def test_epoch(key, surv, user_choices):
"""
Check if previously calculated GP fit survives selection cuts.
input: key, str
object type
surv, dict
dictionary of objects surviving basic cuts
keys are types, values are GP fit mean file name
user_choices, dict
output from snclass.util.read_user_input
output: my_lc, LC object
updated light curve object after checked for epoch cuts
"""
# sample a random obj in the training sample
indx = np.random.randint(0, len(surv[key]))
name = surv[key][indx]
# determine fitting method
fit_method = bool(int(user_choices['do_mcmc'][0]))
# update path to raw data
user_choices['path_to_lc'] = [name]
# read light curve raw data
raw = read_snana_lc(user_choices)
# update raw data with user choices
raw.update(user_choices)
# set number of samples to 0 (we are only interested in the mean for now)
raw['n_samples'] = ['0']
# initiate light curve object
my_lc = LC(raw, user_choices)
screen('Fitting SN' + raw['SNID:'][0], user_choices)
# load GP fit
my_lc.load_fit_GP(user_choices['samples_dir'][0] + '/DES_SN' + raw['SNID:'][0] + '_mean.dat')
# normalize
my_lc.normalize()
# shift to peak mjd
my_lc.mjd_shift()
# check epoch requirements
my_lc.check_epoch()
return my_lc, raw
def select_GP(params, user_choices):
"""
Select original objs to build a synthetic spectroscopic sample.
input: params, dict
output from set_paramameters
user_choices, dict
output from snclass.util.read_user_input
"""
from snclass.util import translate_snid, read_snana_lc
from snclass.functions import screen
from snclass.treat_lc import LC
from snclass.fit_lc_gptools import save_result
import os
import numpy as np
import sys
# set reference filter
if user_choices['ref_filter'][0] == 'None':
fil_choice = None
else:
fil_choice = user_choices['ref_filter'][0]
# select extra GP realizations in order to construct
# a representative spec sample
for key in params['draw_spec_samples'].keys():
cont = 0
fail = 0
# check if there are existing objs in this sample
screen('... Check existing objs', user_choices)
ready = []
for obj in params['surv_spec_names'][key]:
obj_id = translate_snid(obj)
for j in xrange(params['draw_spec_samples'][key]):
mean_file = params['synthetic_dir'] + '/' + \
user_choices['file_root'][0] + str(j) + \
'X' + obj_id + '_mean.dat'
if os.path.isfile(mean_file) and mean_file not in ready:
cont = cont + 1
ready.append(mean_file)
screen('Found ready SN ' + str(cont) + 'X' + \
obj_id, user_choices)
while cont < params['draw_spec_samples'][key]:
# draw one of the objs in the spec sample
indx = np.random.randint(0, params['spec_pop'][key])
name = params['surv_spec_names'][key][indx]
user_choices['path_to_lc'] = [name]
# read light curve raw data
raw = read_snana_lc(user_choices)
if os.path.isfile(params['fitted_data_dir'] + user_choices['file_root'][0] + \
raw['SNID:'][0] + '_samples.dat'):
# initiate light curve object
my_lc = LC(raw, user_choices)
screen('Loading SN' + raw['SNID:'][0], user_choices)
# load GP fit
my_lc.user_choices['n_samples'] = ['100']
my_lc.user_choices['samples_dir'] = [params['fitted_data_dir']]
my_lc.load_fit_GP(params['fitted_data_dir'] + user_choices['file_root'][0] + \
raw['SNID:'][0] + '_mean.dat')
l1 = [
1 if len(my_lc.fitted['GP_fit'][fil]) > 0 else 0
for fil in user_choices['filters']
]
if sum(l1) == len(user_choices['filters']):
# normalize
my_lc.normalize(samples=True, ref_filter=fil_choice)
# shift to peak mjd
my_lc.mjd_shift()
# check epoch requirements
my_lc.check_epoch()
if my_lc.epoch_cuts:
screen('... Passed epoch cuts', user_choices)
screen('... ... This is SN type ' + raw[user_choices['type_flag'][0]][0] + \
' number ' + str(cont + 1) + ' of ' +
str(params['draw_spec_samples'][key]), user_choices)
# draw one realization
size = len(my_lc.fitted['realizations'][
user_choices['filters'][0]])
indx2 = np.random.randint(0, size)
for fil in user_choices['filters']:
print '... ... ... filter ' + fil
raw['GP_fit'][fil] = my_lc.fitted['realizations'][
fil][indx2]
raw['GP_std'][fil] = my_lc.fitted['GP_std'][fil]
raw['xarr'][fil] = my_lc.fitted['xarr'][fil]
# set new file root
raw['file_root'] = [user_choices['file_root'][0] + \
str(cont) + 'X']
raw['samples_dir'] = [params['synthetic_dir'] + '/']
save_result(raw)
# check epoch for this realization
new_lc = LC(raw, user_choices)
new_lc.load_fit_GP(params['synthetic_dir'] + '/' + \
user_choices['file_root'][0] + str(cont) + \
'X' + raw['SNID:'][0] + '_mean.dat')
new_lc.normalize(ref_filter=fil_choice)
new_lc.mjd_shift()
new_lc.check_epoch()
if new_lc.epoch_cuts:
cont = cont + 1
else:
screen('Samples failed to pass epoch cuts!\n',
user_choices)
os.remove(params['synthetic_dir'] + '/' +
user_choices['file_root'][0] + str(cont) + \
'X' + raw['SNID:'][0] + '_mean.dat')
print '\n'
else:
screen('Failed to pass epoch cuts!\n', user_choices)
fail = fail + 1
if fail > 10 * params['spec_pop'][key]:
cont = 100000
sys.exit()
def build_sample(params):
"""
Build a directory holding all raw data passing selection cuts.
input: params, dict
keywords: 'raw_dir' -> new directory to be created
'photo_dir' -> photometric LC fitted with GP
'spec_dir' -> sectroscopic LC fitted with GP
'user_choices' -> output from
snclass.util.read_user_input
"""
import shutil
from snclass.util import read_user_input, read_snana_lc, translate_snid
from snclass.treat_lc import LC
from snclass.functions import screen
# create data directory
if not os.path.isdir(params['raw_dir']):
os.makedirs(params['raw_dir'])
# read fitted light curves
photo_list = os.listdir(params['photo_dir'])
spec_list = os.listdir(params['spec_dir'])
# build filter list
fil_list = params['user_choices']['filters'][0]
for i in xrange(1, len(params['user_choices']['filters'])):
fil_list = fil_list + params['user_choices']['filters'][i]
for sn_set in [photo_list, spec_list]:
for obj in sn_set:
if 'samples' in obj and '~' not in obj and 'Y' not in obj:
screen(obj, params['user_choices'])
rname = translate_snid(obj)[0]
params['user_choices']['path_to_lc'] = [rname]
params['user_choices']['n_samples'] = ['0']
# read raw data
raw = read_snana_lc(params['user_choices'])
new_lc = LC(raw, params['user_choices'])
# load GP fit
if sn_set == photo_list:
new_lc.load_fit_GP(photo_dir +
params['user_choices']['file_root'][0] +
raw['SNID:'][0] + '_mean.dat')
else:
new_lc.load_fit_GP(spec_dir +
params['user_choices']['file_root'][0] +
raw['SNID:'][0] + '_mean.dat')
l1 = [
1 if len(new_lc.fitted['GP_fit'][fil]) > 0 else 0
for fil in params['user_choices']['filters']
]
if sum(l1) == len(params['user_choices']['filters']):
# treat light curve
new_lc.normalize(ref_filter= \
params['user_choices']['ref_filter'][0])
new_lc.mjd_shift()
new_lc.check_basic()
new_lc.check_epoch()
# check epoch cuts
data_path = params['user_choices']['path_to_obs'][0]
if new_lc.epoch_cuts:
shutil.copy2(data_path + rname, raw_dir + rname)
else:
screen('... SN' + raw['SNID:'][0] + \
' fail to pass epoch cuts!',
params['user_choices'])
def classify_test(test_name,
matrix,
user_input,
test_dir='test_samples/',
csamples=True):
"""
Classify one photometric supernova using a trained KernelPCA matrix.
input: test_name, str
name of mean GP fit file
matrix, snclass.matrix.DataMatrix object
trained KernelPCA matrix
user_input, dict
output from snclass.util.read_user_input
test_dir, str, optional
name of directory to store samples from test light curve
Default is 'test_samples/'
csamples, bool, optional
If True, fit GP object and generate sample file as output
otherwise reads samples from file
Default is True
return: new_lc, snclass.treat_lc.LC object
updated with test projections and probability of being Ia
"""
# update path to raw light curve
user_input['path_to_lc'] = [translate_snid(test_name, 'FLUXCAL')[0]]
# store number of samples for latter tests
nsamples = user_input['n_samples'][0]
# reset the number of samples for preliminary tests
user_input['n_samples'] = ['0']
# read raw data
raw = read_snana_lc(user_input)
# load GP fit and test epoch cuts
new_lc = LC(raw, user_input)
new_lc.load_fit_GP(user_input['samples_dir'][0] + test_name)
new_lc.normalize()
new_lc.mjd_shift()
new_lc.check_epoch()
if new_lc.epoch_cuts:
# update test sample directory
user_input['samples_dir'] = [test_dir]
# update user choices
new_lc.user_choices = user_input
# update number of samples
new_lc.user_choices['n_samples'] = [nsamples]
# fit GP or normalize/shift fitted mean
test_matrix = test_samples(new_lc, calc_samples=bool(csamples))
# project test
new_lc.test_proj = matrix.transf_test.transform(test_matrix)
# classify
new_lc.new_label = nneighbor(new_lc.test_proj, matrix.low_dim_matrix,
matrix.sntype, matrix.user_choices)
if csamples:
new_lc.prob_Ia = sum([1 for item in new_label if item == '0'
]) / float(nsamples)
return new_lc
else:
return None
def check_file(self, filename, epoch=True, ref_filter=None):
"""
Construct one line of the data matrix.
input: filename, str
file of raw data for 1 supernova
epoch, bool - optional
If true, check if SN satisfies epoch cuts
Default is True
ref_filter, str - optional
Reference filter for peak MJD calculation
Default is None
"""
screen('Fitting ' + filename, self.user_choices)
# translate identifier
self.user_choices['path_to_lc'] = [
translate_snid(filename, self.user_choices['photon_flag'][0])[0]
]
# read light curve raw data
raw = read_snana_lc(self.user_choices)
# initiate light curve object
lc_obj = LC(raw, self.user_choices)
# load GP fit
lc_obj.load_fit_GP(self.user_choices['samples_dir'][0] + filename)
# normalize
lc_obj.normalize(ref_filter=ref_filter)
# shift to peak mjd
lc_obj.mjd_shift()
if epoch:
# check epoch requirements
lc_obj.check_epoch()
else:
lc_obj.epoch_cuts = True
if lc_obj.epoch_cuts:
# build data matrix lines
lc_obj.build_steps()
# store
obj_line = []
for fil in self.user_choices['filters']:
for item in lc_obj.flux_for_matrix[fil]:
obj_line.append(item)
rflag = self.user_choices['redshift_flag'][0]
redshift = raw[rflag][0]
obj_class = raw[self.user_choices['type_flag'][0]][0]
self.snid.append(raw['SNID:'][0])
return obj_line, redshift, obj_class
else:
screen('... Failed to pass epoch cuts!', self.user_choices)
screen('\n', self.user_choices)
return None
def main(args):
"""
Construct 'fake' training.
Use photometric simulated sample to guess proportions between
spectroscopic sample classes.
"""
# read user input
user_choices = read_user_input(args.input)
##########################################################################
# Spec
# build complete spec list
screen('Building spectroscopic sample.', user_choices)
user_choices['sample_cut'] = ['1', '3', '21', '22', '23', '32', '33']
spec_list = choose_sn(user_choices, output_file='spec_' + \
user_choices['epoch_cut'][0] + '_' + \
user_choices['epoch_cut'][1] '.list')
# check population according to type
spec_pop = check_pop('spec.list', user_choices)
# count spec classes surviving selection cuts
surv_spec = check_fitted(user_choices['samples_dir'][0], user_choices)
##########################################################################
# Photo
#build complete photo list
screen('Build photometric samples.', user_choices)
user_choices['sample_cut'] = ['-9']
name_plist = 'photo_' + user_choices['epoch_cut'][0] + '_' + \
user_choices['epoch_cut'][1] '.list'
photo_list = choose_sn(user_choices, output_file=name_plist)
# check population according to type
photo_pop = check_pop(name_plist, user_choices)
photo_frac = calc_fraction(photo_pop)
##########################################################################
# Building fake training sample
screen('Checking compatibility.', user_choices)
# construct number of SN expected in spec sample
spec_num = {}
for item in photo_pop.keys():
spec_num[item] = int(np.round(sum(spec_pop.values()) * photo_frac[item]))
#construct synthetic spec data directory
synthetic_dir = args.dir
if not os.path.isdir(synthetic_dir):
os.makedirs(synthetic_dir)
# collect gp objects
gp_objs = {}
#run through all types
for key in spec_num.keys():
# start cont of failed tries
fail = 0
if key in surv_spec.keys():
# check which objs and samples were already calculated
ready = check_mean_GP(key, surv_spec, spec_num, user_choices, synthetic_dir)
cont = len(ready)
while cont < spec_num[key]:
my_lc = test_epoch(key, surv_spec, user_choices)
mean_name = synthetic_dir + '/' + \
user_choices['file_root'][0] + my_lc[1]['SNID:'][0] + \
'_mean.dat'
screen('... This is SN type ' + my_lc[1]['SIM_NON1a:'][0] + \
' number ' + str(len(ready) + 1) + ' of ' +
str(spec_num[key]), user_choices)
if my_lc[0].epoch_cuts and mean_name not in ready:
ready.append(mean_name)
cont = len(ready)
shutil.copy2(user_choices['samples_dir'][0] + '/' + \
user_choices['file_root'][0] + \
my_lc[1]['SNID:'][0] + '_mean.dat',
mean_name)
screen('\n', user_choices)
else:
# build GP object
raw, gp_objs = setup_gp(my_lc[1], user_choices, gp_objs)
# set new file root
raw['file_root'] = [user_choices['file_root'][0] + \
str(len(ready)) + 'X']
raw['samples_dir'] = [synthetic_dir + '/']
save_result(raw)
# check epoch for this realization
new_lc = LC(raw, user_choices)
new_lc.load_fit_GP(synthetic_dir + '/' + \
user_choices['file_root'][0] + str(cont) + \
'X' + raw['SNID:'][0] + '_mean.dat')
new_lc.normalize()
new_lc.mjd_shift()
new_lc.check_epoch()
if new_lc.epoch_cuts:
ready.append(synthetic_dir + '/' +
user_choices['file_root'][0] + str(cont) + \
'X' + raw['SNID:'][0] + '_mean.dat')
cont = len(ready)
screen('\n', user_choices)
else:
os.remove(synthetic_dir + '/' + \
user_choices['file_root'][0] + str(cont) + \
'X' + raw['SNID:'][0] + '_mean.dat')
fail = fail + 1
screen(str(fail) + ' samples failed to pass epoch cuts!\n', user_choices)
if fail > 10 * spec_num[key]:
cont = 100000