| 1 |
algomez |
1.2 |
points = [['fourtop1100'], ['fourtop900'], ['fourtop700'], ['fourtop500']]
|
| 2 |
|
|
#points = [['fourtop1100']]
|
| 3 |
algomez |
1.1 |
|
| 4 |
|
|
# for model building:
|
| 5 |
|
|
def get_model():
|
| 6 |
|
|
# Read in and build the model automatically from the histograms in the root file.
|
| 7 |
|
|
# This model will contain all shape uncertainties given according to the templates
|
| 8 |
|
|
# which also includes rate changes according to the alternate shapes.
|
| 9 |
|
|
# For more info about this model and naming conventuion, see documentation
|
| 10 |
|
|
# of build_model_from_rootfile.
|
| 11 |
algomez |
1.2 |
model = build_model_from_rootfile('/uscms/home/algomez/work/CMSSW_4_2_4/src/Yumiceva/TreeAnalyzer/test/mu_disc_templates.root')
|
| 12 |
|
|
#model = build_model_from_rootfile('/uscms/home/algomez/work/CMSSW_4_2_4/src/Yumiceva/TreeAnalyzer/test/mu_stjet_templates.root')
|
| 13 |
algomez |
1.1 |
|
| 14 |
|
|
# If the prediction histogram is zero, but data is non-zero, teh negative log-likelihood
|
| 15 |
|
|
# is infinity which causes problems for some methods. Therefore, we set all histogram
|
| 16 |
|
|
# bin entries to a small, but positive value:
|
| 17 |
|
|
model.fill_histogram_zerobins() # default is 1e-4
|
| 18 |
|
|
|
| 19 |
|
|
# define what the signal processes are. All other processes are assumed to make up the
|
| 20 |
|
|
# 'background-only' model.
|
| 21 |
algomez |
1.2 |
model.set_signal_processes('fourtop*00')
|
| 22 |
algomez |
1.1 |
|
| 23 |
|
|
# Add some lognormal rate uncertainties. The first parameter is the name of the
|
| 24 |
|
|
# uncertainty (which will also be the name of the nuisance parameter), the second
|
| 25 |
|
|
# is the 'effect' as a fraction, the third one is the process name. The fourth parameter
|
| 26 |
|
|
# is optional and denotes the channl. The default '*' means that the uncertainty applies
|
| 27 |
|
|
# to all channels in the same way.
|
| 28 |
|
|
# Note that you can use the same name for a systematic here as for a shape
|
| 29 |
|
|
# systematic. In this case, the same parameter will be used; shape and rate changes
|
| 30 |
|
|
# will be 100% correlated.
|
| 31 |
|
|
model.add_lognormal_uncertainty('lumi', math.log(1.045), '*')
|
| 32 |
|
|
model.add_lognormal_uncertainty('lepeff', math.log(1.03), '*')
|
| 33 |
|
|
model.add_lognormal_uncertainty('hlteff', math.log(1.03), '*')
|
| 34 |
|
|
model.add_lognormal_uncertainty('wjets_rate', math.log(1.5), 'wjets')
|
| 35 |
|
|
model.add_lognormal_uncertainty('zjets_rate', math.log(1.3), 'zjets')
|
| 36 |
|
|
model.add_lognormal_uncertainty('ttbar_rate', math.log(1.3), 'ttbar')
|
| 37 |
|
|
model.add_lognormal_uncertainty('singletop_rate', math.log(1.15), 'singletop')
|
| 38 |
|
|
|
| 39 |
|
|
return model
|
| 40 |
|
|
|
| 41 |
|
|
model = get_model()
|
| 42 |
|
|
|
| 43 |
|
|
# first, it is a good idea to generate a summary report to make sure everything has worked
|
| 44 |
|
|
# as expected. The summary will generate quite some information which should it make easy to spot
|
| 45 |
|
|
# errors like typos in the name of uncertainties, missing shape uncertaintie, etc.
|
| 46 |
|
|
model_summary(model)
|
| 47 |
|
|
|
| 48 |
|
|
# 2. apply the methods
|
| 49 |
|
|
|
| 50 |
|
|
# 2.a. Bayesian limits
|
| 51 |
|
|
# Calculate expected and observed Bayesian limits. For faster run time of this example,
|
| 52 |
|
|
# only make a few mass points. (Omitting the 'signal_procsses' parameter completely would
|
| 53 |
|
|
# process all signals defined as signal processes before; see Section "Common Parameters"
|
| 54 |
|
|
# on the theta auto intro doxygen page for details)
|
| 55 |
|
|
#res = ml_fit_coefficients(model, signal_processes = [''], nuisance_constraint="shape:fix")
|
| 56 |
algomez |
1.2 |
#for p in res['']['mu_disc*']:
|
| 57 |
|
|
# print('%s: %f' % (p, res['']['mu_disc*'][p]))
|
| 58 |
algomez |
1.1 |
plot_exp, plot_obs = bayesian_limits(model, signal_processes = points)
|
| 59 |
|
|
|
| 60 |
|
|
# plot_exp and plot_obs are instances of plotutil.plotdata. they contain x/y values and
|
| 61 |
|
|
# bands. You can do many things with these objects such as inspect the x/y/ban
|
| 62 |
|
|
# data, pass then to plotutil.plot routine to make pdf plots, ...
|
| 63 |
|
|
# Here, we will just create text files of the plot data. This is useful if you want
|
| 64 |
|
|
# to apply your own plotting routines or present the result in a text Table.
|
| 65 |
algomez |
1.2 |
#plot_exp.write_txt('bayesian_limits_expected_'+points[0][0]+'.txt')
|
| 66 |
|
|
plot_exp.write_txt('bayesian_limits_expected_fourtop_test.txt')
|
| 67 |
|
|
#plot_exp.write_txt('bayesian_limits_expected_stjet_fourtop.txt')
|
| 68 |
|
|
#plot_obs.write_txt('bayesian_limits_observed_'+points[0][0]+'.txt')
|
| 69 |
|
|
plot_obs.write_txt('bayesian_limits_observed_fourtop_test.txt')
|
| 70 |
|
|
#plot_obs.write_txt('bayesian_limits_observed_stjet_fourtop.txt')
|
| 71 |
algomez |
1.1 |
|
| 72 |
|
|
#plot_exp, plot_obs = cls_limits(model,ts="lhclike", signal_processes = points)
|
| 73 |
|
|
##plot_exp, plot_obs = cls_limits(model, signal_processes = points)
|
| 74 |
|
|
#plot_exp.write_txt('cls_limits_expected_'+points[0][0]+'.txt')
|
| 75 |
|
|
#plot_obs.write_txt('cls_limits_observed_'+points[0][0]+'.txt')
|
| 76 |
|
|
|
| 77 |
|
|
# model_summary, bayesian_limits, and cls_limits also write their results to the 'report' object
|
| 78 |
|
|
# which we can ask to write its results as html page to a certain directory. Use an existing, empty
|
| 79 |
|
|
# directory and point your web browser to it.
|
| 80 |
algomez |
1.2 |
#report.write_html('htmlout_'+points[0][0])
|
| 81 |
|
|
report.write_html('htmlout_fourtopBDT_test')
|
| 82 |
|
|
#report.write_html('htmlout_fourtop_stjet')
|
| 83 |
|
|
#report.write_html('htmlout_stjet_'+points[0][0])
|
| 84 |
algomez |
1.1 |
|
| 85 |
|
|
# After running theta-auto, you probably want to delete the 'analysis' directory which
|
| 86 |
|
|
# contains intermediate results. Keeping it avoids re-running theta unnecessarily for unchanged configurations
|
| 87 |
|
|
# (e.g., because you just want to change the plot). However, this directory can grow very large over time.
|
