def main():
ap = ArgumentParser()
ap.add_argument('fpm', help='e.g. /scratch/fpm_0.1000/')
ap.add_argument('ll', type=float, help='e.g. 0.2 or 0.168')
ap.add_argument('--with-peak', help='Find Peaks KDDensity estimation (slow)', default=True)
ap.add_argument('fof', help='e.g. /scratch/fpm_0.1000/fof . Will write to {fof}/{ll}')
ap.add_argument('--nmin', type=int, default=20, help='min number of particles to be in the catalogue')
ns = ap.parse_args()
cat = BigFileCatalog(ns.fpm, header='Header', dataset='1/')
cat.attrs['BoxSize'] = numpy.ones(3) * cat.attrs['BoxSize'][0]
cat.attrs['Nmesh'] = numpy.ones(3) * cat.attrs['NC'][0]
cosmo = Planck15.match(Omega0_m=cat.attrs['OmegaM'])
M0 = cat.attrs['OmegaM'][0] * 27.75 * 1e10 * cat.attrs['BoxSize'].prod() / cat.csize
if cat.comm.rank == 0:
print('BoxSize', cat.attrs['BoxSize'])
print('Nmesh', cat.attrs['Nmesh'])
print('Mass of a particle', M0)
print('OmegaM', cosmo.Om0)
if ns.with_peak:
cat['Density'] = KDDensity(cat).density
fof = FOF(cat, linking_length=ns.ll, nmin=ns.nmin)
if ns.with_peak:
features = fof.find_features(peakcolumn='Density')
else:
features = fof.find_features(peakcolumn=None)
features['Mass'] = M0 * features['Length']
if fof.comm.rank == 0:
print('Total number of features found', features.csize)
print('Saving columns', features.columns)
features.save(ns.fof + '/%0.3f' % ns.ll, features.columns)
def main(argv):
del argv
""" -------------- setting paramaeters ------------------------"""
params = FLAGS.flag_values_dict()
params['Nmesh'] = [FLAGS.Nmesh] * 3
params['BoxSize'] = [FLAGS.boxsize] * 3
params['Nmesh2D'] = [FLAGS.Nmesh2D] * 2
params['BoxSize2D'] = [FLAGS.boxsize2D] * 2
params['zs_source'] = [float(zs) for zs in FLAGS.zs_source]
if params['custom_cosmo']:
cosmo = Planck15.match(Omega0_m=FLAGS.Omega_m)
cosmo = cosmo.match(sigma8=FLAGS.sigma_8)
else:
if rank == 0:
print('custom_cosmo is set to False. Using default cosmology.')
cosmo = Planck15
if params['save3D'] or params['save3Dpower']:
try:
assert (params['interpolate'] == False)
except:
raise ValueError(
'interpolate must be set to False if requesting 3D outouts')
"""------- setting output dirs and saving parameters-----------"""
dirs = {}
if rank == 0:
cmd = "git log --pretty=format:'%h' -n 1"
githash = subprocess.run([cmd], stdout=subprocess.PIPE,
shell=True).stdout.decode('utf-8')
print('dumping under githash %s' % githash)
output_path = os.path.join(FLAGS.output_path, githash)
params_path = os.path.join(os.path.join(os.getcwd()), 'runs', githash)
params['output_path'] = output_path
print(params_path)
if not os.path.isdir(params_path):
os.makedirs(params_path)
# make sure parameter file name is unique and we are not repeating a run
num_run = 0
found = True
while found:
path_name = os.path.join(output_path,
params['label'] + '%d/' % num_run)
params_file = os.path.join(params_path,
params['label'] + '%d.json' % num_run)
if not os.path.isdir(path_name):
os.makedirs(path_name)
found = False
if not os.path.isfile(params_file):
found = False
else:
with open(params_file, 'r') as f:
old_params = json.load(f)
if old_params == params and not params['debug']:
raise ValueError(
'run with same settings already exists: %s' %
params_file)
elif params['debug']:
found = False
else:
num_run += 1
for result in ['cls', 'maps', 'snapshots']:
dirs[result] = os.path.join(path_name, result)
if not os.path.isdir(dirs[result]):
os.makedirs(dirs[result])
fjson = json.dumps(params)
f = open(params_file, "w")
f.write(fjson)
f.close()
dirs = comm.bcast(dirs, root=0)
params['snapshot_dir'] = dirs['snapshots']
"""---------------------------run actual simulations-----------------------------"""
sims_start = time.time()
for ii in range(FLAGS.N_maps):
if rank == 0:
print('progress in percent:', ii / params['N_maps'] * 100)
kmaps, kmaps_deriv, pm = run_wl_sim(params, cosmo=cosmo, num=ii)
for jj, z_source in enumerate(params['zs_source']):
kmap = kmaps[jj]
mapfile = os.path.join(
dirs['maps'], 'map_decon_zsource%d_map%d_of%d' %
(z_source * 10, ii, params['N_maps']) + '.npy')
save_2Dmap(kmap, mapfile)
if rank == 0:
print('2D map #%d at z_s=%.1f dumped to %s' %
(ii, z_source, mapfile))
end = time.time()
if rank == 0:
print('time taken per sim in sec %d' %
((end - sims_start) /
(params['N_maps'] * len(params['zs_source']))))
print('time takes before sims in sec %d' % (sims_start - start))
def main():
"""
Script to compute FOF halos from treepm DM catalog. Compute virial mass
which is needed for HOD models.
Note: Before March 2020, used mass given by number of particles in halo,
see psiRec/psirec/main_ms_gadget_fof_halofinder_nbkit0.3.py.
"""
ap = ArgumentParser()
ap.add_argument('treepm',
help='Directory of TreePM matter field, e.g. /scratch/treepm_0.1000/')
ap.add_argument('ll', type=float,
help='Linking length of finding halos, e.g. 0.2 or 0.168',
default=0.2)
ap.add_argument('fof',
help=('Output directory of halo catalogs, e.g. '
'/scratch/treepm_0.1000/fof . Will write to {fof}/{ll_nmin_mvir}'))
ap.add_argument('--nmin', type=int, default=20,
help='min number of particles to be in the catalogue')
ap.add_argument('--with-peak', help='Find Peaks KDDensity estimation (slow)',
default=False)
ns = ap.parse_args()
cat = BigFileCatalog(ns.treepm, header='Header', dataset='1/')
cat.attrs['BoxSize'] = np.ones(3) * cat.attrs['BoxSize'][0]
cat.attrs['Nmesh'] = np.ones(3) * 512.0 # in TreePM catalog, there is no 'NC' attribute
cosmo = Planck15.match(Omega0_m=cat.attrs['Omega0'])
# In TreePM, we need to use 'Omega0' instead of 'OmegaM' in FastPM.
# csize is the total number of particles
M0 = (cat.attrs['Omega0'][0] * 27.75 * 1e10 * cat.attrs['BoxSize'].prod()
/ cat.csize)
redshift = 1.0/cat.attrs['Time'][0]-1.0
if cat.comm.rank == 0:
print('BoxSize', cat.attrs['BoxSize'])
print('Mass of a particle', M0)
print('OmegaM', cosmo.Om0)
print('attrs', cat.attrs.keys())
print('Redshift', redshift)
if ns.with_peak:
posdef = 'peak'
else:
posdef = 'cm'
# Halos which have more than nmin particles are selected.
fof = FOF(cat, linking_length=ns.ll, nmin=ns.nmin)
# Compute halo catalog. Mass column contains virial mass, which is needed
# to get concentration needed for hod.
halos = fof.to_halos(
cosmo=cosmo,
redshift=redshift,
particle_mass=M0,
mdef='vir',
posdef=posdef,
peakcolumn='Density')
halos['log10M'] = np.log10(halos['Mass'])
# print info
if fof.comm.rank == 0:
print('Total number of halos found', halos.csize)
print('Saving columns', halos.columns)
if not os.path.exists(ns.fof):
os.makedirs(ns.fof)
# Save the halo catalog to disk so can easily load it later to populate
# galaxies with hod.
out_fname = ns.fof + '/ll_{0:.3f}_nmin{1}_mvir'.format(ns.ll, ns.nmin+1)
if ns.with_peak:
out_fname += '_peakpos'
# MS: Somehow crashes b/c some ranks don't see header file. running
# a second time works though. maybe write header first with
# single rank?
halos.save(out_fname, halos.columns)
if fof.comm.rank == 0:
print('Saved HaloCatalog to %s' % out_fname)
def run_hod(cat,
HOD_model_name=None,
hod_seed=42,
add_RSD=False,
RSD_LOS=None):
"""
Run HOD to get galaxy catalog from input halo catalog.
Parameters
----------
cat : nbodykit Catalog object
Input halo catalog, should use virial mass as 'Mass' column.
"""
if cat.comm.rank == 0:
print('HOD model: %s' % HOD_model_name)
cat.attrs['BoxSize'] = np.ones(3) * cat.attrs['BoxSize'][0]
cat.attrs['Nmesh'] = np.ones(
3) * 512.0 # in TreePM catalog, there is no 'NC' attribute
cosmo = Planck15.match(Omega0_m=cat.attrs['Omega0'])
# In TreePM, we need to use 'Omega0' instead of 'OmegaM' in FastPM.
# csize is the total number of particles
M0 = (cat.attrs['Omega0'][0] * 27.75 * 1e10 * cat.attrs['BoxSize'].prod() /
cat.csize)
redshift = 1.0 / cat.attrs['Time'][0] - 1.0
# convert to HaloCatalog
halos = HaloCatalog(cat, cosmo, redshift)
if cat.comm.rank == 0:
print('BoxSize', halos.attrs['BoxSize'])
print('attrs', halos.attrs.keys())
print('RSDFactor', halos.attrs['RSDFactor'])
print('Columns', halos.columns)
# Define HOD
if HOD_model_name in [
'Zheng07_HandSeljak17_v2', 'Zheng07_HandSeljak17_centrals_v2',
'Zheng07_HandSeljak17_sats_v2',
'Zheng07_HandSeljak17_parent_halos_v2'
]:
# (1) Hand & Seljak 1706.02362:
# Uses {log10 Mmin, sigma log10 M, log10 M1, alpha, log10 Mcut} = {12.99, 0.308, 14.08, 0.824, 13.20}.
# See Reid et al https://arxiv.org/pdf/1404.3742.pdf eq 17-19
# (2) halotools docs on zheng07 model:
# See https://halotools.readthedocs.io/en/stable/quickstart_and_tutorials/tutorials/model_building/preloaded_models/zheng07_composite_model.html#zheng07-parameters):
# logMmin - Minimum mass required for a halo to host a central galaxy.
# sigma_logM - Rate of transition from <Ncen>=0 -> <Ncen=1>.
# alpha - Power law slope of the relation between halo mass and <Nsat>.
# logM0 - Low-mass cutoff in <Nsat>.
# logM1 - Characteristic halo mass where <Nsat> begins to assume a power law form.
# 11 June 2020: Zheng07_HandSeljak17_v2 uses fixed RSDFactor, which was wrong by factor of 1/a before.
hodmodel = Zheng07Model.to_halotools(cosmo=cosmo,
redshift=redshift,
mdef='vir')
# HOD parameters from Hand & Seljak 1706.02362
hodmodel.param_dict['logMmin'] = 12.99
hodmodel.param_dict['sigma_logM'] = 0.308
hodmodel.param_dict['logM1'] = 14.08
hodmodel.param_dict['alpha'] = 1.06
hodmodel.param_dict[
'logM0'] = 13.20 # this is called Mcut in Hand et al and Reid et al.
if cat.comm.rank == 0:
print('Use zheng07model with:', hodmodel.param_dict)
# Run HOD
galcat = halos.populate(hodmodel, seed=hod_seed)
# select which galaxies to keep
if HOD_model_name == 'Zheng07_HandSeljak17_v2':
# keep all
pass
elif HOD_model_name == 'Zheng07_HandSeljak17_centrals_v2':
# select only centrals
ww = galcat['gal_type'] == 0 # 0: central, 1: satellite
galcat = galcat[ww]
elif HOD_model_name == 'Zheng07_HandSeljak17_sats_v2':
# select only satellites
ww = galcat['gal_type'] == 1 # 0: central, 1: satellite
galcat = galcat[ww]
elif HOD_model_name == 'Zheng07_HandSeljak17_parent_halos_v2':
# select centrals
ww = galcat['gal_type'] == 0 # 0: central, 1: satellite
galcat = galcat[ww]
# set position to that of parent halo (in Mpc/h)
halo_pos = galcat['Position'].compute() + np.nan
halo_pos[:, 0] = galcat['halo_x'].compute()
halo_pos[:, 1] = galcat['halo_y'].compute()
halo_pos[:, 2] = galcat['halo_z'].compute()
galcat['Position'] = halo_pos
del halo_pos
# set velocity to that of parent halo (in km/s)
halo_vel = galcat['Velocity'].compute() + np.nan
halo_vel[:, 0] = galcat['halo_vx'].compute()
halo_vel[:, 1] = galcat['halo_vy'].compute()
halo_vel[:, 2] = galcat['halo_vz'].compute()
galcat['Velocity'] = halo_vel
del halo_vel
# Get RSD displacement = v_z/(aH(a)), where v_z is halo velocity.
# Compute rsd_factor = 1/(aH(a)) = (1+z)/H(z)
# see https://nbodykit.readthedocs.io/en/latest/catalogs/common-operations.html#Adding-Redshift-space-Distortions
rsd_factor = (1. + redshift) / (100. * cosmo.efunc(redshift))
raise Exception(
'this is not correct for ms_gadget which has a^2 dx/dt for velocity.'
)
galcat['VelocityOffset'] = rsd_factor * galcat['Velocity']
# columns: ['Position', 'Selection', 'Value', 'Velocity', 'VelocityOffset', 'Weight', 'conc_NFWmodel', 'gal_type', 'halo_hostid', 'halo_id', 'halo_mvir', 'halo_num_centrals', 'halo_num_satellites', 'halo_rvir', 'halo_upid', 'halo_vx', 'halo_vy', 'halo_vz', 'halo_x', 'halo_y', 'halo_z', 'host_centric_distance', 'vx', 'vy', 'vz', 'x', 'y', 'z']
else:
raise Exception('Unknown hod_model %s' % HOD_model_name)
if add_RSD:
assert type(RSD_LOS) == np.ndarray
assert RSD_LOS.shape == (3, )
print('cat attrs:', galcat.attrs)
# It seems like halos.populate gives satellite velocity in km/s by drawing from NFW profile, and sets central velocity equal to halo velocity.
# But not sure what units are assumed for halo velocity. Note we have different velocity a prefactor in ms_gadget and new MP-Gadget format.
# Also, should probably use peak velocity instead of bulk velocity of halos for the centrals velocity.
# So HOD just seems screwed up.
raise Exception(
'todo: use RSDFactor of the catalog! VelocityOffset can be wrong by factor of a if catalog has a^2 dx/dt (ms_gadget) instead of a dx/dt.'
)
galcat['Position'] = (galcat['Position'] +
galcat['VelocityOffset'] * RSD_LOS)
if cat.comm.rank == 0:
print('galcat', galcat)
print('attrs', galcat.attrs)
print('columns', galcat.columns)
print('fsat', galcat.attrs['fsat'])
return galcat
def __init__(self, githash, label, rnum, local_path, alter_path=None):
"""
loads the parameter file of the run
githash: string, abridged githash of commit under which the run was performed
label : string, label of the run
rnum : int, number of run under this label and githash
local_path: string, path under which parameter files have been stored
"""
#-------------------------------------------------------------#
params_path = os.path.join(local_path, 'runs', githash)
params_file = os.path.join(params_path, label + '%d.json' % rnum)
with open(params_file, 'r') as f:
self.params = json.load(f)
if alter_path is None:
path_name = os.path.join(self.params['output_path'],
self.params['label'] + '%d/' % rnum)
else:
path_name = os.path.join(alter_path,
self.params['label'] + '%d/' % rnum)
self.dirs = {}
for result in ['cls', 'maps', 'snapshots']:
self.dirs[result] = os.path.join(path_name, result)
#-------------------------------------------------------------#
cosmo = Planck15.match(Omega0_m=self.params['Omega_m'])
self.cosmo = cosmo.match(sigma8=self.params['sigma_8'])
self.pm = ParticleMesh(Nmesh=self.params['Nmesh'],
BoxSize=self.params['BoxSize'],
resampler='cic')
BoxSize2D = [deg / 180. * np.pi for deg in self.params['BoxSize2D']]
self.pm2D = ParticleMesh(BoxSize=BoxSize2D,
Nmesh=self.params['Nmesh2D'],
resampler='cic')
z_int = np.logspace(-8, np.log10(1500), 10000)
chis = cosmo.comoving_distance(z_int) #Mpc/h
self.z_chi_int = scipy.interpolate.interp1d(chis,
z_int,
kind=3,
bounds_error=False,
fill_value=0.)
self.theory_cls = {}
self.measured_cls = {}
print(
'Loading run with BoxSize %d, Nmesh %d, SourceRedshift %.2f, PGD %s and interpolation %s.'
% (self.params['BoxSize'][0], self.params['Nmesh'][0],
self.params['zs_source'][0], str(
self.params['PGD']), str(self.params['interpolate'])))
# count how many maps have been dumped
NN = len(os.listdir(self.dirs['maps']))
if NN < self.params['N_maps']:
print(
'less maps produces than requested. Requested:%d Produced:%d' %
(self.params['N_maps'], NN))
self.N_maps = NN
print('%d maps were produced in this run' % self.N_maps)
self.Nyquist_3D = np.pi * self.pm.Nmesh[0] / self.pm.BoxSize[0]
self.Nyquist_2D = np.pi * self.pm2D.Nmesh[0] / self.pm2D.BoxSize[0]