def testSimpleARGaus():
print('testSimpleARGaus')
# Test generating and fitting back with the same model
model = simpleARGausModel(5400., 1.1, 8., 30., 10000.)
dataGen = model.sample(minVal=4800., maxVal=6000.)
plt.hist(dataGen, bins=150)
plt.savefig('dataHist.pdf')
plt.clf()
print('Fitting')
fitter = gl.Fitter(model, backend='minuit')
res = fitter.fit(dataGen, verbose=True)
from pprint import pprint
pprint(model.parameters)
plotter = gl.Plotter(model, dataGen)
plotter.plotDataModel(nDataBins=100)
plt.savefig('simpleARGausTest.pdf')
plt.clf()
exit(0)
fitterB = gl.Fitter(model, backend='emcee')
res = fitterB.fit(dataGen, verbose=True, nIterations=10000,
nWalkers=64) # * nparams
import corner
fig = plt.figure(figsize=(16, 12))
samples = res.chain[:, 1000:, :].reshape((-1, 5))
c = corner.corner(samples, lw=1.0)
c.savefig('argaus-corner.pdf')
plt.clf()
pprint(model.parameters)
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()
from model import Model
y = gl.Parameter(
10000.,
name='yield',
minVal=5000.,
maxVal=15000.,
)
m = Model(name='model',
initialFitYields={'yield': y},
initialFitComponents={'a': a})
from fitter import Fitter
fitter = Fitter(m, backend='minuit')
res = fitter.fit(data, verbose=True)
from plotter import Plotter
plotter = gl.Plotter(m, data)
plotter.plotDataModel(nDataBins=100)
plt.savefig('argaus_fit.pdf')
plt.clf()
# x = np.linspace(5000, 5500, 1000)
# l = a.prob(x)
# plt.plot(x, l, lw = 1.0)
# plt.savefig('argaus_plot.pdf')
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')
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}
fitYields = {signalModel.name : ySig1, combinatorial.name : yBkg}
fitComponents = {signalModel.name : signalModel, combinatorial.name : combinatorial}
model = gl.Model(initialFitYields = fitYields, initialFitComponents = fitComponents, minVal = 4800., maxVal = 6000.)
data = model.sample(minVal = 4800., maxVal = 6000.)
fitter = gl.Fitter(model, backend = 'minuit')
res = fitter.fit(data, verbose = False, nIterations = 250)
pprint(model)
plotter = gl.Plotter(model, data)
plotter.plotDataModel(nDataBins = 50)
plt.savefig('testData.pdf')
plt.clf()
#
# fitter = gl.Fitter(model, backend = 'emcee')
# res = fitter.fit(data, verbose = True, nIterations = 250, nWalkers = 10)
#
# plotter = gl.Plotter(model, data)
# plotter.plotDataModel(nDataBins = 50)
# plt.savefig('testData.pdf')
# plt.clf()
#
# samples = res.chain[:, 50:, :].reshape((-1, model.getNFloatingParameters()))
# c = corner.corner(samples, lw = 1.0)
# c.savefig('corner.pdf')