\documentclass[]{beamer}
\setbeamertemplate{navigation symbols}{}
\usepackage{beamerthemesplit}
\useoutertheme{infolines}
\usecolortheme{dolphin}
%\usetheme{Warsaw}
\usetheme{progressbar} 
\usecolortheme{progressbar}
\usefonttheme{progressbar}
\useoutertheme{progressbar}
\useinnertheme{progressbar}
\usepackage{graphicx}
%\usepackage{amssymb,amsmath}
\usepackage[latin1]{inputenc}
\usepackage{amsmath}
\newcommand\abs[1]{\left|#1\right|}
\usepackage{iwona}
\usepackage{hepparticles}
\usepackage{hepnicenames}
\usepackage{hepunits}
\progressbaroptions{imagename=images/lhcb}
%\usetheme{Boadilla}f

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
\definecolor{mygreen}{cmyk}{0.82,0.11,1,0.25}
\setbeamertemplate{blocks}[rounded][shadow=false]
\addtobeamertemplate{block begin}{\pgfsetfillopacity{0.8}}{\pgfsetfillopacity{1}}
\setbeamercolor{structure}{fg=mygreen}
\setbeamercolor*{block title example}{fg=mygreen!50,
bg= blue!10}
\setbeamercolor*{block body example}{fg= blue,
bg= blue!5}





%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\beamersetuncovermixins{\opaqueness<1>{25}}{\opaqueness<2->{15}}
\title{$\PB \to \PKstar \mu \mu$ selection}  
\author{\underline{Marcin Chrzaszcz}$^{1,2}$, Tatiana Likhomanenko$^{3,4}$,\\ Andrey Ustyuzhanin$^{3,4,5}$}
\date{\today} 

\begin{document}

{
\institute{$^1$ University of Zurich, $^2$ Institute of Nuclear Physics, $^3$ Yandex, $^4$ Kurchatov Institute, $^5$ Imperial College}
\setbeamertemplate{footline}{} 
\begin{frame}
\logo{
\vspace{2 mm}
\includegraphics[height=1cm,keepaspectratio]{images/uzh.jpg}~
\includegraphics[height=1cm,keepaspectratio]{images/ifj.png}}

  \titlepage
\end{frame}
}

\institute{UZH,IFJ} 




%normal slides
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{Reminder}
{~}
On last $\PKstar \mu \mu$ meeting we agreed on:
\begin{center}
\begin{itemize}
\item Use 10 Folds.
\item Use DLL instead of ProbNN.
\item Use isolation inside MVA.
\item Use DLL for $\mu$.
\end{itemize}
Remaining issues:
\begin{itemize}
\item Data agreement
\item Use new isolation or the old one
\item From me: TMVA vs MatrixNet


\end{itemize}
\end{center}
\end{frame}
\section[Outline]{}
\begin{frame}
\tableofcontents
\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Comparison of the performance of each sub-sample from the chopping technique}

\begin{frame}\frametitle{Fold comparison Background}
{~}
\begin{columns}
\column{1.3in}
\center BDT \\
\center \includegraphics[scale=.14]{images_v3/background_BDT.png}


\column{1.3in}
\center BDTG \\

\center \includegraphics[scale=.14]{images_v3/background_BDTG.png}


\column{1.3in}
\center MatrixNet \\

\center \includegraphics[scale=.14]{images_v3/background_MNoldISO.png}


\end{columns}
 
   \begin{exampleblock}{Conclusion}
Everything is very consistent.
   \end{exampleblock}



\end{frame}

\begin{frame}\frametitle{Fold comparison Background ZOOM}
{~}
\begin{columns}
\column{1.3in}
\center BDT \\
\center \includegraphics[scale=.14]{images_v3/background_BDTZOOM.png}


\column{1.3in}
\center BDTG \\

\center \includegraphics[scale=.14]{images_v3/background_BDTGZOOM.png}


\column{1.3in}
\center MatrixNet \\

\center \includegraphics[scale=.14]{images_v3/background_MNoldISOZOOM.png}


\end{columns}
   \begin{exampleblock}{Conclusion}
We only see statistical fluctuations within $2\sigma$. What would one expect with 10 folds.
   \end{exampleblock}


\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
\begin{frame}\frametitle{Fold comparison Sploted $\PKstar \mu \mu$}
{~}
\begin{columns}
\column{1.3in}
\center BDT \\
\center \includegraphics[scale=.14]{images_v3/splot_BDT.png}


\column{1.3in}
\center BDTG \\

\center \includegraphics[scale=.14]{images_v3/splot_BDTG.png}


\column{1.3in}
\center MatrixNet \\

\center \includegraphics[scale=.14]{images_v3/splot_MNOLDISO.png}


\end{columns}
   \begin{exampleblock}{Conclusion}
Again everything very consistent.
   \end{exampleblock}


\end{frame}

\begin{frame}\frametitle{Fold comparison Sploted $\PKstar \mu \mu$ ZOOM}
{~}
\begin{columns}
\column{1.3in}
\center BDT \\
\center \includegraphics[scale=.14]{images_v3/splot_BDTZOOM.png}


\column{1.3in}
\center BDTG \\

\center \includegraphics[scale=.14]{images_v3/splot_BDTGZOOM.png}


\column{1.3in}
\center MatrixNet \\

\center \includegraphics[scale=.14]{images_v3/splot_MNOLDISOZOOM.png}


\end{columns}
   \begin{exampleblock}{Conclusion}
Again everything very consistent with statistical fluctuations.
   \end{exampleblock}


\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\begin{frame}\frametitle{Acceptance}
   \begin{exampleblock}{~}
 All the classifiers show the same behavior: only the BDT is shown here (others in the backup)    \end{exampleblock}
\begin{columns}

\column{2.5in}
\center \includegraphics[scale=.11]{img_BDT/thetal.png}\\

\column{2.5in}
\center \includegraphics[scale=.11]{img_BDT/q2.png}

\end{columns}
%%%%%%%%%%
\begin{columns}

\column{1.3in}
\center \includegraphics[scale=.11]{img_BDT/thetak.png}

\column{1.3in}
\center \includegraphics[scale=.11]{img_BDT/q2.png}


\column{1.3in}
\center \includegraphics[scale=.11]{images_v3/Kpimass.png}


\end{columns}


\end{frame}





%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{Chopping stability I}
	I have tested each formula on MC following the procedure:
  \begin{itemize}
  \item Add 10 formulas to MC ntuples.
  \item For each formula make a cut.
  \item Calculate the efficiency.
  \end{itemize}
  

\end{frame}
\begin{frame}\frametitle{Chopping stability I, BDTG}
\begin{center}

	\begin{tabular}{ |c|c|c| }
  \hline
 Fold & Eff.$[\%]$ & Err.$[\%]$ \\ \hline
 0 & $86.86$ & $0.09$  \\ \hline
 1 & $87.04$ & $0.09$  \\ \hline
 2 & $87.02$ & $0.09$  \\ \hline
 3 & $86.88$ & $0.09$  \\ \hline
 4 & $86.94$ & $0.09$  \\ \hline
 5 & $87.02$ & $0.09$  \\ \hline
 6 & $87.10$ & $0.09$  \\ \hline
 7 & $86.99$ & $0.09$  \\ \hline
 8 & $87.14$ & $0.09$  \\ \hline
 9 & $87.12$ & $0.09$  \\ \hline

  \hline
\end{tabular}
   \begin{exampleblock}{Conclusion}
Everything consistent with statistical fluctuations!
   \end{exampleblock}

\end{center}
\end{frame}

\begin{frame}\frametitle{Chopping stability II, MatrixNet}
\begin{center}

	\begin{tabular}{ |c|c|c| }
  \hline
 Fold & Eff.$[\%]$ & Err.$[\%]$ \\ \hline
 0 & $89.56$ & $0.08$  \\ \hline
 1 & $89.59$ & $0.08$  \\ \hline
 2 & $89.59$ & $0.08$  \\ \hline
 3 & $89.60$ & $0.08$  \\ \hline
 4 & $89.56$ & $0.08$  \\ \hline
 5 & $89.56$ & $0.08$  \\ \hline
 6 & $89.60$ & $0.08$  \\ \hline
 7 & $89.63$ & $0.08$  \\ \hline
 8 & $89.65$ & $0.08$  \\ \hline
 9 & $89.55$ & $0.08$  \\ \hline

  \hline
\end{tabular}
   \begin{exampleblock}{Conclusion}
Everything consistent with statistical fluctuations!
   \end{exampleblock}



\end{center}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{MC Data comparison}
   \begin{exampleblock}{}
We performed MC/Data comparison using weights provided by Sam.
   \end{exampleblock}

\begin{columns}

\column{2.5in}
\center \includegraphics[scale=.14]{images_v3/iso.png}\\
\center \includegraphics[scale=.14]{images_v3/iso_NEW.png}

\column{2.5in}
\center \includegraphics[scale=.155]{images_v3/MN_datamc.png}\\
 \begin{exampleblock}{Conclusion}
\small New Isolation in slightly better agreement. \\
\small MatrixNet in similar agreement as BDT.
   \end{exampleblock}

\end{columns}



\end{frame}




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{Effect of new K-pi mass window on the chopping I}
  
	Due to enlarged K-pi mass window the gain of chopping is reduced (but remember, the chopping helps also in keeping the results more homogeneous)

\begin{columns}
\column{2.5in}
\center SMALL $\PKstar$ mass.
\center \includegraphics[scale=.23]{images_v3/OLD.png}
\column{2.5in}
\center LARGE $\PKstar$ mass.
\center \includegraphics[scale=.23]{images_v3/NEW.png}

\end{columns}

\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{Effect of new K-pi mass window on the chopping II}
\footnotesize We disagree with Sam on this issue.\\
Not easy to drive conclusion on ROC curve. Numerical results from 2 vs 10 Folds training.


  \begin{columns}
\column{2.5in}  
  \center 10 Folds, Large $\PKstar$ mass cut.
  \center \center \includegraphics[scale=.155]{images_v3/MN_0p15.png}

  
  \column{2.5in}  
  \center 2 Folds, Large $\PKstar$ mass cut.
  \center \center \includegraphics[scale=.155]{images_v3/MN_0p16.png}
  \end{columns}

   \begin{exampleblock}{Conclusion}
One gaisn $5\%$ background rejection.
   \end{exampleblock}



\end{frame}


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{TMVA vs MN performance}
\begin{frame}\frametitle{TMVA vs MN}
\footnotesize	Both classifiers show the same level of correlation in angles. Let's see their performance.  
  \begin{columns}
\column{2.5in}  
  \center MatrixNet\\
  \center \center \includegraphics[scale=.2]{images_v3/MN_0p15.png}

  
  \column{2.5in}  
  \center BDTG
  \center \center \includegraphics[scale=.2]{images_v3/BDTG_0p77.png}
  \end{columns}
   \begin{exampleblock}{Conclusion}
	Matrix Net we gain $2.4\%$ in signal and $6.9\%$ of bck rejection.
   \end{exampleblock}

\end{frame}


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}\frametitle{MatrixNet numbers}
\begin{center}

\begin{tabular}{ |c||c|c||c|c||c|c| }
  \hline
  $q^2$  &\multicolumn{2}{|c||}{MatrixNet} &\multicolumn{2}{|c||}{BDTG}  \\ \hline \hline
 $[GeV^2]$ & Signal & Bck & Signal & Bck  \\ \hline
   $0.1,2$ & $484 \pm 24$ & $54 \pm 7$ & $465 \pm 24$ & $58 \pm 7$  \\ \hline
  $2, 4.3$ & $291 \pm 21$ & $84 \pm 8$ & $270 \pm 20$ & $98 \pm 7$ \\ \hline 
  $4.3, 8.68$ & $823 \pm 34$ & $221 \pm 11$ & $807 \pm 34$ & $235 \pm 11$   \\ \hline 
  $10.09, 12.86$ & $660 \pm 28$ & $138 \pm 7$ & $658 \pm 28$ & $142 \pm 8$ \\ \hline 
  $14.18, 16$ & $481 \pm 24$ & $58 \pm 5$ & $467 \pm 24$ & $66 \pm 6$ \\ \hline 	
  $16, 19$ & $532 \pm 25$ & $60 \pm 7$ & $529 \pm 25$ & $61 \pm 7$\\ \hline 	
   $0.1, 19$ & $3252 \pm 65$ & $638 \pm 20$ & $3173 \pm 65$ & $685 \pm 20$ \\ \hline 	
\end{tabular}\\
\end{center}

%This is more then $5\sigma$ significant improvement in back of the envelope calculation.

\end{frame}








%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Conclusion}
\begin{frame}\frametitle{Conclusion}
	\begin{enumerate}
	\item Agreement in the MVA distribution for different sub-samples from the chopping: chopping helps in keeping results more homogeneous!
	\item Effectively we can use one classifier; simplification of the analysis.
	\item Slightly better performance of chopping in case of Larger $\PKstar$ mass window.
	\item New isolation is slightly better then the $\PBs\to \mu \mu$.
	\item MatrixNet performed slightly better then BDT.
	\end{enumerate}
{~}
   \begin{exampleblock}{From last time (in agreement with present Sam's studies)}
\begin{itemize}
\item ProbbNN performs better than DLL (from our studies $15\%$ less background)
\item NewIso slightly better than $\PBs \to \mu \mu$ (but with better Data/MC agreement).
\end{itemize}
   \end{exampleblock}


\end{frame}
%\section{TMVA vs MN performance}
\begin{frame}\frametitle{~}

\center \Huge BACKUP


\end{frame}

\begin{frame}\frametitle{}



\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%55

\begin{frame}\frametitle{Acceptance MatrixNet}
   \begin{exampleblock}{~}
MatrixNet
 \end{exampleblock}
\begin{columns}

\column{2.5in}
\center \includegraphics[scale=.11]{images_v3/thetal.png}\\
\center \includegraphics[scale=.11]{images_v3/thetak.png}

\column{2.5in}
\center \includegraphics[scale=.11]{images_v3/phi.png}\\
\center \includegraphics[scale=.11]{images_v3/q2.png}

\end{columns}


\end{frame}



\end{document}