def testModel():
with gl.name_scope('signal'):
with gl.name_scope('g1'):
m1 = gl.Parameter(5265., name = 'mean', fixed = True)
s1 = gl.Parameter(67.86, name = 'sigma', fixed = True)
signalGauss1 = gl.Gaussian({'mean' : m1, 'sigma' : s1})
with gl.name_scope('g2'):
m2 = gl.Parameter(5257., name = 'mean', fixed = True)
s2 = gl.Parameter(185.9, name = 'sigma', fixed = True)
signalGauss2 = gl.Gaussian({'mean' : m2, 'sigma' : s2})
y1 = gl.Parameter(1000., name = 'g1Yield', minVal = -100., maxVal = 2000.)
y2 = gl.Parameter(1000., name = 'g2Yield', minVal = -100., maxVal = 2000.)
fitYields = {signalGauss1.name : y1, signalGauss2.name : y2}
# f1 = gl.Parameter(0.5, name = 'frac1', minVal = 0, maxVal = 1.0)
# fitFracs = {signalGauss1.name : f1}
fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
signalModel = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
# signalModel = gl.Model(initialFitFracs = fitFracs, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
return signalModel
def simpleARGausModel(c, p, chi, sigma, nEvents):
with gl.name_scope('simpleARGausTest'):
cA = gl.Parameter(c, name='c', minVal=4800., maxVal=6000.)
pA = gl.Parameter(p, name='p', minVal=0., maxVal=5.)
chiA = gl.Parameter(chi, name='chi', minVal=0., maxVal=25.)
sigmaA = gl.Parameter(sigma, name='sigma', minVal=0., maxVal=100.)
argaus = gl.ARGaus({
'c': cA,
'chi': chiA,
'p': pA,
'sigma': sigmaA
},
minVal=4800.,
maxVal=6000.,
gridSize=1000)
argausYield = gl.Parameter(nEvents,
name='argausYield',
minVal=0.8 * nEvents,
maxVal=1.2 * nEvents)
fitYields = {argaus.name: argausYield}
fitComponents = {argaus.name: argaus}
model = gl.Model(initialFitYields=fitYields,
initialFitComponents=fitComponents,
minVal=4800.,
maxVal=6000.)
return model
def mcModel():
with gl.name_scope('mcSignal'):
with gl.name_scope('g1'):
m1 = gl.Parameter(5279., name = 'mean', minVal = 4200, maxVal = 5700) # These HAVE to be floats
s1 = gl.Parameter(30., name = 'sigma', minVal = 5, maxVal = 60) # These HAVE to be floats
signalGauss1 = gl.Gaussian({'mean' : m1, 'sigma' : s1})
with gl.name_scope('g2'):
m2 = gl.Parameter(5279., name = 'mean', minVal = 4200, maxVal = 5700) # These HAVE to be floats
s2 = gl.Parameter(90., name = 'sigma', minVal = 60, maxVal = 150) # These HAVE to be floats
signalGauss2 = gl.Gaussian({'mean' : m2, 'sigma' : s2})
f1 = gl.Parameter(0.5, name = 'frac1', minVal = 0, maxVal = 1.0)
fitFracs = {signalGauss1.name : f1}
fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
model = gl.Model(initialFitFracs = fitFracs, initialFitComponents = fitComponents, minVal = 4200, maxVal = 5700)
return model
def testPhysical():
with gl.name_scope('massFit'):
with gl.name_scope("expBkg"):
yBkg = gl.Parameter(500, name = 'yieldExp', minVal = -1000)
minParam = gl.Parameter(0.0, name = 'minVal', fixed = True)
maxParam = gl.Parameter(10.0, name = 'maxVal', fixed = True)
aExp = gl.Parameter(-0.2, name = 'aExp')
# expBkg = gl.Exponential({'a' : aExp, 'min' : gl.Parameter(0.0, fixed = True), 'max' : gl.Parameter(10.0, fixed = True)})
expBkg = gl.Exponential({'a' : aExp, 'min' : minParam, 'max' : maxParam})
with gl.name_scope("gaussSignal"):
ySig = gl.Parameter(500, name = 'yieldSig', minVal = -1000)
m = gl.Parameter(5.0, name = 'mean')
s = gl.Parameter(1.0, name = 'sigma')
g = gl.Gaussian({'mean' : m, 'sigma' : s})
initialFitYields = {g.name : ySig, expBkg.name : yBkg}
initialFitComponents = {g.name : g, expBkg.name : expBkg}
model = gl.Model(initialFitYields = initialFitYields, initialFitComponents = initialFitComponents, minVal = 0., maxVal = 10.)
data = model.sample(0, 10)
# plt.hist(data, bins = 50, histtype = 'stepfilled')
# plt.savefig('textExp.pdf')
fitter = gl.Fitter(model, backend = 'minuit')
res = fitter.fit(data, verbose = True)#, nIterations = 5000)
# samples = res.chain[:, 500:, :].reshape((-1, model.getNFloatingParameters()))
plotter = gl.Plotter(model, data)
plotter.plotDataModel(nDataBins = 30)
plt.savefig('testPlot.pdf')
plt.clf()
def crossfeedModel(signalYield, crossfeedYield, cbYield):
with gl.name_scope('pidToy'):
minParam = gl.Parameter(4800., name = 'minVal', fixed = True)
maxParam = gl.Parameter(6000., name = 'maxVal', fixed = True)
with gl.name_scope('signal'):
with gl.name_scope('g1'):
m1 = gl.Parameter(5265., name = 'mean', fixed = True)
s1 = gl.Parameter(67.86, name = 'sigma', fixed = True)
signalGauss1 = gl.Gaussian({'mean' : m1, 'sigma' : s1})
with gl.name_scope('g2'):
m2 = gl.Parameter(5257., name = 'mean', fixed = True)
s2 = gl.Parameter(185.9, name = 'sigma', fixed = True)
signalGauss2 = gl.Gaussian({'mean' : m2, 'sigma' : s2})
y1 = gl.Parameter(signalYield // 2, name = 'g1Yield', minVal = -10., maxVal = signalYield)
y2 = gl.Parameter(signalYield // 2, name = 'g2Yield', minVal = -10., maxVal = signalYield)
#
# # Would be good to have these represented using fractions, i.e., y1/y2
#
# fitYields = {signalGauss1.name : y1, signalGauss2.name : y2}
# fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
# #
# signalModel = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
f1 = gl.Parameter(0.5665, name = 'frac1', fixed = True)
fitFracs = {signalGauss1.name : f1}
fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
signalModel = gl.Model(initialFitFracs = fitFracs, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
with gl.name_scope('crossfeedKpipi'):
with gl.name_scope('g1'):
m1_cf = gl.Parameter(5021., name = 'mean', fixed = True)
s1_cf = gl.Parameter(300., name = 'sigma', fixed = True)
crossfeedKpipiGauss1 = gl.Gaussian({'mean' : m1_cf, 'sigma' : s1_cf})
with gl.name_scope('g2'):
m2_cf = gl.Parameter(5174., name = 'mean', fixed = True)
s2_cf = gl.Parameter(97.77, name = 'sigma', fixed = True)
crossfeedKpipiGauss2 = gl.Gaussian({'mean' : m2_cf, 'sigma' : s2_cf})
y1 = gl.Parameter(crossfeedYield * 0.5, name = 'g1Yield', minVal = -10.)
y2 = gl.Parameter(crossfeedYield * 0.5, name = 'g2Yield', minVal = -10.)
# Would be good to have these represented using fractions, i.e., y1/y2
# fitYields = {crossfeedKpipiGauss1.name : y1, crossfeedKpipiGauss2.name : y2}
# fitComponents = {crossfeedKpipiGauss1.name : crossfeedKpipiGauss1, crossfeedKpipiGauss2.name : crossfeedKpipiGauss2}
#
# cfModel = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
f1_cf = gl.Parameter(0.3373, name = 'frac1', fixed = True)
fitFracs = {crossfeedKpipiGauss1.name : f1_cf}
fitComponents = {crossfeedKpipiGauss1.name : crossfeedKpipiGauss1, crossfeedKpipiGauss2.name : crossfeedKpipiGauss2}
cfModel = gl.Model(initialFitFracs = fitFracs, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
with gl.name_scope('combinatorial'):
a = gl.Parameter(-0.003, name = 'aExp', fixed = True) # ??
combinatorial = gl.Exponential({'a' : a, 'min' : minParam, 'max' : maxParam})
# ySig1 = gl.Parameter(signalYield // 2, name = 'sigYield1', minVal = -10.)
# ySig2 = gl.Parameter(signalYield // 2, name = 'sigYield2', minVal = -10.)
# yCF1 = gl.Parameter(crossfeedYield // 2, name = 'cfYield1', minVal = -10.)
# yCF2 = gl.Parameter(crossfeedYield // 2, name = 'cfYield2', minVal = -10.)
# yBkg = gl.Parameter(cbYield, name = 'cbYield', minVal = -10.)
#
# # Have sub-models defined using fractions of components, then report yields for models in addition to yields for individual components?
# # AND FIX NORMALISATION!
#
# fitYields = {signalGauss1.name : ySig1, signalGauss2.name : ySig2, crossfeedKpipiGauss1.name : yCF1, crossfeedKpipiGauss2.name : yCF2, combinatorial.name : yBkg}
# fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2, crossfeedKpipiGauss1.name : crossfeedKpipiGauss1, crossfeedKpipiGauss2.name : crossfeedKpipiGauss2, combinatorial.name : combinatorial}
#
# model = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
ySig1 = gl.Parameter(signalYield, name = 'sigYield', minVal = -10.)
yCF1 = gl.Parameter(crossfeedYield, name = 'cfYield', minVal = -10.)
yBkg = gl.Parameter(cbYield, name = 'cbYield', minVal = -10.)
# Have sub-models defined using fractions of components, then report yields for models in addition to yields for individual components?
# AND FIX NORMALISATION!
fitYields = {signalModel.name : ySig1, cfModel.name : yCF1, combinatorial.name : yBkg}
fitComponents = {signalModel.name : signalModel, cfModel.name : cfModel, combinatorial.name : combinatorial}
model = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
return model
def testWithFracs():
data = np.concatenate((np.random.normal(-2, 1., 1000), np.random.normal(-2, 3., 1000), np.random.normal(2., 1., 1000), np.random.normal(2., 3., 1000)))
data = data[data > -10]
data = data[data < 10]
with gl.name_scope('fit'):
with gl.name_scope('signal1'):
m1_1 = gl.Parameter(-2.2, name = 'mean', minVal = -4, maxVal = 0)
s1_1 = gl.Parameter(1.0, name = 'sigma1', minVal = 0., maxVal = 3.)
s2_1 = gl.Parameter(3.0, name = 'sigma2', minVal = 1., maxVal = 5.)
f1_1 = gl.Parameter(0.75, name = 'frac1', minVal = 0.0, maxVal = 1.0)
with gl.name_scope("firstGaussian"):
# y1 = gl.Parameter(len(data) // 2, name = 'yield', minVal = 0)
g1_1 = gl.Gaussian({'mean' : m1_1, 'sigma' : s1_1})
with gl.name_scope("secondGaussian"):
# y2 = gl.Parameter(len(data) // 2, name = 'yield', minVal = 0)
g2_1 = gl.Gaussian({'mean' : m1_1, 'sigma' : s2_1})
# initialFitYields1 = {g1.name : y1, g2.name : y2}
initialFitFracs = {g1_1.name : f1_1}
initialFitComponents = {g1_1.name : g1_1, g2_1.name : g2_1}
# model = gl.Model(initialFitYields = initialFitYields1, initialFitComponents = initialFitComponents1, minVal = -10., maxVal = 10.)
model1 = gl.Model(initialFitFracs = initialFitFracs, initialFitComponents = initialFitComponents, minVal = -10., maxVal = 10.)
with gl.name_scope('signal2'):
m1_2 = gl.Parameter(5.0, name = 'mean', minVal = 0, maxVal = 10)
s1_2 = gl.Parameter(1.0, name = 'sigma1', minVal = 0., maxVal = 3.)
s2_2 = gl.Parameter(3.0, name = 'sigma2', minVal = 1., maxVal = 5.)
f1_2 = gl.Parameter(0.75, name = 'frac1', minVal = 0.0, maxVal = 1.0)
with gl.name_scope("firstGaussian"):
# y1 = gl.Parameter(len(data) // 2, name = 'yield', minVal = 0)
g1_2 = gl.Gaussian({'mean' : m1_2, 'sigma' : s1_2})
with gl.name_scope("secondGaussian"):
# y2 = gl.Parameter(len(data) // 2, name = 'yield', minVal = 0)
g2_2 = gl.Gaussian({'mean' : m1_2, 'sigma' : s2_2})
# initialFitYields1 = {g1.name : y1, g2.name : y2}
initialFitFracs = {g1_2.name : f1_2}
initialFitComponents = {g1_2.name : g1_2, g2_2.name : g2_2}
# model = gl.Model(initialFitYields = initialFitYields1, initialFitComponents = initialFitComponents1, minVal = -10., maxVal = 10.)
model2 = gl.Model(initialFitFracs = initialFitFracs, initialFitComponents = initialFitComponents, minVal = -10., maxVal = 10.)
y1 = gl.Parameter(2000, name = 'yield1')
y2 = gl.Parameter(2000, name = 'yield2')
initialFitYields = {model1.name : y1, model2.name : y2}
initialFitComponents = {model1.name : model1, model2.name : model2}
model = gl.Model(initialFitYields = initialFitYields, initialFitComponents = initialFitComponents, minVal = -10., maxVal = 10.)
# fitter = gl.Fitter(model, backend = 'minuit')
#
# res = fitter.fit(data, verbose = True)
#
# # model.setTotalYield(len(data))
# plotter = gl.Plotter(model, data)
# plotter.plotDataModel()
#
# plt.savefig('plotWithFracs.pdf')
fitter = gl.Fitter(model, backend = 'emcee')
res = fitter.fit(data, verbose = True, nIterations = 2000)
samples = res.chain[:, 1000:, :].reshape((-1, model.getNFloatingParameters()))
c = corner.corner(samples, lw = 1.0)
c.savefig('corner.pdf')
plt.clf()
# model.setTotalYield(len(data))
plotter = gl.Plotter(model, data)
plotter.plotDataModel()
plt.savefig('plotWithFracs.pdf')
def testPhysicalSimBF():
with gl.name_scope('massFit'):
minParam = gl.Parameter(0.0, name = 'minVal', fixed = True)
maxParam = gl.Parameter(10.0, name = 'maxVal', fixed = True)
m = gl.Parameter(5.0, name = 'mean')
s = gl.Parameter(1.0, name = 'sigma')
with gl.name_scope('model1'):
with gl.name_scope("expBkg"):
yBkg1 = gl.Parameter(500, name = 'yieldExp', minVal = -1000)
aExp1 = gl.Parameter(-0.2, name = 'aExp')
expBkg1 = gl.Exponential({'a' : aExp1, 'min' : minParam, 'max' : maxParam})
with gl.name_scope("gaussSignal"):
ySig1 = gl.Parameter(500, name = 'yieldSig', minVal = -1000)
g1 = gl.Gaussian({'mean' : m, 'sigma' : s})
with gl.name_scope('model2'):
with gl.name_scope("expBkg"):
yBkg2 = gl.Parameter(1500, name = 'yieldExp', minVal = -1000)
aExp2 = gl.Parameter(-0.1, name = 'aExp')
expBkg2 = gl.Exponential({'a' : aExp2, 'min' : minParam, 'max' : maxParam})
with gl.name_scope("gaussSignal"):
ySig2 = gl.Parameter(1000, name = 'yieldSig', minVal = -1000)
g2 = gl.Gaussian({'mean' : m, 'sigma' : s})
initialFitYields1 = {g1.name : ySig1, expBkg1.name : yBkg1}
initialFitComponents1 = {g1.name : g1, expBkg1.name : expBkg1}
initialFitYields2= {g2.name : ySig2, expBkg2.name : yBkg2}
initialFitComponents2 = {g2.name : g2, expBkg2.name : expBkg2}
model1 = gl.Model(initialFitYields = initialFitYields1, initialFitComponents = initialFitComponents1)
model2 = gl.Model(initialFitYields = initialFitYields2, initialFitComponents = initialFitComponents2)
data1 = model1.sample(0, 10)
data2 = model2.sample(0, 10)
model = gl.SimultaneousModel(initialFitComponents = [model1, model2])
# plt.hist(data, bins = 50, histtype = 'stepfilled')
# plt.savefig('textExp.pdf')
fitter = gl.Fitter(model, backend = 'emcee')
res = fitter.fit([data1, data2], verbose = True, nIterations = 2000)
samples = res.chain[:, 100:, :].reshape((-1, model.getNFloatingParameters()))
plotter = gl.Plotter(model1, data1)
plotter.plotDataModel(nDataBins = 50)
plt.savefig('testPlot1.pdf')
plt.clf()
plotter = gl.Plotter(model2, data2)
plotter.plotDataModel(nDataBins = 50)
plt.savefig('testPlot2.pdf')
plt.clf()
c = corner.corner(samples)#, truths = [-0.2, 500., 5.0, 1.0, 500.])
c.savefig('corner.pdf')
plt.clf()
def testToy():
# data = np.concatenate((np.random.normal(-1., 4., 1000), np.random.normal(0., 1., 1000), np.random.normal(3., 1., 1000)))
data1 = np.concatenate((np.random.normal(0., 1., 1000), np.random.normal(0., 3., 1000)))
data2 = np.concatenate((np.random.normal(0., 1., 500), np.random.normal(0., 3., 500)))
with gl.name_scope('massFit'):
m1 = gl.Parameter(0.2, name = 'mean')
s1 = gl.Parameter(1.0, name = 'sigma1')
s2 = gl.Parameter(3.0, name = 'sigma2')
with gl.name_scope("model1"):
with gl.name_scope("firstGaussian"):
y1 = gl.Parameter(len(data1) // 2, name = 'yield', minVal = 0)
g1 = gl.Gaussian({'mean' : m1, 'sigma' : s1})
with gl.name_scope("secondGaussian"):
y2 = gl.Parameter(len(data1) // 2, name = 'yield', minVal = 0)
g2 = gl.Gaussian({'mean' : m1, 'sigma' : s2})
initialFitYields1 = {g1.name : y1, g2.name : y2}
initialFitComponents1 = {g1.name : g1, g2.name : g2}
model1 = gl.Model(initialFitYields = initialFitYields1, initialFitComponents = initialFitComponents1, minVal = -10., maxVal = 10.)
with gl.name_scope("model2"):
with gl.name_scope("firstGaussian"):
y3 = gl.Parameter(len(data2) // 2, name = 'yield', minVal = 0)
with gl.name_scope("secondGaussian"):
y4 = gl.Parameter(len(data2) // 2, name = 'yield', minVal = 0)
initialFitYields2 = {g1.name : y3, g2.name : y4}
model2 = gl.Model(initialFitYields = initialFitYields2, initialFitComponents = initialFitComponents1, minVal = -10., maxVal = 10.)
models = {model1.name : model1, model2.name : model2}
model = gl.SimultaneousModel(initialFitComponents = [model1, model2])
# s1, s2 = model.sample(-10, 10)
#
# plt.hist(s1, bins = 100, histtype = 'stepfilled')
# plt.savefig('samples1.pdf')
# plt.clf()
# plt.hist(s2, bins = 100, histtype = 'stepfilled')
# plt.savefig('samples2.pdf')
fitter = gl.Fitter(model, backend = 'minuit')
res = fitter.fit([data1, data2], verbose = True)
# samples = res.chain[:, 200:, :].reshape((-1, 7))
#
# means = np.mean(samples, axis = 0)
# stds = np.std(samples, axis = 0)
#
# for i in range(7):
#
# plt.plot(samples[:,i], lw = 2.0)
# plt.savefig('samples' + str(i) + '.png')
# plt.clf()
#
# plt.hist(samples[:,i], bins = 50)
# plt.savefig('samplesHist' + str(i) + '.png')
# plt.clf()
#
# c = corner.corner(samples, truths = [0.0, 1000., 1000., 500., 500., 1., 3.])
# c.savefig('corner.pdf')
f, axarr = plt.subplots(1, 2, sharey = True)
plotter = gl.Plotter(model1, data1)
plotter.plotDataModel(ax = axarr[0])
axarr[0].set_xlabel('The things')
axarr[0].set_ylabel('How many things')
plt.xlim(np.min(data1), np.max(data1))
# axarr[0].set_yscale("log", nonposy='clip')
# axarr[0].set_ylim(1)
plotter = gl.Plotter(model2, data2)
plotter.plotDataModel(ax = axarr[1])
axarr[1].set_xlabel('The things')
plt.xlim(np.min(data2), np.max(data2))
# axarr[1].set_yscale("log", nonposy='clip')
# axarr[1].set_ylim(1)
plt.savefig('testPlot.pdf', bbox_inches = 'tight')
signalGauss1 = gl.Gaussian({'mean' : m1, 'sigma' : s1})
with gl.name_scope('g2'):
m2 = gl.Parameter(5257., name = 'mean', fixed = True)
s2 = gl.Parameter(185.9, name = 'sigma', fixed = True)
signalGauss2 = gl.Gaussian({'mean' : m2, 'sigma' : s2})
f1 = gl.Parameter(0.85, name = 'frac1', fixed = True)
fitFracs = {signalGauss1.name : f1}
fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
signalModel = gl.Model(initialFitFracs = fitFracs, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
# y1 = gl.Parameter(8000, name = 'y1', fixed = False, minVal = -100)
# y2 = gl.Parameter(2000, name = 'y2', fixed = False, minVal = -100)
# fitYields = {signalGauss1.name : y1, signalGauss2.name : y2}
# fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2}
# signalModel = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
ySig1 = gl.Parameter(1000000, name = 'sigYield', minVal = -10.)
yBkg = gl.Parameter(50000, name = 'cbYield', minVal = -10.)
# fitYields = {signalGauss1.name : y1, signalGauss2.name : y2, combinatorial.name : yBkg}
# fitComponents = {signalGauss1.name : signalGauss1, signalGauss2.name : signalGauss2, combinatorial.name : combinatorial}
