contours) distmean = sky_map.meta.get('distmean', np.nan) # Sort the pixels by descending posterior probability. sky_map = np.flipud(np.sort(sky_map, order='PROBDENSITY')) # Find the pixel that contains the injection. order, ipix = moc.uniq2nest(sky_map['UNIQ']) max_order = np.max(order) max_nside = hp.order2nside(max_order) max_ipix = ipix << np.uint64(2 * (max_order - order)) ipix = ipix.astype(np.int64) max_ipix = max_ipix.astype(np.int64) if true_ra is not None: true_theta = 0.5 * np.pi - true_dec true_phi = true_ra true_pix = hp.ang2pix(max_nside, true_theta, true_phi, nest=True) i = np.argsort(max_ipix) true_idx = i[np.digitize(true_pix, max_ipix[i]) - 1] # Find the angular offset between the mode and true locations. mode_theta, mode_phi = hp.pix2ang( hp.order2nside(order[0]), ipix[0].astype(np.int64), nest=True) if true_ra is None: offset = np.nan else: offset = np.rad2deg( angle_distance(true_theta, true_phi, mode_theta, mode_phi)) # Calculate the cumulative area in deg2 and the cumulative probability. dA = moc.uniq2pixarea(sky_map['UNIQ']) dP = sky_map['PROBDENSITY'] * dA prob = np.cumsum(dP) area = np.cumsum(dA) * np.square(180 / np.pi) # Construct linear interpolants to map between probability and area. # This allows us to compute more accurate contour areas and probabilities # under the approximation that the pixels have constant probability # density. prob_padded = np.concatenate(([0], prob)) area_padded = np.concatenate(([0], area)) # FIXME: we should use the assume_sorted=True argument below, but # it was added in Scipy 0.14.0, and our Scientific Linux 7 clusters # only have Scipy 0.12.1. prob_for_area = interp1d(area_padded, prob_padded) area_for_prob = interp1d(prob_padded, area_padded) if true_ra is None: searched_area = searched_prob = np.nan else: # Find the smallest area that would have to be searched to find # the true location. searched_area = area[true_idx] # Find the smallest posterior mass that would have to be searched to find # the true location. searched_prob = prob[true_idx] # Find the contours of the given credible levels. contour_idxs = np.digitize(contours, prob) - 1 # For each of the given confidence levels, compute the area of the # smallest region containing that probability. contour_areas = area_for_prob(contours).tolist() # For each listed area, find the probability contained within the # smallest credible region of that area. area_probs = prob_for_area(areas).tolist() if modes: if true_ra is None: searched_modes = np.nan else: # Count up the number of modes in each of the given contours. searched_modes = count_modes_moc(sky_map['UNIQ'], true_idx) contour_modes = [ count_modes_moc(sky_map['UNIQ'], i) for i in contour_idxs] else: searched_modes = np.nan contour_modes = np.nan # Distance stats now... if 'DISTMU' in sky_map.dtype.names: probdensity = sky_map['PROBDENSITY'] mu = sky_map['DISTMU'] sigma = sky_map['DISTSIGMA'] norm = sky_map['DISTNORM'] args = (dP, mu, sigma, norm) if true_dist is None: searched_prob_dist = np.nan else: searched_prob_dist = distance.marginal_cdf(true_dist, *args) # FIXME: old verisons of Numpy can't handle passing zero-length # arrays to generalized ufuncs. Remove this workaround once LIGO # Data Grid clusters provide a more modern version of Numpy. if len(contours) == 0: contour_dists = [] else: lo, hi = distance.marginal_ppf( np.row_stack(( 0.5 * (1 - contours), 0.5 * (1 + contours) )), *args) contour_dists = (hi - lo).tolist() # Set up distance grid. n_r = 1000 max_r = max(6 * distmean, np.max(true_dist)) d_r = max_r / n_r r = d_r * np.arange(1, n_r) # Calculate volume of frustum-shaped voxels with distance centered on r # and extending from (r - d_r) to (r + d_r). dV = (np.square(r) + np.square(d_r) / 12) * d_r * dA.reshape(-1, 1) # Calculate probability within each voxel. dP = probdensity.reshape(-1, 1) * dV * np.exp( -0.5 * np.square( (r.reshape(1, -1) - mu.reshape(-1, 1)) / sigma.reshape(-1, 1) ) ) * (norm / sigma).reshape(-1, 1) / np.sqrt(2 * np.pi) dP[np.isnan(dP)] = 0 # Suppress invalid values # Calculate probability density per unit volume. dP_dV = dP / dV i = np.flipud(np.argsort(dP_dV.ravel())) P_flat = np.cumsum(dP.ravel()[i]) V_flat = np.cumsum(dV.ravel()[i]) contour_vols = interp1d( P_flat, V_flat, bounds_error=False)(contours).tolist() P = np.empty_like(P_flat) V = np.empty_like(V_flat) P[i] = P_flat V[i] = V_flat P = P.reshape(dP.shape) V = V.reshape(dV.shape) if true_dist is None: searched_vol = searched_prob_vol = np.nan else: i_radec = true_idx i_dist = np.digitize(true_dist, r) - 1 searched_prob_vol = P[i_radec, i_dist] searched_vol = V[i_radec, i_dist] else: searched_vol = searched_prob_vol = searched_prob_dist = np.nan contour_dists = [np.nan] * len(contours) contour_vols = [np.nan] * len(contours) # Done. return FoundInjection( searched_area, searched_prob, offset, searched_modes, contour_areas, area_probs, contour_modes, searched_prob_dist, contour_dists, searched_vol, searched_prob_vol, contour_vols)
def flat_bitmap(self): """Return flattened HEALPix representation.""" m = np.empty(hp.nside2npix(hp.order2nside(self.order))) for nside, full_nside, ipix, ipix0, ipix1, samples in self.visit(): m[ipix0:ipix1] = len(samples) / hp.nside2pixarea(nside) return m
""" from __future__ import division import numpy as np import healpy as hp import collections import itertools __all__ = ('HEALPIX_MACHINE_ORDER', 'HEALPIX_MACHINE_NSIDE', 'HEALPixTree', 'adaptive_healpix_histogram', 'interpolate_nested', 'reconstruct_nested') # Maximum 64-bit HEALPix resolution. HEALPIX_MACHINE_ORDER = 29 HEALPIX_MACHINE_NSIDE = hp.order2nside(HEALPIX_MACHINE_ORDER) _HEALPixTreeVisitExtra = collections.namedtuple( 'HEALPixTreeVisit', 'nside full_nside ipix ipix0 ipix1 value') _HEALPixTreeVisit = collections.namedtuple( 'HEALPixTreeVisit', 'nside ipix') class HEALPixTree(object): """Data structure used internally by the function adaptive_healpix_histogram().""" def __init__(
def find_injection_moc(sky_map, true_ra, true_dec, contours=(), areas=(), modes=False, nest=False): """ Given a sky map and the true right ascension and declination (in radians), find the smallest area in deg^2 that would have to be searched to find the source, the smallest posterior mass, and the angular offset in degrees from the true location to the maximum (mode) of the posterior. Optionally, also compute the areas of and numbers of modes within the smallest contours containing a given total probability. """ # Sort the pixels by descending posterior probability. sky_map = np.flipud(np.sort(sky_map, order='PROBDENSITY')) # Find the pixel that contains the injection. order, ipix = moc.uniq2nest(sky_map['UNIQ']) max_order = np.max(order) max_nside = hp.order2nside(max_order) max_ipix = ipix << np.uint64(2 * (max_order - order)) ipix = ipix.astype(np.int64) max_ipix = max_ipix.astype(np.int64) true_theta = 0.5 * np.pi - true_dec true_phi = true_ra true_pix = hp.ang2pix(max_nside, true_theta, true_phi, nest=True) # At this point, we could sort the dataset by max_ipix and then do a binary # (e.g., np.searchsorted) to find true_pix in max_ipix. However, would be # slower than the linear search below because the sort would be N log N. i = np.flatnonzero(max_ipix <= true_pix) true_idx = i[np.argmax(max_ipix[i])] # Find the angular offset between the mode and true locations. mode_theta, mode_phi = hp.pix2ang(hp.order2nside(order[0]), ipix[0].astype(np.int64), nest=True) offset = np.rad2deg( angle_distance(true_theta, true_phi, mode_theta, mode_phi)) # Calculate the cumulative area in deg2 and the cumulative probability. area = moc.uniq2pixarea(sky_map['UNIQ']) prob = np.cumsum(sky_map['PROBDENSITY'] * area) area = np.cumsum(area) * np.square(180 / np.pi) # Construct linear interpolants to map between probability and area. # This allows us to compute more accurate contour areas and probabilities # under the approximation that the pixels have constant probability # density. prob_padded = np.concatenate(([0], prob)) area_padded = np.concatenate(([0], area)) # FIXME: we should use the assume_sorted=True argument below, but # it was added in Scipy 0.14.0, and our Scientific Linux 7 clusters # only have Scipy 0.12.1. prob_for_area = interp1d(area_padded, prob_padded) area_for_prob = interp1d(prob_padded, area_padded) # Find the smallest area that would have to be searched to find # the true location. searched_area = area[true_idx] # Find the smallest posterior mass that would have to be searched to find # the true location. searched_prob = prob[true_idx] # Find the contours of the given credible levels. contour_idxs = np.searchsorted(prob, contours) # For each of the given confidence levels, compute the area of the # smallest region containing that probability. contour_areas = area_for_prob(contours).tolist() # For each listed area, find the probability contained within the # smallest credible region of that area. area_probs = prob_for_area(areas).tolist() if modes: # Count up the number of modes in each of the given contours. searched_modes = count_modes_moc(sky_map['UNIQ'], true_idx) contour_modes = [ count_modes_moc(sky_map['UNIQ'], i) for i in contour_idxs ] else: searched_modes = None contour_modes = None # Done. return FoundInjection(searched_area, searched_prob, offset, searched_modes, contour_areas, area_probs, contour_modes)
def find_injection_moc(sky_map, true_ra=None, true_dec=None, true_dist=None, contours=(), areas=(), modes=False, nest=False): """ Given a sky map and the true right ascension and declination (in radians), find the smallest area in deg^2 that would have to be searched to find the source, the smallest posterior mass, and the angular offset in degrees from the true location to the maximum (mode) of the posterior. Optionally, also compute the areas of and numbers of modes within the smallest contours containing a given total probability. """ if (true_ra is None) ^ (true_dec is None): raise ValueError('Both true_ra and true_dec must be provided or None') contours = np.asarray(contours) distmean = sky_map.meta.get('distmean', np.nan) # Sort the pixels by descending posterior probability. sky_map = np.flipud(np.sort(sky_map, order='PROBDENSITY')) # Find the pixel that contains the injection. order, ipix = moc.uniq2nest(sky_map['UNIQ']) max_order = np.max(order) max_nside = hp.order2nside(max_order) max_ipix = ipix << np.uint64(2 * (max_order - order)) ipix = ipix.astype(np.int64) max_ipix = max_ipix.astype(np.int64) if true_ra is not None: true_theta = 0.5 * np.pi - true_dec true_phi = true_ra true_pix = hp.ang2pix(max_nside, true_theta, true_phi, nest=True) i = np.argsort(max_ipix) true_idx = i[digitize(true_pix, max_ipix[i]) - 1] # Find the angular offset between the mode and true locations. mode_theta, mode_phi = hp.pix2ang( hp.order2nside(order[0]), ipix[0].astype(np.int64), nest=True) if true_ra is None: offset = np.nan else: offset = np.rad2deg( angle_distance(true_theta, true_phi, mode_theta, mode_phi)) # Calculate the cumulative area in deg2 and the cumulative probability. dA = moc.uniq2pixarea(sky_map['UNIQ']) dP = sky_map['PROBDENSITY'] * dA prob = np.cumsum(dP) area = np.cumsum(dA) * np.square(180 / np.pi) # Construct linear interpolants to map between probability and area. # This allows us to compute more accurate contour areas and probabilities # under the approximation that the pixels have constant probability # density. prob_padded = np.concatenate(([0], prob)) area_padded = np.concatenate(([0], area)) # FIXME: we should use the assume_sorted=True argument below, but # it was added in Scipy 0.14.0, and our Scientific Linux 7 clusters # only have Scipy 0.12.1. prob_for_area = interp1d(area_padded, prob_padded) area_for_prob = interp1d(prob_padded, area_padded) if true_ra is None: searched_area = searched_prob = np.nan else: # Find the smallest area that would have to be searched to find # the true location. searched_area = area[true_idx] # Find the smallest posterior mass that would have to be searched to find # the true location. searched_prob = prob[true_idx] # Find the contours of the given credible levels. contour_idxs = digitize(contours, prob) - 1 # For each of the given confidence levels, compute the area of the # smallest region containing that probability. contour_areas = area_for_prob(contours).tolist() # For each listed area, find the probability contained within the # smallest credible region of that area. area_probs = prob_for_area(areas).tolist() if modes: if true_ra is None: searched_modes = np.nan else: # Count up the number of modes in each of the given contours. searched_modes = count_modes_moc(sky_map['UNIQ'], true_idx) contour_modes = [ count_modes_moc(sky_map['UNIQ'], i) for i in contour_idxs] else: searched_modes = np.nan contour_modes = np.nan # Distance stats now... if 'DISTMU' in sky_map.dtype.names: probdensity = sky_map['PROBDENSITY'] mu = sky_map['DISTMU'] sigma = sky_map['DISTSIGMA'] norm = sky_map['DISTNORM'] args = (dP, mu, sigma, norm) if true_dist is None: searched_prob_dist = np.nan else: searched_prob_dist = distance.marginal_cdf(true_dist, *args) # FIXME: old verisons of Numpy can't handle passing zero-length # arrays to generalized ufuncs. Remove this workaround once LIGO # Data Grid clusters provide a more modern version of Numpy. if len(contours) == 0: contour_dists = [] else: lo, hi = distance.marginal_ppf( np.row_stack(( 0.5 * (1 - contours), 0.5 * (1 + contours) )), *args) contour_dists = (hi - lo).tolist() # Set up distance grid. n_r = 1000 max_r = 6 * distmean if true_dist is not None and true_dist > max_r: max_r = true_dist d_r = max_r / n_r r = d_r * np.arange(1, n_r) # Calculate volume of frustum-shaped voxels with distance centered on r # and extending from (r - d_r) to (r + d_r). dV = (np.square(r) + np.square(d_r) / 12) * d_r * dA.reshape(-1, 1) # Calculate probability within each voxel. dP = probdensity.reshape(-1, 1) * dV * np.exp( -0.5 * np.square( (r.reshape(1, -1) - mu.reshape(-1, 1)) / sigma.reshape(-1, 1) ) ) * (norm / sigma).reshape(-1, 1) / np.sqrt(2 * np.pi) dP[np.isnan(dP)] = 0 # Suppress invalid values # Calculate probability density per unit volume. dP_dV = dP / dV i = np.flipud(np.argsort(dP_dV.ravel())) P_flat = np.cumsum(dP.ravel()[i]) V_flat = np.cumsum(dV.ravel()[i]) contour_vols = interp1d( P_flat, V_flat, bounds_error=False)(contours).tolist() P = np.empty_like(P_flat) V = np.empty_like(V_flat) P[i] = P_flat V[i] = V_flat P = P.reshape(dP.shape) V = V.reshape(dV.shape) if true_dist is None: searched_vol = searched_prob_vol = np.nan else: i_radec = true_idx i_dist = digitize(true_dist, r) - 1 searched_prob_vol = P[i_radec, i_dist] searched_vol = V[i_radec, i_dist] else: searched_vol = searched_prob_vol = searched_prob_dist = np.nan contour_dists = [np.nan] * len(contours) contour_vols = [np.nan] * len(contours) # Done. return FoundInjection( searched_area, searched_prob, offset, searched_modes, contour_areas, area_probs, contour_modes, searched_prob_dist, contour_dists, searched_vol, searched_prob_vol, contour_vols)
""" from __future__ import division __author__ = "Leo Singer <*****@*****.**>" import numpy as np import healpy as hp import collections import itertools import lal import lalsimulation # Maximum 64-bit HEALPix resolution. HEALPIX_MACHINE_ORDER = 29 HEALPIX_MACHINE_NSIDE = hp.order2nside(HEALPIX_MACHINE_ORDER) def nside2order(nside): """Convert lateral HEALPix resolution to order. FIXME: see https://github.com/healpy/healpy/issues/163""" order = np.log2(nside) int_order = int(order) if order != int_order: raise ValueError('not a valid value for nside: {0}'.format(nside)) return int_order def order2nside(order): return 1 << order
def crossmatch(sky_map, coordinates=None, contours=(), areas=(), modes=False, cosmology=False): """ Given a sky map and the true right ascension and declination (in radians), find the smallest area in deg^2 that would have to be searched to find the source, the smallest posterior mass, and the angular offset in degrees from the true location to the maximum (mode) of the posterior. Optionally, also compute the areas of and numbers of modes within the smallest contours containing a given total probability. Parameters ---------- sky_map : :class:`astropy.table.Table` A multiresolution sky map, as returned by :func:`ligo.skymap.io.fits.read_sky_map` called with the keyword argument ``moc=True``. coordinates : :class:`astropy.coordinates.SkyCoord`, optional The catalog of target positions to match against. contours : :class:`tuple`, optional Credible levels between 0 and 1. If this argument is present, then calculate the areas and volumes of the 2D and 3D credible regions that contain these probabilities. For example, for ``contours=(0.5, 0.9)``, then areas and volumes of the 50% and 90% credible regions. areas : :class:`tuple`, optional Credible areas in square degrees. If this argument is present, then calculate the probability contained in the 2D credible levels that have these areas. For example, for ``areas=(20, 100)``, then compute the probability within the smallest credible levels of 20 deg² and 100 deg², respectively. modes : :class:`bool`, optional If True, then enable calculation of the number of distinct modes or islands of probability. Note that this option may be computationally expensive. cosmology : :class:`bool`, optional If True, then search space by descending probability density per unit comoving volume. If False, then search space by descending probability per luminosity distance cubed. Returns ------- result : :class:`~ligo.skymap.postprocess.crossmatch.CrossmatchResult` Notes ----- This function is also be used for injection finding; see :doc:`/ligo/skymap/tool/ligo_skymap_stats`. Examples -------- First, some imports: >>> from astroquery.vizier import VizierClass >>> from astropy.coordinates import SkyCoord >>> from ligo.skymap.io import read_sky_map >>> from ligo.skymap.postprocess import crossmatch Next, retrieve the GLADE catalog using Astroquery and get the coordinates of all its entries: >>> vizier = VizierClass( ... row_limit=-1, columns=['GWGC', '_RAJ2000', '_DEJ2000', 'Dist']) >>> cat, = vizier.get_catalogs('VII/281/glade2') >>> coordinates = SkyCoord(cat['_RAJ2000'], cat['_DEJ2000'], cat['Dist']) Load the multiresolution sky map for S190814bv: >>> url = 'https://gracedb.ligo.org/api/superevents/S190814bv/files/bayestar.multiorder.fits' >>> skymap = read_sky_map(url, moc=True) Perform the cross match: >>> result = crossmatch(skymap, coordinates) Using the cross match results, we can list the galaxies within the 90% credible volume: >>> print(cat[result.searched_prob_vol < 0.9]) GWGC _RAJ2000 _DEJ2000 Dist deg deg Mpc ---------- -------------------- -------------------- -------------------- NGC0171 9.3396699999999999 -19.9342460000000017 57.56212553960000 --- 20.2009090000000064 -31.1146050000000010 137.16022925600001 ESO540-003 8.9144679999999994 -20.1252980000000008 49.07809291930000 --- 10.6762720000000009 -21.7740819999999999 276.46938505499998 --- 13.5855169999999994 -23.5523850000000010 138.44550704800000 --- 20.6362969999999990 -29.9825149999999958 160.23313164900000 --- 13.1923879999999993 -22.9750179999999986 236.96795954500001 --- 11.7813630000000007 -24.3706470000000017 244.25031189699999 --- 19.1711120000000008 -31.4339490000000019 152.13614001400001 --- 13.6367060000000002 -23.4948789999999974 141.25162979500001 ... ... ... ... --- 11.3517000000000010 -25.8596999999999966 335.73800000000000 --- 11.2073999999999998 -25.7149000000000001 309.02999999999997 --- 11.1875000000000000 -25.7503999999999991 295.12099999999998 --- 10.8608999999999991 -25.6904000000000003 291.07200000000000 --- 10.6938999999999975 -25.6778300000000002 323.59399999999999 --- 15.4935000000000009 -26.0304999999999964 304.78899999999999 --- 15.2794000000000008 -27.0410999999999966 320.62700000000001 --- 14.8323999999999980 -27.0459999999999994 320.62700000000001 --- 14.5341000000000005 -26.0949000000000026 307.61000000000001 --- 23.1280999999999963 -31.1109199999999966 320.62700000000001 Length = 1478 rows """ # noqa: E501 # Astropy coordinates that are constructed without distance have # a distance field that is unity (dimensionless). if coordinates is None: true_ra = true_dec = true_dist = None else: coordinates = coordinates.icrs true_ra = coordinates.ra.rad true_dec = coordinates.dec.rad if np.any(coordinates.distance != 1): true_dist = coordinates.distance.to_value(u.Mpc) else: true_dist = None contours = np.asarray(contours) distmean = sky_map.meta.get('distmean', np.nan) # Sort the pixels by descending posterior probability. sky_map = np.flipud(np.sort(sky_map, order='PROBDENSITY')) # Find the pixel that contains the injection. order, ipix = moc.uniq2nest(sky_map['UNIQ']) max_order = np.max(order) max_nside = hp.order2nside(max_order) max_ipix = ipix << np.int64(2 * (max_order - order)) if true_ra is not None: true_theta = 0.5 * np.pi - true_dec true_phi = true_ra true_pix = hp.ang2pix(max_nside, true_theta, true_phi, nest=True) i = np.argsort(max_ipix) true_idx = i[np.digitize(true_pix, max_ipix[i]) - 1] # Find the angular offset between the mode and true locations. mode_theta, mode_phi = hp.pix2ang( hp.order2nside(order[0]), ipix[0], nest=True) if true_ra is None: offset = np.nan else: offset = np.rad2deg( angle_distance(true_theta, true_phi, mode_theta, mode_phi)) # Calculate the cumulative area in deg2 and the cumulative probability. dA = moc.uniq2pixarea(sky_map['UNIQ']) dP = sky_map['PROBDENSITY'] * dA prob = np.cumsum(dP) area = np.cumsum(dA) * np.square(180 / np.pi) if true_ra is None: searched_area = searched_prob = probdensity = np.nan else: # Find the smallest area that would have to be searched to find # the true location. searched_area = area[true_idx] # Find the smallest posterior mass that would have to be searched to # find the true location. searched_prob = prob[true_idx] # Find the probability density. probdensity = sky_map['PROBDENSITY'][true_idx] # Find the contours of the given credible levels. contour_idxs = np.digitize(contours, prob) - 1 # For each of the given confidence levels, compute the area of the # smallest region containing that probability. contour_areas = np.interp( contours, prob, area, left=0, right=4*180**2/np.pi).tolist() # For each listed area, find the probability contained within the # smallest credible region of that area. area_probs = np.interp(areas, area, prob, left=0, right=1).tolist() if modes: if true_ra is None: searched_modes = np.nan else: # Count up the number of modes in each of the given contours. searched_modes = count_modes_moc(sky_map['UNIQ'], true_idx) contour_modes = [ count_modes_moc(sky_map['UNIQ'], i) for i in contour_idxs] else: searched_modes = np.nan contour_modes = np.nan # Distance stats now... if 'DISTMU' in sky_map.dtype.names: dP_dA = sky_map['PROBDENSITY'] mu = sky_map['DISTMU'] sigma = sky_map['DISTSIGMA'] norm = sky_map['DISTNORM'] # Set up distance grid. n_r = 1000 max_r = 6 * distmean if true_dist is not None and np.max(true_dist) > max_r: max_r = np.max(true_dist) d_r = max_r / n_r # Calculate searched_prob_dist and contour_dists. r = d_r * np.arange(1, n_r) P_r = distance.marginal_cdf(r, dP, mu, sigma, norm) if true_dist is None: searched_prob_dist = np.nan else: searched_prob_dist = np.interp(true_dist, r, P_r, left=0, right=1) if len(contours) == 0: contour_dists = [] else: lo, hi = np.interp( np.row_stack(( 0.5 * (1 - contours), 0.5 * (1 + contours) )), P_r, r, left=0, right=np.inf) contour_dists = (hi - lo).tolist() # Calculate volume of each voxel, defined as the region within the # HEALPix pixel and contained within the two centric spherical shells # with radii (r - d_r / 2) and (r + d_r / 2). dV = (np.square(r) + np.square(d_r) / 12) * d_r * dA.reshape(-1, 1) # Calculate probability within each voxel. dP = np.exp( -0.5 * np.square( (r.reshape(1, -1) - mu.reshape(-1, 1)) / sigma.reshape(-1, 1) ) ) * (dP_dA * norm / (sigma * np.sqrt(2 * np.pi))).reshape(-1, 1) * dV dP[np.isnan(dP)] = 0 # Suppress invalid values # Calculate probability density per unit volume. if cosmology: dV *= dVC_dVL_for_DL(r) dP_dV = dP / dV i = np.flipud(np.argsort(dP_dV.ravel())) P_flat = np.cumsum(dP.ravel()[i]) V_flat = np.cumsum(dV.ravel()[i]) contour_vols = np.interp( contours, P_flat, V_flat, left=0, right=np.inf).tolist() P = np.empty_like(P_flat) V = np.empty_like(V_flat) P[i] = P_flat V[i] = V_flat P = P.reshape(dP.shape) V = V.reshape(dV.shape) if true_dist is None: searched_vol = searched_prob_vol = probdensity_vol = np.nan else: i_radec = true_idx i_dist = np.digitize(true_dist, r) - 1 probdensity_vol = dP_dV[i_radec, i_dist] searched_prob_vol = P[i_radec, i_dist] searched_vol = V[i_radec, i_dist] else: searched_vol = searched_prob_vol = searched_prob_dist \ = probdensity_vol = np.nan contour_dists = [np.nan] * len(contours) contour_vols = [np.nan] * len(contours) # Done. return CrossmatchResult( searched_area, searched_prob, offset, searched_modes, contour_areas, area_probs, contour_modes, searched_prob_dist, contour_dists, searched_vol, searched_prob_vol, contour_vols, probdensity, probdensity_vol)