class AeffServiceMC:
'''
Takes the weighted effective area files, and creates a dictionary
of the 2D effective area in terms of energy and coszen, for each
flavor (nue,nue_bar,numu,...) and interaction type (CC, NC)
'''
def __init__(self, ebins, czbins, aeff_weight_file=None, **kwargs):
self.ebins = ebins
self.czbins = czbins
logging.info('Initializing AeffServiceMC...')
logging.info('Opening file: %s' % (aeff_weight_file))
try:
fh = h5py.File(find_resource(aeff_weight_file), 'r')
except IOError, e:
logging.error("Unable to open aeff_weight_file %s" %
aeff_weight_file)
logging.error(e)
sys.exit(1)
self.aeff_dict = {}
logging.info("Creating effective area dict...")
for flavor in [
'nue', 'nue_bar', 'numu', 'numu_bar', 'nutau', 'nutau_bar'
]:
flavor_dict = {}
logging.debug("Working on %s effective areas" % flavor)
for int_type in ['cc', 'nc']:
weighted_aeff = np.array(fh[flavor + '/' + int_type +
'/weighted_aeff'])
true_energy = np.array(fh[flavor + '/' + int_type +
'/true_energy'])
true_coszen = np.array(fh[flavor + '/' + int_type +
'/true_coszen'])
bins = (self.ebins, self.czbins)
aeff_hist, _, _ = np.histogram2d(true_energy,
true_coszen,
weights=weighted_aeff,
bins=bins)
# Divide by bin widths to convert to aeff:
ebin_sizes = get_bin_sizes(ebins)
czbin_sizes = 2.0 * np.pi * get_bin_sizes(czbins)
bin_sizes = np.meshgrid(czbin_sizes, ebin_sizes)
aeff_hist /= np.abs(bin_sizes[0] * bin_sizes[1])
flavor_dict[int_type] = aeff_hist
self.aeff_dict[flavor] = flavor_dict
return
def get_aeff1D(data,cuts_list,ebins,files_per_run,mcnu='MCNeutrino',
nc=False,solid_angle=None):
'''
Return 1D Aeff directly from simulations (using OneWeight) as a
histogram, including the error bars.
\params:
* data - data table from a hdf5 PyTables file.
* cuts_list - np array of indices for events passing selection cuts
* ebins - energy bin edges in GeV that the Aeff histogram
will be formed from
* files_per_run - file number normalization to correctly calculate
simulation weight for Aeff. Should be number of simulation files
per run for flavour.
* mcnu - Monte Carlo neutrino field/key in hdf5 file to use for MC
true information.
* solid_angle - total solid angle to integrate over. Most of the
time, we only calculate Aeff for upgoing events, so we do half
the 4pi solid angle of the whole sky, which would be 2.0*np.pi
* nc - set this flag to true if we are using NC files.
'''
#if not nc:
nfiles = len(set(data.root.I3EventHeader.col('Run')))*files_per_run
logging.info("runs: %s"%str(set(data.root.I3EventHeader.col('Run'))))
logging.info("num runs: %d"%len(set(data.root.I3EventHeader.col('Run'))))
total_events = data.root.I3MCWeightDict.col('NEvents')[cuts_list]*nfiles/2.0
# NOTE: solid_angle should be coordinated with get_arb_cuts(cut_sim_down=bool)
sim_wt_array = (data.root.I3MCWeightDict.col('OneWeight')[cuts_list]/
total_events/solid_angle)
# Not sure why nu_<> cc needs __getattr__ and only NC combined
# file needs __getattribute__??
try:
egy_array = data.root.__getattr__(mcnu).col('energy')[cuts_list]
except:
egy_array = data.root.__getattribute__(mcnu).col('energy')[cuts_list]
aeff,xedges = np.histogram(egy_array,weights=sim_wt_array,bins=ebins)
egy_bin_widths = get_bin_sizes(ebins)
aeff /= egy_bin_widths
aeff*=1.0e-4 # [cm^2] -> [m^2]
unweighted_events,xedges = np.histogram(egy_array,bins=ebins)
# So that the error calculation comes out right without divide by zeros...
unweighted_events = [1. if val < 1.0 else val for val in unweighted_events]
aeff_error = np.divide(aeff,np.sqrt(unweighted_events))
logging.debug("percent error on aeff: %s"%str(np.nan_to_num(aeff/aeff_error)))
return aeff,aeff_error,xedges
def get_flux(self, ebins, czbins, prim):
'''Get the flux in units [m^-2 s^-1] for the given
bin edges in energy and cos(zenith) and the primary.'''
#Evaluate the flux at the bin centers
evals = get_bin_centers(ebins)
czvals = get_bin_centers(czbins)
# Get the spline interpolation, which is in
# log(flux) as function of log(E), cos(zenith)
return_table = bisplev(np.log10(evals), czvals, self.spline_dict[prim])
return_table = np.power(10., return_table).T
#Flux is given per sr and GeV, so we need to multiply
#by bin width in both dimensions
#Get the bin size in both dimensions
ebin_sizes = get_bin_sizes(ebins)
czbin_sizes = 2.*np.pi*get_bin_sizes(czbins)
bin_sizes = np.meshgrid(ebin_sizes, czbin_sizes)
return_table *= np.abs(bin_sizes[0]*bin_sizes[1])
return return_table.T
def get_flux(self, ebins, czbins, prim):
"""Get the flux in units [m^-2 s^-1] for the given
bin edges in energy and cos(zenith) and the primary."""
#Evaluate the flux at the bin centers
evals = get_bin_centers(ebins)
czvals = get_bin_centers(czbins)
# Get the spline interpolation, which is in
# log(flux) as function of log(E), cos(zenith)
return_table = bisplev(np.log10(evals), czvals, self.spline_dict[prim])
return_table = np.power(10., return_table).T
#Flux is given per sr and GeV, so we need to multiply
#by bin width in both dimensions
#Get the bin size in both dimensions
ebin_sizes = get_bin_sizes(ebins)
czbin_sizes = 2. * np.pi * get_bin_sizes(czbins)
bin_sizes = np.meshgrid(ebin_sizes, czbin_sizes)
return_table *= np.abs(bin_sizes[0] * bin_sizes[1])
return return_table.T
def _get_reco_kernels(self, flipback=True,
e_reco_scale=None, cz_reco_scale=None,
**kwargs):
"""
Use the parametrization functions to calculate the actual reco
kernels (i.e. 4D histograms). If flipback==True, the zenith angle
part that goes below the zenith will be mirrored back in.
"""
if all([hasattr(self, 'kernels'), e_reco_scale==1., cz_reco_scale==1.]):
logging.info('Using existing kernels for reconstruction')
return self.kernels
logging.info('Creating parametrized reconstruction kernels')
# get binning information
evals, esizes = get_bin_centers(self.ebins), get_bin_sizes(self.ebins)
czvals, czsizes = get_bin_centers(self.czbins), get_bin_sizes(self.czbins)
czbins = self.czbins
n_e, n_cz = len(evals), len(czvals)
# prepare for folding back at lower edge
if not is_linear(self.czbins):
logging.warn("cos(zenith) bins have different "
"sizes! Unable to fold around edge "
"of histogram, will not do that.")
flipback = False
if flipback:
czvals = np.append(czvals-(self.czbins[-1]-self.czbins[0]),
czvals)
czbins = np.append(czbins[:-1]-(self.czbins[-1]-self.czbins[0]),
czbins)
czsizes = np.append(czsizes, czsizes)
# get properly scaled parametrization, initialize kernels
parametrization = self.apply_reco_scales(e_reco_scale, cz_reco_scale)
kernel_dict = {}
for flavour in parametrization:
kernel_dict[flavour] = {}
for int_type in ['cc', 'nc']:
logging.debug('Calculating parametrized reconstruction kernel for %s %s'
%(flavour, int_type))
# create empty kernel
kernel = np.zeros((n_e, n_cz, n_e, n_cz))
# quick handle to parametrization
e_pars = parametrization[flavour][int_type]['energy']
cz_pars = parametrization[flavour][int_type]['coszen']
# loop over every bin in true (energy, coszen)
for (i, j) in itertools.product(range(n_e), range(n_cz)):
e_kern_cdf = double_gauss(self.ebins,
loc1=e_pars['loc1'][i,j]+evals[i],
width1=e_pars['width1'][i,j],
loc2=e_pars['loc2'][i,j]+evals[i],
width2=e_pars['width2'][i,j],
fraction=e_pars['fraction'][i,j])
e_kern_int = np.sum(e_kern_cdf)
offset = n_cz if flipback else 0
cz_kern_cdf = double_gauss(czbins,
loc1=cz_pars['loc1'][i,j]+czvals[j+offset],
width1=cz_pars['width1'][i,j],
loc2=cz_pars['loc2'][i,j]+czvals[j+offset],
width2=cz_pars['width2'][i,j],
fraction=cz_pars['fraction'][i,j])
cz_kern_int = np.sum(cz_kern_cdf)
if flipback:
# fold back
cz_kern_cdf = cz_kern_cdf[:len(czbins)/2][::-1] + cz_kern_cdf[len(czbins)/2:]
kernel[i,j] = np.outer(e_kern_cdf, cz_kern_cdf)
kernel_dict[flavour][int_type] = copy(kernel)
kernel_dict['ebins'] = self.ebins
kernel_dict['czbins'] = self.czbins
return kernel_dict
def _get_reco_kernels(self,
flipback=True,
e_reco_scale=None,
cz_reco_scale=None,
**kwargs):
"""
Use the parametrization functions to calculate the actual reco
kernels (i.e. 4D histograms). If flipback==True, the zenith angle
part that goes below the zenith will be mirrored back in.
"""
if all([
hasattr(self, 'kernels'), e_reco_scale == 1.,
cz_reco_scale == 1.
]):
logging.info('Using existing kernels for reconstruction')
return self.kernels
logging.info('Creating parametrized reconstruction kernels')
# get binning information
evals, esizes = get_bin_centers(self.ebins), get_bin_sizes(self.ebins)
czvals, czsizes = get_bin_centers(self.czbins), get_bin_sizes(
self.czbins)
czbins = self.czbins
n_e, n_cz = len(evals), len(czvals)
# prepare for folding back at lower edge
if not is_linear(self.czbins):
logging.warn("cos(zenith) bins have different "
"sizes! Unable to fold around edge "
"of histogram, will not do that.")
flipback = False
if flipback:
czvals = np.append(czvals - (self.czbins[-1] - self.czbins[0]),
czvals)
czbins = np.append(
czbins[:-1] - (self.czbins[-1] - self.czbins[0]), czbins)
czsizes = np.append(czsizes, czsizes)
# get properly scaled parametrization, initialize kernels
parametrization = self.apply_reco_scales(e_reco_scale, cz_reco_scale)
kernel_dict = {}
for flavour in parametrization:
kernel_dict[flavour] = {}
for int_type in ['cc', 'nc']:
logging.debug(
'Calculating parametrized reconstruction kernel for %s %s'
% (flavour, int_type))
# create empty kernel
kernel = np.zeros((n_e, n_cz, n_e, n_cz))
# quick handle to parametrization
e_pars = parametrization[flavour][int_type]['energy']
cz_pars = parametrization[flavour][int_type]['coszen']
# loop over every bin in true (energy, coszen)
for (i, j) in itertools.product(range(n_e), range(n_cz)):
e_kern_cdf = double_gauss(
self.ebins,
loc1=e_pars['loc1'][i, j] + evals[i],
width1=e_pars['width1'][i, j],
loc2=e_pars['loc2'][i, j] + evals[i],
width2=e_pars['width2'][i, j],
fraction=e_pars['fraction'][i, j])
e_kern_int = np.sum(e_kern_cdf)
offset = n_cz if flipback else 0
cz_kern_cdf = double_gauss(
czbins,
loc1=cz_pars['loc1'][i, j] + czvals[j + offset],
width1=cz_pars['width1'][i, j],
loc2=cz_pars['loc2'][i, j] + czvals[j + offset],
width2=cz_pars['width2'][i, j],
fraction=cz_pars['fraction'][i, j])
cz_kern_int = np.sum(cz_kern_cdf)
if flipback:
# fold back
cz_kern_cdf = cz_kern_cdf[:len(czbins) /
2][::-1] + cz_kern_cdf[
len(czbins) / 2:]
kernel[i, j] = np.outer(e_kern_cdf, cz_kern_cdf)
kernel_dict[flavour][int_type] = copy(kernel)
kernel_dict['ebins'] = self.ebins
kernel_dict['czbins'] = self.czbins
return kernel_dict
