Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 228.882 LHC PhysicsExperimental Methods and MeasurementsData Analysis Strategies and Essentials[Lecture 6, February 23, 2009]C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials2Organizational Issues Project 1 – Charge Multiplicity•due date: March 12 (2.5 weeks) Project 2 – Bottomonia Cross Sections•due date: April 6 (3.5 weeks) Project 3 – B lifetime•due date: May 2 (3.5 weeks)C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials3Lecture Outline Data Analysis Strategies and Essential•motherhood and apple pie•proper work style•preparation of setup•manuals and tutorials•prototyping•design of an analysis•data processing•histograms and ntuples•rule of thumb for coding Most of the following seems obvious but so many people do it incorrectlyC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials4Motherhood and Apple Pie Performing analysis = repeating the same thing over and over with small variations. Everybody has to develop a work style for him/herself, and they are not all going to be the same. There are some general patterns though which make sense to consider. Most of this applies in one way or another beyond High Energy or Heavy Ion Physics.C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials5Proper Work Style Determine the objectives before you start•measure physics quantity with statistical and systematic uncertainties•document it in form of an internal note (including numbers and figures) Determine your tools•analysis tool is root: numbers and figures•documentation tool is latex•make sure to have a decent setup for your tools•you'll need it anyway•steal and copy as best as you can (no glory in this one)•you do not show your smarts here, though it has to look goodC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials6Proper Work Style Manuals and tutorials•reading a program manual is very ineffective, don't do it!•learn to use a program by example•best: use examples from people you know are good•second best: use tutorials which are relevant for what you do•in case of technical question•first try yourself, but not for too long (<15min)•then ask someone who you think might know the answer•or try to find an example (google is a great resource)•or check the manual (online if possible, search works best)C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials7Proper Work Style Prototyping•in most cases with an analysis in the beginning you do not know what is important and what not•prototyping the analysis will get you a quick feeling for what it is all about•prototyping means do it quick and dirty•after the prototype is done: cleanup and define the essential blocks (spent enough time here)•then prototype the blocks and refine as neededC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials8Design of an Analysis Data processing•analysis usually starts from bulky format•how much time does it takes to run over the entire sample?•how much time does a short peak job take?•decide whether you need a dedicated small format: ntuple (today: TTree) usually the answer is yes, sometimes it is enough to work with histograms Ntuple design•ntuple always has run and event number•design it to be fast, but as inclusive as you can•make sure no information is duplicated•split event information and run informationC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials9Design of an Analysis Larger ntuple projects (not for this class)•always organize ntuple on event basis•write clean atomic objects to be streamed•understand how to generically link objects•make sure to include versions to all objects•define split between input format and output format cleanly•consider large subdivision of code (avoid circular dependencies)•lowest level: objects•next higher: algorithms (uses objects)•highest: modules (uses objects and algorithms)C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials10Last Step in Analysis Histograms and fit•in all analysis you need to present figures or plots•plots are, apart from the numbers, the result of your analysis•good plots explain: “a picture says more then 1000 words”•they have to be presentable, invest time in them•time invested in good figures always pays•most analysis end up with a fit•fitting is an essential component of doing analysis•fitting is not difficult but needs quite some experience•always worth to invest some time into it•start with histogram fits in root and develop from there•TMinuit should be sufficient for almost any fit you'll doC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials11Example: Making a Plot – Style gray background, bold font, adjusted axis label position, larger font, adjusted axis division, adjusted line width, no useless information, no boxesC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials12Example: Making a Plot How to make a good plot, efficiently?•important figures are never made interactively•write a little macro (in CINT) for each plot•re-making the figure has to be very quick•be guaranteed: you will remake a plot many, many times• use standards•invest some work into your personal root style, copy someone style if it is good and you like it: MitStyle.C (SetStyle)•setups are often repeated, write some generic tools to help you make it go quickerC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials13Example: Making a Plot – Macro Load your preferred style•generally: choose no serif fonts for plots•this would be: Helvetica, Arial, etc.•choose all fonts to be completely consistent•choose default plot layout, maybe put together two three layouts•choose the setup ones and do not touch them afterwards•remember small tunings have to be applied to plots, one size never really fits all•use your default helpers to make the plot•tune the plots so they look niceC.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials14Example: Making a Plot – Macro cosine.C:C.Paus, 8.882 LHC Physics: Data Analysis Strategies and Essentials15Example: Ntuple for N Track Analysis Considerations•analyzing all data (241k, MC: 203k) takes ≈20 secs•reasonable test sample maybe