def test_toymc_exceptions(self):
with self.assertRaises(MonteCarloParameterMismatch):
ToyMC(self.model, GaussianGenerator(list(reversed(self.model.parameter_names)), self.mu, self.sigma))
with self.assertRaises(MonteCarloParameterMismatch):
ToyMC(self.model, GaussianGenerator(["x", "y"], self.mu, self.sigma))
with self.assertRaises(MonteCarloParameterMismatch):
ToyMC(self.model, GaussianGenerator(self.model.parameter_names + ["z"], self.mu + [3.0], self.sigma + [3.0]))
def _buildmodel(self, withosc, cachestr="testnoosc"):
#build a simple dummy model
binning = [("reco_energy", np.linspace(0.0, 5.0, num=100.0)),
("true_energy", np.linspace(0.0, 5.0, num=25.0)),
("true_nupdg", [0.0, 1.0, 2.0, 3.0, 4.0]),
("beammode", [0.0, 1.0, 2.0]),
]
signalsyst = np.array([[-4.0, 1.0, 6.0], [1.0, 1.0, 1.0]])
bkgdsyst = np.array([[1.0, 1.0, 1.0], [-4.0, 1.0, 6.0]])
def gen(N):
for _ in xrange(N):
nupdg = np.random.randint(4)
if np.random.uniform() > 0.5:
#signal
while True:
true = np.random.normal(1.0, 0.25)
if 0.0 <= true <= 5.0:
break
syst = signalsyst
else:
#bkgd
true = np.random.uniform(0.0, 5.0)
syst = bkgdsyst
while True:
reco = np.random.normal(true, 0.0001)
if 0.0 <= reco <= 5.0:
break
beammode = np.random.randint(2)
if beammode == 1:
nupdg = {0:1, 1:0, 2:3, 3:2}[nupdg]
if reco > 0.0 and true > 0.0:
yield (reco, true, nupdg, beammode), 1.0, syst
def gensk(N):
eff = 0.5
for (reco, true, nupdg, beammode), weight, syst in gen(N):
yield (reco, true, nupdg), eff*weight, weight, syst
iternd280 = gen(10**5)
itersuperk = gensk(1000)
systematics = [("signal", [-5.0, 0.0, 5.0]), ("bkgd", [-5.0, 0.0, 5.0])]
flux_error_binning = [((beamname, flavname), beambin, flavbin, [0.0, 5.0]) for beambin, beamname in enumerate(["RHC", "FHC"]) for flavbin, flavname in enumerate(["numu", "nue", "numubar", "nuebar"])]
fluxparametermap = FluxSystematics.make_flux_parameter_map(binning[1][1], flux_error_binning)
systematics = FluxAndSplineSystematics(systematics, enudim=1, nupdgdim=2, beammodedim=3, fluxparametermap=fluxparametermap)
observables = ["reco_energy"]
nd280 = BinnedSample("nd280", binning, observables, iternd280, cache_name="nd280_" + cachestr, systematics=systematics)
xsecprior = GaussianPrior(["signal", "bkgd"], [0.0, 0.0], [0.1, 0.1], seed=1231)
fluxprior = GaussianPrior([("f_%s_%s_0" %(beamname, flavname)) for beambin, beamname in enumerate(["RHC", "FHC"]) for flavbin, flavname in enumerate(["numu", "nue", "numubar", "nuebar"])], [1.0]*8, [0.1]*8, seed=23123)
prior = CombinedPrior([xsecprior, fluxprior])
if withosc:
superk = BinnedSampleWithOscillation("superk", binning, observables, itersuperk, "true_energy", "true_nupdg", 295.0, cache_name="superk_" + cachestr)
model = CombinedBinnedSample([nd280, superk])
prior = CombinedPrior([prior, OscillationParametersPrior()])
else:
model = CombinedBinnedSample([nd280])
#generate monte carlo
ratevector = model
toymc = ToyMC(ratevector, prior.generator)
lhd_data = EventRateLikelihood(model, data=toymc.asimov().vec)
lhd_prior = prior.likelihood
lhd = SumLikelihood([lhd_data, lhd_prior])
return model, toymc, lhd
def build(self,
name,
toymc,
sample,
cache_name=None,
npe=1000,
probabilitycalc=None,
fixed=None,
oscparmode=OscParMode.SINSQTHETA):
self.name = name
oscpars = toymc.generator.parameter_names[:6]
cov, mean = self._generate_covariance_with_cache(toymc=toymc,
keep=oscpars,
npe=npe,
cache_name=cache_name,
fixed=fixed)
generator = self._buildgenerator(toymc, oscpars, cov)
ratevector = self._buildratevector(oscpars,
toymc,
sample,
generator,
probabilitycalc=probabilitycalc,
oscparmode=oscparmode)
toymc = ToyMC(ratevector, generator)
return toymc, cov
def build(self,
name,
toymc,
keep=None,
cache_name=None,
npe=1000,
fixed=None):
self.name = name
try:
#assume keep is dict(parnames, splinepoints)
keys = keep.keys()
spline_points = OrderedDict()
for k in keep:
spline_points[k] = keep[k]
except AttributeError:
#assume keep is list(parnames)
spline_points = None
cov, mean = self._generate_covariance_with_cache(toymc=toymc,
keep=keep,
npe=npe,
cache_name=cache_name,
fixed=fixed)
splines = self._generate_splines_with_cache(
toymc=toymc,
nominal=toymc.asimov().vec,
keep=keep,
spline_points=spline_points,
cache_name=cache_name)
ratevector = self._buildratevector(mean, splines)
generator = self._buildgenerator(toymc, keep, cov)
toymc = ToyMC(ratevector, generator)
return toymc, cov
def _buildtestmc(self, cachestr=None):
systematics = [
("x", [-5.0, 0.0, 5.0]),
("y", [-5.0, 0.0, 5.0]),
("z", [-10.0, -5.0, 0.0, 5.0, 10.0]),
]
systematics = SplineSystematics(systematics)
random = np.random.RandomState(seed=1223)
def gen(N):
for _ in xrange(N):
coord = random.poisson(size=2)
yield coord, 1.0, [
(_smear(-4.0,
random), _smear(1.0, random), _smear(5.0, random)),
(_smear(-4.0,
random), _smear(1.0, random), _smear(5.0, random)),
(_smear(-9.0, random), _smear(-4.0, random),
_smear(1.0, random), _smear(5.0,
random), _smear(10.0, random))
]
binning = [("a", np.arange(0.0, 5.0)), ("b", np.arange(0.0, 5.0))]
observables = ["a"]
model = BinnedSample("simplemodel",
binning,
observables,
gen(10**4),
systematics=systematics,
cache_name=cachestr)
generator = GaussianGenerator(["x", "y", "z"], [1.0, 2.0, 3.0],
[1.0, 2.0, 3.0],
seed=1224)
toymc = ToyMC(model, generator)
return toymc
def test_toymcexperiment_clone(self):
toymc = ToyMC(self.model, self.gen)
exp1 = toymc()
exp2 = exp1.clone()
self.assertTrue(np.array_equal(exp1.vec, exp2.vec))
self.assertTrue(np.array_equal(exp1.pars, exp2.pars))
self.assertFalse(exp1.vec is exp2.vec)
self.assertFalse(exp1.pars is exp2.pars)
return
def test_toymc_covariance(self):
npe = 10**4
toymc = ToyMC(self.model, self.gen)
expected = np.diag(np.power(self.sigma, 2))
cov = calculate_statistics_from_toymc(toymc, Covariance(), npe)[0]
cval = cov.eval()
cerr = cov.err()
for ii, jj in itertools.product(xrange(len(self.mu)), repeat=2):
self.assertAlmostEquals(cval[ii, jj], expected[ii][jj], delta=3.0*cerr[ii, jj])
return
def test_toymc_stat(self):
npe = 10**4
toymc = ToyMC(self.model, self.gen)
statistics = [Mean(), StandardDeviation(), FractionalStandardDeviation()]
calculate_statistics_from_toymc(toymc, statistics, npe)
expected = [self.mu, self.sigma, np.divide(self.sigma,self.mu), ]
for s, exp in itertools.izip_longest(statistics, expected):
val = s.eval()
error = s.err()
for v, er, ex in itertools.izip_longest(val, error, exp):
self.assertAlmostEquals(v, ex, delta=3.0*er)
return
def build1flavmodel(self, iflav, jflav, mode):
systematics = [
("x", [-5.0, 0.0, 5.0]),
("y", [-5.0, 0.0, 5.0]),
("z", [-5.0, 0.0, 5.0]),
]
enubinning = self.enubinning
systematics = SplineSystematics(systematics)
random = np.random.RandomState(1222)
def gen(N, iflav=iflav, jflav=jflav):
for trueenu in self.enubincentres:
recoenu = trueenu
for flavbin in [0.0, 1.0, 2.0, 3.0]:
if flavbin == jflav:
eff = 1.0
else:
eff = 1e-9
if flavbin == iflav:
weight = 1.0
else:
weight = 1e-9
coord = (trueenu, flavbin, recoenu)
yield coord, eff * weight, weight, [(-4.0, 1.0, 6.0),
(-4.0, 1.0, 6.0),
(-4.0, 1.0, 6.0)]
binning = [("trueenu", enubinning), ("nupdg", np.arange(0.0, 5.0)),
("recoenu", enubinning)]
observables = ["recoenu"]
model = BinnedSampleWithOscillation("flatmodelwithoscillation",
binning,
observables,
gen(10**4),
enuaxis="trueenu",
flavaxis="nupdg",
distance=295.0,
systematics=systematics,
probabilitycalc=None,
oscparmode=mode)
oscgen = OscillationParametersPrior(seed=1225,
oscparmode=mode).generator
systgen = GaussianGenerator(["x", "y", "z"], [0.0, 0.0, 0.0],
[0.1, 0.1, 0.1],
seed=1226)
#toymc = ToyMC(model, GeneratorList(oscgen))
toymc = ToyMC(model, GeneratorList(oscgen, systgen))
return toymc
def _buildtestmc(self, cachestr=None):
systematics = [
("x", [-5.0, 0.0, 5.0]),
("y", [-5.0, 0.0, 5.0]),
("z", [-5.0, 0.0, 5.0]),
]
systematics = SplineSystematics(systematics)
random = np.random.RandomState(1222)
def gen(N):
for _ in xrange(N):
nupdg = random.uniform(0.0, 4.0)
trueenu = random.uniform(0.0, 5.0)
recoenu = _smear(trueenu, random)
coord = (trueenu, nupdg, recoenu)
yield coord, 1.0, 1.0, [
(_smear(-4.0,
random), _smear(1.0, random), _smear(6.0, random)),
(_smear(-4.0,
random), _smear(1.0, random), _smear(6.0, random)),
(_smear(-4.0, random), _smear(1.0,
random), _smear(6.0, random))
]
binning = [("trueenu", np.linspace(0.0, 5.0, num=10.0)),
("nupdg", np.arange(0.0, 5.0)),
("recoenu", np.linspace(0.0, 5.0, num=10.0))]
observables = ["recoenu"]
model = BinnedSampleWithOscillation("simplemodelwithoscillation",
binning,
observables,
gen(10**4),
enuaxis="trueenu",
flavaxis="nupdg",
distance=295.0,
systematics=systematics,
probabilitycalc=None)
oscgen = OscillationParametersPrior(seed=1225).generator
systgen = GaussianGenerator(["x", "y", "z"], [0.0, 0.0, 0.0],
[0.1, 0.1, 0.1],
seed=1226)
#toymc = ToyMC(model, GeneratorList(oscgen))
toymc = ToyMC(model, GeneratorList(oscgen, systgen))
return toymc
def _buildmodel(self, withosc, cachestr="testnoosc"):
#build a simple dummy model
binning = [
("reco_energy", np.linspace(0.0, 5.0, num=100.0)),
("true_energy", np.linspace(0.0, 5.0, num=25.0)),
("true_nupdg", [0.0, 1.0, 2.0, 3.0, 4.0]),
("beammode", [0.0, 1.0, 2.0]),
]
signalsyst = np.array([[-4.0, 1.0, 6.0], [1.0, 1.0, 1.0]])
bkgdsyst = np.array([[1.0, 1.0, 1.0], [-4.0, 1.0, 6.0]])
def gen(N):
for _ in xrange(N):
nupdg = np.random.randint(4)
if np.random.uniform() > 0.5:
#signal
while True:
true = np.random.normal(1.0, 0.25)
if 0.0 <= true <= 5.0:
break
syst = signalsyst
else:
#bkgd
true = np.random.uniform(0.0, 5.0)
syst = bkgdsyst
while True:
reco = np.random.normal(true, 0.0001)
if 0.0 <= reco <= 5.0:
break
beammode = np.random.randint(2)
if beammode == 1:
nupdg = {0: 1, 1: 0, 2: 3, 3: 2}[nupdg]
if reco > 0.0 and true > 0.0:
yield (reco, true, nupdg, beammode), 1.0, syst
def gensk(N):
eff = 0.5
for (reco, true, nupdg, beammode), weight, syst in gen(N):
yield (reco, true, nupdg), eff * weight, weight, syst
iternd280 = gen(10**5)
itersuperk = gensk(1000)
systematics = [("signal", [-5.0, 0.0, 5.0]), ("bkgd", [-5.0, 0.0,
5.0])]
flux_error_binning = [
((beamname, flavname), beambin, flavbin, [0.0, 5.0])
for beambin, beamname in enumerate(["RHC", "FHC"])
for flavbin, flavname in enumerate(
["numu", "nue", "numubar", "nuebar"])
]
fluxparametermap = FluxSystematics.make_flux_parameter_map(
binning[1][1], flux_error_binning)
systematics = FluxAndSplineSystematics(
systematics,
enudim=1,
nupdgdim=2,
beammodedim=3,
fluxparametermap=fluxparametermap)
observables = ["reco_energy"]
nd280 = BinnedSample("nd280",
binning,
observables,
iternd280,
cache_name="nd280_" + cachestr,
systematics=systematics)
xsecprior = GaussianPrior(["signal", "bkgd"], [0.0, 0.0], [0.1, 0.1],
seed=1231)
fluxprior = GaussianPrior(
[("f_%s_%s_0" % (beamname, flavname))
for beambin, beamname in enumerate(["RHC", "FHC"])
for flavbin, flavname in enumerate(
["numu", "nue", "numubar", "nuebar"])], [1.0] * 8, [0.1] * 8,
seed=23123)
prior = CombinedPrior([xsecprior, fluxprior])
if withosc:
superk = BinnedSampleWithOscillation("superk",
binning,
observables,
itersuperk,
"true_energy",
"true_nupdg",
295.0,
cache_name="superk_" +
cachestr)
model = CombinedBinnedSample([nd280, superk])
prior = CombinedPrior([prior, OscillationParametersPrior()])
else:
model = CombinedBinnedSample([nd280])
#generate monte carlo
ratevector = model
toymc = ToyMC(ratevector, prior.generator)
lhd_data = EventRateLikelihood(model, data=toymc.asimov().vec)
lhd_prior = prior.likelihood
lhd = SumLikelihood([lhd_data, lhd_prior])
return model, toymc, lhd
def test_toymcexperiment_infostring(self):
toymc = ToyMC(self.model, self.gen)
str(toymc) # just check no exceptions are raised
return
def test_toymc_stat(self):
npe = 1000
toymc = ToyMC(self.model, self.gen)
for _ in xrange(npe):
mc()
return