\relax 
\providecommand\zref@newlabel[2]{}
\providecommand\hyper@newdestlabel[2]{}
\providecommand\AtEndDvi@Check{}
\AtEndDvi@Check
\providecommand\HyperFirstAtBeginDocument{\AtBeginDocument}
\HyperFirstAtBeginDocument{\ifx\hyper@anchor\@undefined
\global\let\oldcontentsline\contentsline
\gdef\contentsline#1#2#3#4{\oldcontentsline{#1}{#2}{#3}}
\global\let\oldnewlabel\newlabel
\gdef\newlabel#1#2{\newlabelxx{#1}#2}
\gdef\newlabelxx#1#2#3#4#5#6{\oldnewlabel{#1}{{#2}{#3}}}
\AtEndDocument{\ifx\hyper@anchor\@undefined
\let\contentsline\oldcontentsline
\let\newlabel\oldnewlabel
\fi}
\fi}
\global\let\hyper@last\relax 
\gdef\HyperFirstAtBeginDocument#1{#1}
\providecommand\HyField@AuxAddToFields[1]{}
\providecommand\HyField@AuxAddToCoFields[2]{}
\providecommand{\mciteSetMaxWidth}[3]{\relax}
\providecommand{\mciteSetMaxCount}[3]{\relax}
\@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{1}{section.1}\protected@file@percent }
\newlabel{sec:Introduction}{{1}{1}{Introduction}{section.1}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces The particles of the SM.\relax }}{1}{figure.caption.2}\protected@file@percent }
\providecommand*\caption@xref[2]{\@setref\relax\@undefined{#1}}
\newlabel{fig:sm}{{1}{1}{The particles of the SM.\relax }{figure.caption.2}{}}
\citation{tensorflow2015-whitepaper}
\citation{tensorflow2015-whitepaper}
\citation{paszke2017automatic}
\citation{paszke2017automatic}
\citation{hep_survey_jim}
\citation{hep_survey_jim}
\@writefile{toc}{\contentsline {section}{\numberline {2}Model fitting}{5}{section.2}\protected@file@percent }
\newlabel{sec:modelfitting}{{2}{5}{Model fitting}{section.2}{}}
\newlabel{sec:theory}{{2}{5}{Model fitting}{section.2}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {2.1}Maximum Likelihood}{5}{subsection.2.1}\protected@file@percent }
\newlabel{eq:likelihood}{{1}{5}{Maximum Likelihood}{equation.2.1}{}}
\newlabel{eq:pdf}{{2}{5}{Maximum Likelihood}{equation.2.2}{}}
\newlabel{eq:pdf from func}{{3}{5}{Maximum Likelihood}{equation.2.3}{}}
\newlabel{eq:likelihood_from_products}{{4}{5}{Maximum Likelihood}{equation.2.4}{}}
\newlabel{eq:likelihood joint probability}{{2.1}{5}{Maximum Likelihood}{equation.2.4}{}}
\newlabel{eq:nll}{{5}{6}{Maximum Likelihood}{equation.2.5}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {2.2}Requirements}{6}{subsection.2.2}\protected@file@percent }
\newlabel{sec:requirements}{{2.2}{6}{Requirements}{subsection.2.2}{}}
\citation{software:scipy}
\citation{software:scipy}
\citation{software:lmfit}
\citation{software:lmfit}
\@writefile{toc}{\contentsline {subsection}{\numberline {2.3}Existing libraries}{7}{subsection.2.3}\protected@file@percent }
\newlabel{sec:landscape}{{2.3}{7}{Existing libraries}{subsection.2.3}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {2.3.1}General fitting}{7}{subsubsection.2.3.1}\protected@file@percent }
\citation{DBLP:journals/corr/abs-1711-10604}
\citation{DBLP:journals/corr/abs-1711-10604}
\citation{Verkerke:2003ir}
\citation{Verkerke:2003ir}
\citation{software:probfit}
\citation{software:probfit}
\citation{software:pyhf}
\citation{software:pyhf}
\citation{higgsanalysis_combinedlimit}
\citation{higgsanalysis_combinedlimit}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {2.3.2}HEP specific}{8}{subsubsection.2.3.2}\protected@file@percent }
\citation{tensorflow_analysis}
\citation{tensorflow_analysis}
\citation{James:1975dr}
\citation{James:1975dr}
\@writefile{toc}{\contentsline {section}{\numberline {3}\texttt  {zfit}{} introduction}{10}{section.3}\protected@file@percent }
\newlabel{sec:quickstart}{{3}{10}{\zfit {} introduction}{section.3}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Fitting workflow in \texttt  {zfit}{}. Model building is the largest part. Models combined with data can be used to create a loss. A minimiser finds the optimal values and returns them as a result. Estimations on the parameters uncertainties can then be made.\relax }}{12}{figure.caption.3}\protected@file@percent }
\newlabel{fig:fit_workflow}{{2}{12}{Fitting workflow in \zfit {}. Model building is the largest part. Models combined with data can be used to create a loss. A minimiser finds the optimal values and returns them as a result. Estimations on the parameters uncertainties can then be made.\relax }{figure.caption.3}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}TensorFlow backend}{12}{subsection.3.1}\protected@file@percent }
\newlabel{sec:implementation}{{3.1}{12}{TensorFlow backend}{subsection.3.1}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces A schematic view of a DNN function. Input takes the data and has the dimension of an event. Each end-point of a line is multiplied by a weight, a free parameter, and added with the other lines. At a node, a non-linear function is then applied.\relax }}{13}{figure.caption.4}\protected@file@percent }
\newlabel{fig:dnn}{{3}{13}{A schematic view of a DNN function. Input takes the data and has the dimension of an event. Each end-point of a line is multiplied by a weight, a free parameter, and added with the other lines. At a node, a non-linear function is then applied.\relax }{figure.caption.4}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Example of a graph representing $result = (5 + 3) * (7 + 2)$.\relax }}{15}{figure.caption.5}\protected@file@percent }
\newlabel{fig:graph_example}{{4}{15}{Example of a graph representing $result = (5 + 3) * (7 + 2)$.\relax }{figure.caption.5}{}}
\@writefile{toc}{\contentsline {section}{\numberline {4}\texttt  {zfit}{} implementation}{17}{section.4}\protected@file@percent }
\newlabel{sec:parts}{{4}{17}{\zfit {} implementation}{section.4}{}}
\newlabel{sec:concepts}{{4}{17}{\zfit {} implementation}{section.4}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {4.1}Spaces and Dimensions}{18}{subsection.4.1}\protected@file@percent }
\newlabel{sec:spaces and dimensions}{{4.1}{18}{Spaces and Dimensions}{subsection.4.1}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.1.1}Limits}{18}{subsubsection.4.1.1}\protected@file@percent }
\newlabel{sec:multiple limits}{{4.1.1}{18}{Limits}{subsubsection.4.1.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {4.2}Data handling}{19}{subsection.4.2}\protected@file@percent }
\newlabel{sec:data}{{4.2}{19}{Data handling}{subsection.4.2}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {4.3}Model}{20}{subsection.4.3}\protected@file@percent }
\newlabel{sec:model}{{4.3}{20}{Model}{subsection.4.3}{}}
\newlabel{eq:norm_range}{{6}{20}{Model}{equation.4.6}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.3.1}Parametrization}{21}{subsubsection.4.3.1}\protected@file@percent }
\newlabel{sec:parametrization}{{4.3.1}{21}{Parametrization}{subsubsection.4.3.1}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.3.2}Implementing a custom PDF}{22}{subsubsection.4.3.2}\protected@file@percent }
\newlabel{eq:simple_linear_model}{{7}{22}{Implementing a custom PDF}{equation.4.7}{}}
\newlabel{py:custom_pdf_linear_model}{{4.3.2}{23}{Implementing a custom PDF}{equation.4.7}{}}
\newlabel{py:instance_custom_pdf_linear_model}{{4.3.2}{23}{Implementing a custom PDF}{lstnumber.-16.10}{}}
\newlabel{eq:model integral}{{8}{24}{Implementing a custom PDF}{equation.4.8}{}}
\newlabel{integral_xy}{{10}{24}{Implementing a custom PDF}{equation.4.10}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.3.3}Sampling}{25}{subsubsection.4.3.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.3.4}Extended PDFs}{26}{subsubsection.4.3.4}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {4.4}Loss}{26}{subsection.4.4}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {4.5}Minimisation}{28}{subsection.4.5}\protected@file@percent }
\newlabel{sec:minimisation}{{4.5}{28}{Minimisation}{subsection.4.5}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.5.1}Different optimisations}{28}{subsubsection.4.5.1}\protected@file@percent }
\citation{kingma2014adam}
\citation{kingma2014adam}
\@writefile{toc}{\contentsline {subsection}{\numberline {4.6}Results and uncertainties}{29}{subsection.4.6}\protected@file@percent }
\@writefile{toc}{\contentsline {subsubsection}{\numberline {4.6.1}Parameter uncertainties}{30}{subsubsection.4.6.1}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {5}Performance}{31}{section.5}\protected@file@percent }
\newlabel{sec:performance}{{5}{31}{Performance}{section.5}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {5.1}Gaussian models}{31}{subsection.5.1}\protected@file@percent }
\newlabel{sec:perf gaussian models}{{5.1}{31}{Gaussian models}{subsection.5.1}{}}
\newlabel{fig:gperf_left}{{\caption@xref {fig:gperf_left}{ on input line 104}}{32}{Gaussian models}{figure.caption.6}{}}
\newlabel{sub@fig:gperf_left}{{}{32}{Gaussian models}{figure.caption.6}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {5}{\ignorespaces A sum of 9 Gaussian PDFs with shared (left) or individual (right) mu and sigma. On the y axes, the time for a single fit is shown, averaged over 20 fits. It is plotted against the number of events that have been drawn per toy.\relax }}{32}{figure.caption.6}\protected@file@percent }
\newlabel{fig:gperf}{{5}{32}{A sum of 9 Gaussian PDFs with shared (left) or individual (right) mu and sigma. On the y axes, the time for a single fit is shown, averaged over 20 fits. It is plotted against the number of events that have been drawn per toy.\relax }{figure.caption.6}{}}
\citation{Aaij:2015oid}
\citation{Aaij:2015oid}
\@writefile{lof}{\contentsline {figure}{\numberline {6}{\ignorespaces Measurement of the computation time with a sum of $n\_{gauss}$ Gaussians and in total two free parameters. Notice that the y-scale is the same for both plots but the x-axis for \texttt  {zfit}{} goes an order of magnitude higher. Also, \texttt  {zfit}{} sums up to 16 Gaussians whereas \unhbox \voidb@x \hbox {\textsc  {RooFit}}\xspace  only goes to 9.\relax }}{34}{figure.caption.7}\protected@file@percent }
\newlabel{fig:time events}{{6}{34}{Measurement of the computation time with a sum of $n\_{gauss}$ Gaussians and in total two free parameters. Notice that the y-scale is the same for both plots but the x-axis for \zfit {} goes an order of magnitude higher. Also, \zfit {} sums up to 16 Gaussians whereas \roofit only goes to 9.\relax }{figure.caption.7}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {7}{\ignorespaces Time of a single toy in dependence of the number of events used. Plotted for a sum of $n\_{gauss}$ Gaussians and two free parameters \textit  {per Gaussian}.\relax }}{34}{figure.caption.8}\protected@file@percent }
\newlabel{fig:time events nparam}{{7}{34}{Time of a single toy in dependence of the number of events used. Plotted for a sum of $n\_{gauss}$ Gaussians and two free parameters \textit {per Gaussian}.\relax }{figure.caption.8}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {5.2}Angular analysis}{34}{subsection.5.2}\protected@file@percent }
\newlabel{sec:angular analysis}{{5.2}{34}{Angular analysis}{subsection.5.2}{}}
\citation{Khachatryan:2038750}
\citation{Khachatryan:2038750}
\citation{Khachatryan:2038750}
\@writefile{lof}{\contentsline {figure}{\numberline {8}{\ignorespaces Schematic view of the angles of the $\unhbox \voidb@x \hbox {\ensuremath  {{\ensuremath  {{\ensuremath  {\ensuremath  {B}\xspace  }}\xspace  ^0}}\xspace  \tmspace  -\thinmuskip {.1667em}\ensuremath  {\rightarrow }\xspace  {\ensuremath  {{\ensuremath  {\ensuremath  {K}\xspace  }}\xspace  ^*}}\xspace  (\unhbox \voidb@x \hbox {\ensuremath  {\tmspace  -\thinmuskip {.1667em}\ensuremath  {\rightarrow }\xspace  {\ensuremath  {{\ensuremath  {\ensuremath  {K}\xspace  }}\xspace  ^+}}\xspace  {\ensuremath  {{\ensuremath  {\ensuremath  {\pi }\xspace  }}\xspace  ^-}}\xspace  }}\xspace  ){\ensuremath  {\ell }}\xspace  ^+{\ensuremath  {\ell }}\xspace  ^-}}\xspace  $ decay. The image was taken from\mciteCiteA  {\@auxout }{main}{\relax }{\relax }{\cite  }{Khachatryan:2038750}\relax }}{35}{figure.caption.9}\protected@file@percent }
\newlabel{fig:kst_angles}{{8}{35}{Schematic view of the angles of the $\decay {\Bz }{\Kstar (\decay {}{\Kp \pim })\lepton ^+\lepton ^-}$ decay. The image was taken from\cite {Khachatryan:2038750}\relax }{figure.caption.9}{}}
\newlabel{eq:kstangular}{{11}{35}{Angular analysis}{equation.5.11}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {9}{\ignorespaces The figures show the time needed per toy for the four dimensional P5' folded angular distribution. 25 toys are produced for each number of events.\relax }}{36}{figure.caption.10}\protected@file@percent }
\newlabel{fig:kst_angular_performance}{{9}{36}{The figures show the time needed per toy for the four dimensional P5' folded angular distribution. 25 toys are produced for each number of events.\relax }{figure.caption.10}{}}
\@writefile{toc}{\contentsline {section}{\numberline {6}Beyond standard fitting}{36}{section.6}\protected@file@percent }
\newlabel{sec:beyond standard fitting}{{6}{36}{Beyond standard fitting}{section.6}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {6.1}Amplitude fits}{37}{subsection.6.1}\protected@file@percent }
\newlabel{sec:dalitz}{{6.1}{37}{Amplitude fits}{subsection.6.1}{}}
\citation{James:1968gu}
\citation{James:1968gu}
\@writefile{toc}{\contentsline {subsection}{\numberline {6.2}phasespace}{38}{subsection.6.2}\protected@file@percent }
\newlabel{sec:phasespace}{{6.2}{38}{phasespace}{subsection.6.2}{}}
\newlabel{eq:momentum_conservation}{{12}{38}{phasespace}{equation.6.12}{}}
\citation{PhysRevLett.103.211801}
\citation{PhysRevLett.103.211801}
\newlabel{py:reso_particle}{{6.2}{39}{Usage}{section*.11}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {6.3}Dalitz implementation}{39}{subsection.6.3}\protected@file@percent }
\newlabel{eq:total amp}{{13}{40}{Dalitz implementation}{equation.6.13}{}}
\newlabel{fig:dalitz_comparison}{{\caption@xref {fig:dalitz_comparison}{ on input line 389}}{42}{Dalitz implementation}{figure.caption.12}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {10}{\ignorespaces Dalitz plot of the invariant mass ${\ensuremath  {{\ensuremath  {\ensuremath  {K}\xspace  }}\xspace  ^+}}\xspace  {\ensuremath  {{\ensuremath  {\ensuremath  {\pi }\xspace  }}\xspace  ^-}}\xspace  $ and ${\ensuremath  {{\ensuremath  {\ensuremath  {K}\xspace  }}\xspace  ^+}}\xspace  {\ensuremath  {{\ensuremath  {\ensuremath  {\pi }\xspace  }}\xspace  ^0}}\xspace  $. The left histogram is taken from the paper while the one on the right was sampled from a distribution built with \texttt  {zfit}{} and using \texttt  {phasespace}{} as sampler.\relax }}{42}{figure.caption.12}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {7}Conclusion and outlook}{44}{section.7}\protected@file@percent }
\newlabel{sec:conclusion}{{7}{44}{Conclusion and outlook}{section.7}{}}
\@writefile{toc}{\contentsline {section}{\numberline {A}Likelihood}{47}{appendix.A}\protected@file@percent }
\newlabel{appendix:likelihood}{{A}{47}{Likelihood}{appendix.A}{}}
\newlabel{eq:bayes theorem}{{14}{47}{Likelihood}{equation.A.14}{}}
\newlabel{eq:combined_prob}{{15}{47}{Likelihood}{equation.A.15}{}}
\newlabel{eq:combined_prob_rewritten}{{16}{47}{Likelihood}{equation.A.16}{}}
\newlabel{eq:loglikelihood_from_sums}{{23}{48}{Likelihood}{equation.A.23}{}}
\newlabel{eq:loglikelihood_from_sums_weighted}{{24}{48}{Likelihood}{equation.A.24}{}}
\newlabel{eq:simultaneous_likelihood}{{25}{49}{Likelihood}{equation.A.25}{}}
\newlabel{eq:extended_likelihood}{{26}{49}{Likelihood}{equation.A.26}{}}
\newlabel{eq:loglikelihood_from_sums_weighted_extended}{{28}{49}{Likelihood}{equation.A.28}{}}
\newlabel{eq:constraints}{{29}{49}{Likelihood}{equation.A.29}{}}
\citation{software:lauztat}
\citation{software:lauztat}
\newlabel{eq:total_likelihood}{{32}{50}{Likelihood}{equation.A.32}{}}
\@writefile{toc}{\contentsline {section}{\numberline {B}Backend}{50}{appendix.B}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {B.1}HPC and paradigms}{50}{subsection.B.1}\protected@file@percent }
\newlabel{appendix:paradigms}{{B.1}{50}{HPC and paradigms}{subsection.B.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {B.2}Working with TensorFlow}{52}{subsection.B.2}\protected@file@percent }
\newlabel{appendix:tensorflow}{{B.2}{52}{Working with TensorFlow}{subsection.B.2}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {B.2.1}Caching}{53}{subsubsection.B.2.1}\protected@file@percent }
\newlabel{appendix:caching}{{B.2.1}{53}{Caching}{subsubsection.B.2.1}{}}
\@writefile{toc}{\contentsline {section}{\numberline {C}Implementation}{54}{appendix.C}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {C.1}Spaces definition}{54}{subsection.C.1}\protected@file@percent }
\newlabel{appendix:spaces defined}{{C.1}{54}{Spaces definition}{subsection.C.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.2}General limits}{55}{subsection.C.2}\protected@file@percent }
\newlabel{appendix:general limits}{{C.2}{55}{General limits}{subsection.C.2}{}}
\citation{software:uproot}
\citation{software:uproot}
\citation{mckinney-proc-scipy-2010}
\citation{mckinney-proc-scipy-2010}
\citation{software:numpy}
\citation{software:numpy}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.3}Data formats}{56}{subsection.C.3}\protected@file@percent }
\newlabel{appendix:data formats}{{C.3}{56}{Data formats}{subsection.C.3}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.4}Data batching}{57}{subsection.C.4}\protected@file@percent }
\newlabel{appendix:data batching}{{C.4}{57}{Data batching}{subsection.C.4}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.5}Dependency management}{57}{subsection.C.5}\protected@file@percent }
\newlabel{appendix:dependency management}{{C.5}{57}{Dependency management}{subsection.C.5}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.6}Base Model}{58}{subsection.C.6}\protected@file@percent }
\newlabel{appendix:basemodel}{{C.6}{58}{Base Model}{subsection.C.6}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.1}Public methods}{58}{subsubsection.C.6.1}\protected@file@percent }
\newlabel{sec:public_methods}{{C.6.1}{58}{Public methods}{subsubsection.C.6.1}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.2}Hooks}{59}{subsubsection.C.6.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.3}Norm range handling}{59}{subsubsection.C.6.3}\protected@file@percent }
\newlabel{sec:norm_range_handling}{{C.6.3}{59}{Norm range handling}{subsubsection.C.6.3}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.4}Multiple limits handling}{60}{subsubsection.C.6.4}\protected@file@percent }
\newlabel{sec:multiple_limits_handling}{{C.6.4}{60}{Multiple limits handling}{subsubsection.C.6.4}{}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.5}Most efficient method}{60}{subsubsection.C.6.5}\protected@file@percent }
\@writefile{toc}{\contentsline {subsubsection}{\numberline {C.6.6}Functors}{61}{subsubsection.C.6.6}\protected@file@percent }
\newlabel{model:functors}{{C.6.6}{61}{Functors}{subsubsection.C.6.6}{}}
\newlabel{appendix:functor}{{C.6.6}{61}{Functors}{subsubsection.C.6.6}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.7}Sampling techniques}{61}{subsection.C.7}\protected@file@percent }
\newlabel{sec:sampling techniques}{{C.7}{61}{Sampling techniques}{subsection.C.7}{}}
\newlabel{appendix:sampling techniques}{{C.7}{61}{Sampling techniques}{subsection.C.7}{}}
\newlabel{fig:imporance_accept_reject}{{11b}{62}{\relax }{figure.caption.15}{}}
\newlabel{sub@fig:imporance_accept_reject}{{b}{62}{\relax }{figure.caption.15}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {11}{\ignorespaces Visualization of the accept reject method. Proposed events are randomly sampled in the valid range. In a) a uniformly sampled y value and in b) a Gaussian shaped y is used to either accept or reject them. The black like is the true shape of the model. The orange line represents the distribution the y were drawn from. Blue values are accepted, red are rejected.\relax }}{62}{figure.caption.15}\protected@file@percent }
\newlabel{fig:accept_reject}{{11}{62}{Visualization of the accept reject method. Proposed events are randomly sampled in the valid range. In a) a uniformly sampled y value and in b) a Gaussian shaped y is used to either accept or reject them. The black like is the true shape of the model. The orange line represents the distribution the y were drawn from. Blue values are accepted, red are rejected.\relax }{figure.caption.15}{}}
\@writefile{lof}{\contentsline {figure}{\numberline {12}{\ignorespaces Importance sampling with a wrong scaled weight. The sampled Gaussian distribution (blue) is cut on the top and does not resemble the correct shape.\relax }}{63}{figure.caption.16}\protected@file@percent }
\newlabel{fig:importance_biased}{{12}{63}{Importance sampling with a wrong scaled weight. The sampled Gaussian distribution (blue) is cut on the top and does not resemble the correct shape.\relax }{figure.caption.16}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {C.8}Loss defined}{63}{subsection.C.8}\protected@file@percent }
\newlabel{appendix:loss defined}{{C.8}{63}{Loss defined}{subsection.C.8}{}}
\@writefile{toc}{\contentsline {section}{\numberline {D}Performance studies}{63}{appendix.D}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {D.1}Hardware specification}{63}{subsection.D.1}\protected@file@percent }
\newlabel{appendix:hardware specs}{{D.1}{63}{Hardware specification}{subsection.D.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {D.2}Profiling TensorFlow}{64}{subsection.D.2}\protected@file@percent }
\newlabel{appendix:profiling tensorflow}{{D.2}{64}{Profiling TensorFlow}{subsection.D.2}{}}
\bibstyle{LHCb}
\bibdata{main,standard,LHCb-PAPER,LHCb-CONF,LHCb-DP,LHCb-TDR}
\bibcite{tensorflow2015-whitepaper}{1}
\bibcite{paszke2017automatic}{2}
\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces Execution time measurement of a loss-like function execution. The complexity of the problem is scaled by $n$ times adding the same loss again to the reduce function.\relax }}{65}{table.caption.17}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {D.3}Additional profiling}{65}{subsection.D.3}\protected@file@percent }
\newlabel{appendix:additional profiling}{{D.3}{65}{Additional profiling}{subsection.D.3}{}}
\@writefile{toc}{\contentsline {section}{References}{65}{figure.caption.18}\protected@file@percent }
\bibcite{hep_survey_jim}{3}
\bibcite{software:scipy}{4}
\bibcite{software:lmfit}{5}
\bibcite{DBLP:journals/corr/abs-1711-10604}{6}
\bibcite{Verkerke:2003ir}{7}
\bibcite{software:probfit}{8}
\bibcite{software:pyhf}{9}
\bibcite{higgsanalysis_combinedlimit}{10}
\@writefile{lof}{\contentsline {figure}{\numberline {13}{\ignorespaces Full toy study with sum of 9 Gaussians and 2 free parameters. We can see that \texttt  {zfit}{} s temporary bottleneck in sampling causes an extraordinary increase in execution time mostly for low number of events, but the conclusions and the overall scaling behaviour is still the same as described in Sec. \ref  {sec:perf gaussian models}.\relax }}{66}{figure.caption.18}\protected@file@percent }
\newlabel{fig:gperf across sampling}{{13}{66}{Full toy study with sum of 9 Gaussians and 2 free parameters. We can see that \zfit {} s temporary bottleneck in sampling causes an extraordinary increase in execution time mostly for low number of events, but the conclusions and the overall scaling behaviour is still the same as described in Sec. \ref {sec:perf gaussian models}.\relax }{figure.caption.18}{}}
\bibcite{tensorflow_analysis}{11}
\bibcite{James:1975dr}{12}
\bibcite{kingma2014adam}{13}
\bibcite{Aaij:2015oid}{14}
\bibcite{Khachatryan:2038750}{15}
\bibcite{James:1968gu}{16}
\bibcite{PhysRevLett.103.211801}{17}
\bibcite{software:lauztat}{18}
\bibcite{software:uproot}{19}
\bibcite{mckinney-proc-scipy-2010}{20}
\bibcite{software:numpy}{21}
\mciteSetMaxCount{main}{bibitem}{21}
\mciteSetMaxCount{main}{subitem}{1}
\mciteSetMaxWidth{main}{bibitem}{770040}
\mciteSetMaxWidth{main}{subitem}{0}
\zref@newlabel{LastPage}{\default{D.3}\page{67}\abspage{72}}