PROOF lite
“Diving Into ROOT”
Abstact:
ROOT is a software framework for data analysis, a powerful tool to cope with the demanding tasks typical of state of the art scientific data analysis. Among its prominent features are an advanced graphical user interface, ideal for interactive analysis, an interpreter for the C++ programming language, for rapid and efficient prototyping and a persistency mechanism for C++ objects, used also to write every year petabytes of data recorded by the Large Hadron Collider experiments. This introductory guide illustrates the main features of ROOT, relevant for the typical problems of data analysis: input and plotting of data from measurements and fitting of analytical functions.
Welcome to data analysis!1
Comparison of measurements to theoretical models is one of the standard tasks in experimental physics. In the most simple case, a “model” is just a function providing predictions of measured data. Very often, the model depends on parameters. Such a model may simply state “the current I is proportional to the voltage U”, and the task of the experimentalist consists of determining the resistance, R, from a set of measurements.
As a first step, a visualisation of the data is needed. Next, some manipulations typically have to be applied, e.g. corrections or parameter transformations. Quite often, these manipulations are complex ones, and a powerful library of mathematical functions and procedures should be provided - think for example of an integral or peak-search or a Fourier transformation applied to an input spectrum to obtain the actual measurement described by the model.
One specialty of experimental physics are the inevitable errors affecting each measurement, and visualisation tools have to include these. In subsequent analysis, the statistical nature of the errors must be handled properly.
As the last step, measurements are compared to models, and free model parameters need to be determined in this process. See Figure 1.1 for an example of a function (model) fit to data points. Several standard methods are available, and a data analysis tool should provide easy access to more than one of them. Means to quantify the level of agreement between measurements and model must also be available.
Measured data points with error bars and fitted quadratic function.
Quite often, the data volume to be analyzed is large - think of fine-granular measurements accumulated with the aid of computers. A usable tool therefore must contain easy-to-use and efficient methods for data handling.
In Quantum mechanics, models typically only predict the probability density function (“pdf”) of measurements depending on a number of parameters, and the aim of the experimental analysis is to extract the parameters from the observed distribution of frequencies at which certain values of the measurement are observed. Measurements of this kind require means to generate and visualize frequency distributions, so-called histograms, and stringent statistical treatment to extract the model parameters from purely statistical distributions.
Simulation of expected data is another important aspect in data analysis. By repeated generation of “pseudo-data”, which are analysed in the same manner as intended for the real data, analysis procedures can be validated or compared. In many cases, the distribution of the measurement errors is not precisely known, and simulation offers the possibility to test the effects of different assumptions.
A powerful software framework addressing all of the above requirements is ROOT (Brun, René and Rademakers, Fons 1997), an open source project coordinated by the European Organisation for Nuclear Research, CERN in Geneva. ROOT is very flexible and provides both a programming interface to use in own applications and a graphical user interface for interactive data analysis. The purpose of this document is to serve as a beginners guide and provides extendable examples for your own use cases, based on typical problems addressed in student labs. This guide will hopefully lay the ground for more complex applications in your future scientific work building on a modern, state-of the art tool for data analysis.
This guide in form of a tutorial is intended to introduce you to the ROOT package in about 30 pages. This goal will be accomplished using concrete examples, according to the “learning by doing” principle. Also because of this reason, this guide cannot cover the complexity of the ROOT package. Nevertheless, once you feel confident with the concepts presented in the following chapters, you will be able to appreciate the ROOT Users Guide (The ROOT Team 2013a) and navigate through the Class Reference (The ROOT Team 2013b) to find all the details you might be interested in. You can even look at the code itself, since ROOT is a free, open-source product. Use these documents in parallel to this tutorial!
The ROOT Data Analysis Framework itself is written in and heavily relys on the programming language C++
, and therefore some knowledge about C
and C++
is required. Eventually, just profit from the immense available literature about C++
if you do not have any idea of what object oriented programming could be.
ROOT is available for many platforms (Linux, Mac OS X, Windows…), but in this guide we will implicitly assume that you are using Linux. The first thing you need to do with ROOT is install it, don’t you? Obtaining the latest ROOT version is straightforward. Just seek the “Pro” version on this webpage http://root.cern.ch/drupal/content/downloading-root. You will find precompiled versions for the different architectures, or the ROOT source code to compile yourself. Just pick up the flavour you need and follow the installation instructions.
Let’s dive into ROOT!
Now that you have installed ROOT, what’s this interactive shell thing you’re running ? It’s like this: ROOT leads a double life. It has an interpreter for macros (Cint (Goto 2005)) that you can run from the command line or run like applications. But it is also an interactive shell that can evaluate arbitrary statements and expressions. This is extremely useful for debugging, quick hacking and testing. Let us first have a look at some very simple examples.
You can even use the ROOT interactive shell in lieu of a calculator! Launch the ROOT interactive shell with the command
> root
on your Linux box. The prompt should appear shortly:
root [1]
and let’s dive in with the steps shown here:
root [0] 1+1
(const int)2
root [1] 2*(4+2)/12.
(const double)1.00000000000000000e+00
root [2] sqrt(3)
(const double)1.73205080756887719e+00
root [3] 1 > 2
(const int)0
root [4] TMath::Pi()
(Double_t)3.14159265358979312e+00
root [5] TMath::Erf(.2)
(Double_t)2.22702589210478447e-01
Not bad. You can see that ROOT offers you the possibility not only to type in C++
statements, but also advanced mathematical functions, which live in the TMath
namespace.
Now let’s do something more elaborated. A numerical example with the well known geometrical series:
root [6] double x=.5
root [7] int N=30
root [8] double geom_series=0
root [9] for (int i=0;i<N;++i)geom_series+=TMath::Power(x,i)
root [10] TMath::Abs(geom_series - (1-TMath::Power(x,N-1))/(1-x))
(Double_t)1.86264514923095703e-09
Here we made a step forward. We even declared variables and used a for control structure. Note that there are some subtle differences between Cint and the standard C++ language. You do not need the “;” at the end of line in interactive mode – try the difference e.g. using the command at line root [6]
.
Using one of ROOT’s powerful classes, here TF1
,2 will allow us to display a function of one variable, x. Try the following:
root [11] TF1 *f1 = new TF1("f1","sin(x)/x",0.,10.);
root [12] f1->Draw();
f1
is a pointer to an instance of a TF1 class, the arguments are used in the constructor; the first one of type string is a name to be entered in the internal ROOT memory management system, the second string type parameter defines the function, here sin(x)/x
, and the two parameters of type double define the range of the variable x. The Draw()
method, here without any parameters, displays the function in a window which should pop up after you typed the above two lines. Note again differences between Cint and C++: you could have omitted the “;” at the end of lines, Cint woud also have accepted the “.” to access the method Draw()
. However, it is best to stick to standard C++ syntax and avoid Cint-specific code, as will become clear in a moment.
A slightly extended version of this example is the definition of a function with parameters, called [0]
, [1]
and so on in the ROOT formula syntax. We now need a way to assign values to these parameters; this is achieved with the method SetParameter(<parameter_number>,<parameter_value>)
of class TF1
. Here is an example:
root [13] TF1 *f1 = new TF1("f2","[0]*sin([1]*x)/x",0.,10.);
root [14] f1->SetParameter(0,1);
root [15] f1->SetParameter(1,1);
root [16] f1->Draw();
Of course, this version shows the same results as the initial one. Try playing with the parameters and plot the function again. The class TF1
has a large number of very useful methods, including integration and differentiation. To make full use of this and other ROOT classes, visit the documentation on the Internet under http://root.cern.ch/drupal/content/reference-guide. Formulae in ROOT are evaluated using the class TFormula
, so also look up the relevant class documentation for examples, implemented functions and syntax.
You should definitely download this guide to your own system to have it at you disposal whenever you need it.
To extend a little bit on the above example, consider a more complex function you would like to define. You can also do this using standard C
or C++
code. In many cases this is the only practical way, as the ROOT formula interpreter has clear limitations concerning complexity and speed of evaluation.
Consider the example below, which calculates and displays the interference pattern produced by light falling on a multiple slit. Please do not type in the example below at the ROOT command line, there is a much simpler way: Make sure you have the file slits.C
on disk, and type root slits.C
in the shell. This will start root and make it read the “macro” slits.C
, i.e. all the lines in the file will be executed one after the other.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 |
|
Output of slits.C with parameters 0.2 and 2.
The example first asks for user input, namely the ratio of slit width over slit distance, and the number of slits. After entering this information, you should see the graphical output as is shown in Figure 2.1.
This is a more complicated example than the ones we have seen before, so spend some time analysing it carefully, you should have understood it before continuing. Let us go through it in detail:
Lines 6-19 define the necessary functions in C++
code, split into three separate functions, as suggested by the problem considered. The full interference pattern is given by the product of a function depending on the ratio of the width and distance of the slits, and a second one depending on the number of slits. More important for us here is the definition of the interface of these functions to make them usable for the ROOT class TF1
: the first argument is the pointer to x, the second one points to the array of parameters.
The main program starts in line 17 with the definition of a function slits()
of type void
. After asking for user input, a ROOT function is defined using the C-type function given in the beginning. We can now use all methods of the TF1
class to control the behaviour of our function – nice, isn’t it ?
If you like, you can easily extend the example to also plot the interference pattern of a single slit, using function double single
, or of a grid with narrow slits, function double nslit0
, in TF1
instances.
Here, we used a macro, some sort of lightweight program, that the interpreter distributed with ROOT, Cint, is able to execute. This is a rather extraordinary situation, since C++ is not natively an interpreted language! There is much more to say: chapter is indeed dedicated to macros.
One more remark at this point: as every command you type into ROOT is usually interpreted by Cint, an “escape character” is needed to pass commands to ROOT directly. This character is the dot at the beginning of a line:
root [1] .<command>
To
quit root, simply type .q
obtain a list of commands, use .?
access the shell of the operating system, type .!<OS_command>
; try, e.g. .!ls
or .!pwd
execute a macro, enter .x <file_name>
; in the above example, you might have used .x slits.C
at the ROOT prompt
load a macro, type .L <file_name>
; in the above example, you might instead have used the command .L slits.C
followed by the function call slits();
. Note that after loading a macro all functions and procedures defined therein are available at the ROOT prompt.
compile a macro, type .L <file_name>+
; ROOT is able to manage for you the C++ compiler behind the scenes and to produce machine code starting from your macro. One could decide to compile a macro in order to obtain better performance or to get nearer to the production environment.
To display measurements in ROOT, including errors, there exists a powerful class TGrapErrors
with different types of constructors. In the example here, we use data from the file ExampleData.txt
in text format:
root [0] TGraphErrors *gr=new TGraphErrors("ExampleData.txt");
root [1] gr->Draw("AP");
You should see the output shown in Figure 2.2.
Visualisation of data points with errors using the class TGraphErrors.
Make sure the file ExampleData.txt
is available in the directory from which you started ROOT. Inspect this file now with your favourite editor, or use the command less ExampleData.txt
to inspect the file, you will see that the format is very simple and easy to understand. Lines beginning with |
are ignored, very convenient to add some comments about the type of data. The data itself consist of lines with four real numbers each, representing the x- and y- coordinates and their errors of each data point.
The argument of the method Draw("AP")
is important here. It tells the TGraphPainter
class to show the axes and to plot markers at the x and y positions of the specified data points. Note that this simple example relies on the default settings of ROOT, concerning the size of the canvas holding the plot, the marker type and the line colours and thickness used and so on. In a well-written, complete example, all this would need to be specified explicitly in order to obtain nice and well readable results. A full chapter on graphs (chapter ) will explain many more of the features of the class TGraphErrors
and its relation to other ROOT classes in much more detail.
Frequency distributions in ROOT are handled by a set of classes derived from the histogram class TH1
, in our case TH1F
. The letter F
stands for “float”, meaning that the data type float
is used to store the entries in one histogram bin.
root [0] TF1 efunc("efunc","exp([0]+[1]*x)",0.,5.);
root [1] efunc.SetParameter(0,1);
root [2] efunc.SetParameter(1,-1);
root [3] TH1F* h=new TH1F("h","example histogram",100,0.,5.);
root [4] for (int i=0;i<1000;i++) {h->Fill(efunc.GetRandom());}
root [5] h->Draw();
The first three lines of this example define a function, an exponential in this case, and set its parameters. In Line 4 a histogram is instantiated, with a name, a title, a certain number of bins (100 of them, equidistant, equally sized) in the range from 0 to 5.
Visualisation of a histogram filled with exponentially distributed, random numbers.
We use yet another new feature of ROOT to fill this histogram with data, namely pseudo-random numbers generated with the method TF1::GetRandom
, which in turn uses an instance of the ROOT class TRandom
created when ROOT is started. Data is entered in the histogram at line 5 using the method TH1F::Fill
in a loop construct. As a result, the histogram is filled with 1000 random numbers distributed according to the defined function. The histogram is displayed using the method TH1F::Draw()
. You may think of this example as repeated measurements of the life time of a quantum mechanical state, which are entered into the histogram, thus giving a visual impression of the probability density distribution. The plot is shown in Figure 2.3.
Note that you will not obtain an identical plot when executing the above lines, depending on how the random number generator is initialised.
The class TH1F
does not contain a convenient input format from plain text files. The following lines of C++ code do the job. One number per line stored in the text file “expo.dat” is read in via an input stream and filled in the histogram until end of file is reached.
root [1] TH1F* h=new TH1F("h","example histogram",100,0.,5.);
root [2] ifstream inp; double x;
root [3] inp.open("expo.dat");
root [4] while (inp >> x) { h->Fill(x); }
root [5] h->Draw();
root [6] inp.close();
Histograms and random numbers are very important tools in statistical data analysis, and the whole Chapter will be dedicated to this topic.
Look at one of your plots again and move the mouse across. You will notice that this is much more than a static picture, as the mouse pointer changes its shape when touching objects on the plot. When the mouse is over an object, a right-click opens a pull-down menu displaying in the top line the name of the ROOT class you are dealing with, e.g. TCanvas
for the display window itself, TFrame
for the frame of the plot, TAxis
for the axes, TPaveText
for the plot name. Depending on which plot you are investigating, menus for the ROOT classes TF1
, TGraphErrors
or TH1F
will show up when a right-click is performed on the respective graphical representations. The menu items allow direct access to the members of the various classes, and you can even modify them, e.g. change colour and size of the axis ticks or labels, the function lines, marker types and so on. Try it!
Interactive ROOT panel for setting function parameters.
You will probably like the following: in the output produced by the example slits.C
, right-click on the function line and select “SetLineAttributes”, then left-click on “Set Parameters”. This gives access to a panel allowing you to interactively change the parameters of the function, as shown in Figure 2.4. Change the slit width, or go from one to two and then three or more slits, just as you like. When clicking on “Apply”, the function plot is updated to reflect the actual value of the parameters you have set.
Fit Panel.
Another very useful interactive tool is the FitPanel
, available for the classes TGraphErrors
and TH1F
. Predefined fit functions can be selected from a pull-down menu, including “gaus
”, “expo
” and “pol0
” - “pol9
” for Gaussian and exponential functions or polynomials of degree 0 to 9, respectively. In addition, user-defined functions using the same syntax as for functions with parameters are possible.
After setting the initial parameters, a fit of the selected function to the data of a graph or histogram can be performed and the result displayed on the plot. The fit panel is shown in Figure 2.5. The fit panel has a number of control options to select the fit method, fix or release individual parameters in the fit, to steer the level of output printed on the console, or to extract and display additional information like contour lines showing parameter correlations. As function fitting is of prime importance in any kind of data analysis, this topic will again show up in chapter .
If you are satisfied with your plot, you probably want to save it. Just close all selector boxes you opened previously and select the menu item Save as...
from the menu line of the window. It will pop up a file selector box to allow you to choose the format, file name and target directory to store the image. There is one very noticeable feature here: you can store a plot as a root macro! In this macro, you find the C++ representation of all methods and classes involved in generating the plot. This is a valuable source of information for your own macros, which you will hopefully write after having worked through this tutorial.
Using ROOT’s interactive capabilities is useful for a first exploration of possibilities. Other ROOT classes you will encounter in this tutorial have such graphical interfaces. We will not comment further on this, just be aware of the existence of ROOT’s interactive features and use them if you find convenient. Some trial-and-error is certainly necessary to find your way through the huge number of menus and parameter settings.
At this point of the guide, some basic questions could have already come to your mind. We will try to clarify some of them with further explanations in the following.
In the official ROOT documentation, you find special data types replacing the normal ones, e.g. Double_t
, Float_t
or Int_t
replacing the standard double
, float
or int
types. Using the ROOT types makes it easier to port code between platforms (64/32 bit) or operating systems (windows/Linux), as these types are mapped to suitable ones in the ROOT header files. If you want adaptive code of this type, use the ROOT type declarations. However, usually you do not need such adaptive code, and you can safely use the standard C type declarations for your private code, as we did and will do throughout this guide. If you intend to become a ROOT developer, however, you better stick to the official coding rules!
The behaviour of a ROOT session can be tailored with the options in the .rootrc
file. Examples of the tunable parameters are the ones related to the operating and window system, to the fonts to be used, to the location of start-up files. At start-up, ROOT looks for a .rootrc
file in the following order:
./.rootrc //local directory
$HOME/.rootrc //user directory
$ROOTSYS/etc/system.rootrc //global ROOT directory
If more than one .rootrc
files are found in the search paths above, the options are merged, with precedence local, user, global. The parsing and interpretation of this file is handled by the ROOT class TEnv
. Have a look to its documentation if you need such rather advanced features. The file .rootrc
defines the location of two rather important files inspected at start-up: rootalias.C
and rootlogon.C
. They can contain code that needs to be loaded and executed at ROOT startup. rootalias.C
is only loaded and best used to define some often used functions. rootlogon.C
contains code that will be executed at startup: this file is extremely useful for example to pre-load a custom style for the plots created with ROOT. This is done most easily by creating a new TStyle
object with your preferred settings, as described in the class reference guide, and then use the command gROOT->SetStyle("MyStyleName");
to make this new style definition the default one. As an example, have a look in the file rootlogon.C
coming with this tutorial. Another relevant file is rootlogoff.C
that it called when the session is finished.
Every command typed at the ROOT prompt is stored in a file .root_hist
in your home directory. ROOT uses this file to allow for navigation in the command history with the up-arrow and down-arrow keys. It is also convenient to extract successful ROOT commands with the help of a text editor for use in your own macros.
All global pointers in ROOT begin with a small “g”. Some of them were already implicitly introduced (for example in the section Configure ROOT at start-up). The most important among them are presented in the following:
gROOT: the gROOT
variable is the entry point to the ROOT system. Technically it is an instance of the TROOT
class. Using the gROOT
pointer one has access to basically every object created in a ROOT based program. The TROOT
object is essentially a container of several lists pointing to the main ROOT
objects.
gRandom: the gRandom variable is a variable that points to a random number generator instance of the type TRandom3
. Such a variable is useful to access in every point of a program the same random number generator, in order to achieve a good quality of the random sequence.
gStyle: By default ROOT creates a default style that can be accessed via the gStyle
pointer. This class includes functions to set some of the following object attributes.
Canvas
Pad
Histogram axis
Lines
Fill areas
Text
Markers
Functions
Histogram Statistics and Titles
gSystem: An instance of a base class defining a generic interface to the underlying Operating System, in our case TUnixSystem
.
At this point you have already learnt quite a bit about some basic features of ROOT.
Please move on to become an expert!
You know how other books go on and on about programming fundamentals and finally work up to building a complete, working program ? Let’s skip all that. In this guide, we will describe macros executed by the ROOT C++ interpreter CINT.
It is relatively easy to compile a macro, either as a pre-compiled library to load into ROOT, or as a stand-alone application, by adding some include statements for header file or some “dressing code” to any macro.
If you have a number of lines which you were able to execute at the ROOT prompt, they can be turned into a ROOT macro by giving them a name which corresponds to the file name without extension. The general structure for a macro stored in file MacroName.C
is
void MacroName() {
< ...
your lines of C++ code
... >
}
The macro is executed by typing
> root MacroName.C
at the system prompt, or it can be loaded into a ROOT session and then be executed by typing
root [0].L MacroName.C
root [1] MacroName();
at the ROOT prompt. Note that more than one macro can be loaded this way, as each macro has a unique name in the ROOT name space. A small set of options can help making your plot nicer.
gROOT->SetStyle("Plain"); // set plain TStyle
gStyle->SetOptStat(111111); // draw statistics on plots,
// (0) for no output
gStyle->SetOptFit(1111); // draw fit results on plot,
// (0) for no ouput
gStyle->SetPalette(53); // set color map
gStyle->SetOptTitle(0); // suppress title box
...
Next, you should create a canvas for graphical output, with size, subdivisions and format suitable to your needs, see documentation of class TCanvas
:
// create a canvas, specify position and size in pixels
TCanvas c1("c1","<Title>",0,0,400,300);
c1.Divide(2,2); //set subdivisions, called pads
c1.cd(1); //change to pad 1 of canvas c1
These parts of a well-written macro are pretty standard, and you should remember to include pieces of code like in the examples above to make sure your output always comes out as you had intended.
Below, in section Interpretation and Compilation, some more code fragments will be shown, allowing you to use the system compiler to compile macros for more efficient execution, or turn macros into stand-alone applications linked against the ROOT libraries.
Let us now look at a rather complete example of a typical task in data analysis, a macro that constructs a graph with errors, fits a (linear) model to it and saves it as an image. To run this macro, simply type in the shell:
> root macro1.C
The code is built around the ROOT class TGraphErrors
, which was already introduced previously. Have a look at it in the class reference guide, where you will also find further examples. The macro shown below uses additional classes, TF1
to define a function, TCanvas
to define size and properties of the window used for our plot, and TLegend
to add a nice legend. For the moment, ignore the commented include statements for header files, they will only become important at the end in section Interpretation and Compilation.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
|
Let’s comment it in detail:
Line 13: the name of the principal function (it plays the role of the “main” function in compiled programs) in the macro file. It has to be the same as the file name without extension.
Line 24-25: instance of the TGraphErrors
class. The constructor takes the number of points and the pointers to the arrays of $x$~values, $y$~values, $x$~errors (in this case none, represented by the NULL pointer) and $y$~errors. The second line defines in one shot the title of the graph and the titles of the two axes, separated by a “;”.
Line 28-30: the first line refers to the style of the plot, set as Plain. This is done through a manipulation of the global variable gSystem
(ROOT global variables begin always with “g”). The following three lines are rather intuitive right ? To understand better the enumerators for colours and styles see the reference for the TColor
and TMarker
classes.
Line 33: the canvas object that will host the drawn objects. The “memory leak” is intentional, to make the object existing also out of the macro1 scope.
Line 36: the method DrawClone draws a clone of the object on the canvas. It has to be a clone, to survive after the scope of macro1
, and be displayed on screen after the end of the macro execution. The string option “APE” stands for:
A imposes the drawing of the Axes.
P imposes the drawing of the graphs markers.
E imposes the drawing of the graphs error bars.
Line 39: define a mathematical function. There are several ways to accomplish this, but in this case the constructor accepts the name of the function, the formula, and the function range.
Line 41: maquillage. Try to give a look to the line styles at your disposal visiting the documentation of the TLine
class.
Line 43: fits the f function to the graph, observe that the pointer is passed. It is more interesting to look at the output on the screen to see the parameters values and other crucial information that we will learn to read at the end of this guide.
Line 44: again draws the clone of the object on the canvas. The “Same” option avoids the cancellation of the already drawn objects, in our case, the graph.
Line 47-52: completes the plot with a legend, represented by a TLegend
instance. The constructor takes as parameters the lower left and upper right corners coordinates with respect to the total size of the canvas, assumed to be 1, and the legend header string. You can add to the legend the objects, previously drawn or not drawn, through the addEntry
method. Observe how the legend is drawn at the end: looks familiar now, right ?
Line 55-57: defines an arrow with a triangle on the right hand side, a thickness of 2 and draws it.
Line 60-61: interpret a Latex string which hast its lower left corner located in the specified coordinate. The #splitline{}{}
construct allows to store multiple lines in the same TLatex
object.
Line 63: save the canvas as image. The format is automatically inferred from the file extension (it could have been eps, gif, …).
Let’s give a look to the obtained plot in Figure 3.1. Beautiful outcome for such a small bunch of lines, isn’t it ?
Your first plot with data points, a fit of an analytical function, a legend and some additional information in the form of graphics primitives and text. A well formatted plot, clear for the reader is crucial to communicate the relevance of your results to the reader.
We have seen that to specify a colour, some identifiers like kWhite, kRed or kBlue can be specified for markers, lines, arrows etc. The complete summary of colours is represented by the ROOT “colour wheel”. To know more about the full story, refer to the online documentation of TColor
.
ROOT provides an analogue of the colour wheel for the graphics markers. Select the most suited symbols for your plot among dots, triangles, crosses or stars. An alternative set of names for the markers is available.
The macro line 56 shows how to define an arrow and draw it. The class representing arrows is TArrow
, which inherits from TLine
. The constructors of lines and arrows always contain the coordinates of the endpoints. Arrows also foresee parameters to specify their shapes. Do not underestimate the role of lines and arrows in your plots. Since each plot should contain a message, it is convenient to stress it with additional graphics primitives.
Also text plays a fundamental role in making the plots self-explanatory. A possibility to add text in your plot is provided by the TLatex
class. The objects of this class are constructed with the coordinates of the bottom-left corner of the text and a string which contains the text itself. The real twist is that ordinary Latex mathematical symbols are automatically interpreted, you just need to replace the “\” by a “#”.
If “\” is used as control character , then the TMathText interface is invoked. It provides the plain TeX syntax and allow to access character’s set like Russian and Japenese.
As you observed, up to now we heavily exploited the capabilities of ROOT for interpreting our code, more than compiling and then executing. This is sufficient for a wide range of applications, but you might have already asked yourself “how can this code be compiled ?”. There are two answers.
ACLiC will create for you a compiled dynamic library for your macro, without any effort from your side, except the insertion of the appropriate header files in lines 3–9. In this example, they are already included. This does not harm, as they are not loaded by CINT. To generate an object libary from the macro code, from inside the interpreter type (please note the “+”):
root [1] .L macro1.C+
Once this operation is accomplished, the macro symbols will be available in memory and you will be able to execute it simply by calling from inside the interpreter:
root [2] macro1()
A plethora of excellent compilers are available, both free and commercial. We will refer to the GCC
compiler in the following. In this case, you have to include the appropriate headers in the code and then exploit the root-config tool for the automatic settings of all the compiler flags. root-config is a script that comes with ROOT; it prints all flags and libraries needed to compile code and link it with the ROOT libraries. In order to make the code executable stand-alone, an entry point for the operating system is needed, in C++ this is the procedure int main();
. The easiest way to turn a ROOT macro code into a stand-alone application is to add the following “dressing code” at the end of the macro file. This defines the procedure main, the only purpose of which is to call your macro:
#ifndef __CINT__
int main() {
ExampleMacro();
return 0;
}
#endif
Within ROOT, the variable __CINT__
is defined, and the code enclosed by #ifndef __CINT__
and #endif
is not executed; on the contrary, when running the system compiler GCC
, this symbol is not defined, and the code is compiled. To create a stand-alone program from a macro called ExampleMacro.C
, simply type
> g++ -o ExampleMacro.exe ExampleMacro.C \
`root-config --cflags --libs`
and execute it by typing
> ./ExampleMacro.exe
This procedure will, however, not give access to the ROOT graphics, as neither control of mouse or keyboard events nor access to the graphics windows of ROOT is available. If you want your stand-alone application have display graphics output and respond to mouse and keyboard, a slightly more complex piece of code can be used. In the example below, a macro ExampleMacro_GUI
is executed by the ROOT class TApplication. As a additional feature, this code example offers access to parameters eventually passed to the program when started from the command line. Here is the code fragment:
#ifndef __CINT__
void StandaloneApplication(int argc, char** argv) {
// eventually, evaluate the application parameters argc, argv
// ==>> here the ROOT macro is called
ExampleMacro_GUI();
}
// This is the standard "main" of C++ starting
// a ROOT application
int main(int argc, char** argv) {
TApplication app("ROOT Application", &argc, argv);
StandaloneApplication(app.Argc(), app.Argv());
app.Run();
return 0;
}
#endif
Compile the code with
> g++ -o ExampleMacro_GUI.exe \
ExampleMacro_GUI `root-config --cflags --libs`
and execute the program with
> ./ExampleMacro_GUI.exe
In this Chapter we will learn how to exploit some of the functionalities that ROOT provides to display data exploiting the class TGraphErrors
, which you already got to know previously.
The fastest way in which you can fill a graph with experimental data is to use the constructor which reads data points and their errors from a file in ASCII (i.e. standard text) format:
TGraphErrors(const char *filename,
const char *format="%lg %lg %lg %lg", Option_t *option="");
The format string can be:
"\%lg \%lg"
read only 2 first columns into X,Y
"\%lg \%lg \%lg"
read only 3 first columns into X,Y and EY
"\%lg \%lg \%lg \%lg"
read only 4 first columns into X,Y,EX,EY
This approach has the nice feature of allowing the user to reuse the macro for many different data sets. Here is an example of an input file. The nice graphic result shown is produced by the macro below, which reads two such input files and uses different options to display the data points.
# Measurement of Friday 26 March
# Experiment 2 Physics Lab
1 6 5
2 12 5
3 14 4.7
4 20 4.5
5 22 4.2
6 24 5.1
7 35 2.9
8 45 4.1
9 44 4.8
10 53 5.43
// Reads the points from a file and produces a simple graph.
int macro2(){
TCanvas* c=new TCanvas();
c->SetGrid();
TGraphErrors graph_expected("./macro2_input_expected.txt",
"%lg %lg %lg");
graph_expected.SetTitle(
"Measurement XYZ and Expectation;
lenght [cm];
Arb.Units");
graph_expected.SetFillColor(kYellow);
graph_expected.DrawClone("E3AL"); // E3 draws the band
TGraphErrors graph("./macro2_input.txt","%lg %lg %lg");
graph.SetMarkerStyle(kCircle);
graph.SetFillColor(0);
graph.DrawClone("PESame");
// Draw the Legend
TLegend leg(.1,.7,.3,.9,"Lab. Lesson 2");
leg.SetFillColor(0);
leg.AddEntry(&graph_expected,"Expected Points");
leg.AddEntry(&graph,"Measured Points");
leg.DrawClone("Same");
graph.Print();
}
In addition to the inspection of the plot, you can check the actual contents of the graph with the TGraph::Print()
method at any time, obtaining a printout of the coordinates of data points on screen. The macro also shows us how to print a coloured band around a graph instead of error bars, quite useful for example to represent the errors of a theoretical prediction.
With ROOT you can profit from rather advanced plotting routines, like the ones implemented in the TPolarGraph
, a class to draw graphs in polar coordinates. It is very easy to use, as you see in the example macro and the resulting Figure 4.1:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 |
|
A new element was added on line 4, the size of the canvas: it is sometimes optically better to show plots in specific canvas sizes.
The graph of a fan obtained with ROOT.
Under specific circumstances, it might be useful to plot some quantities versus two variables, therefore creating a bi-dimensional graph. Of course ROOT can help you in this task, with the TGraph2DErrors
class. The following macro produces a bi-dimensional graph representing a hypothetical measurement, fits a bi-dimensional function to it and draws it together with its x and y projections. Some points of the code will be explained in detail. This time, the graph is populated with data points using random numbers, introducing a new and very important ingredient, the ROOT TRandom3
random number generator using the Mersenne Twister algorithm (Matsumoto 1997).
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 |
|
Let’s go through the code, step by step to understand what is going on:
Line 3: This sets the palette colour code to a much nicer one than the default. Comment this line to give it a try.
Line 7: The instance of the random generator. You can then draw out of this instance random numbers distributed according to different probability density functions, like the Uniform one at lines 27-29. See the on-line documentation to appreciate the full power of this ROOT feature.
Line 12: You are already familiar with the TF1
class. This is its two-dimensional correspondent. At line 24 two random numbers distributed according to the TF2
formula are drawn with the method TF2::GetRandom2(double& a, double&b)
.
Line 27-29: Fitting a 2-dimensional function just works like in the one-dimensional case, i.e. initialisation of parameters and calling of the Fit()
method.
Line 32: The Surf1 option draws the TF2
objects (but also bi-dimensional histograms) as coloured surfaces with a wire-frame on three-dimensional canvases. See Figure 4.2.
Line 37-41: Here you learn how to create a canvas, partition it in two sub-pads and access them. It is very handy to show multiple plots in the same window or image.
A dataset fitted with a bidimensional function visualised as a colored surface.
Histograms play a fundamental role in any type of physics analysis, not only to visualise measurements but being a powerful form of data reduction. ROOT offers many classes that represent histograms, all inheriting from the TH1
class. We will focus in this chapter on uni- and bi- dimensional histograms whose bin-contents are represented by floating point numbers,3 the TH1F
and TH2F
classes respectively.
Let’s suppose you want to measure the counts of a Geiger detector put in proximity of a radioactive source in a given time interval. This would give you an idea of the activity of your source. The count distribution in this case is a Poisson distribution. Let’s see how operatively you can fill and draw a histogram with the following example macro.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 |
|
Which gives you the following plot (Figure 5.1):
The result of a counting (pseudo) experiment. Only bins corresponding to integer values are filled given the discrete nature of the poissonian distribution.
Using histograms is rather simple. The main differences with respect to graphs that emerge from the example are:
line 5: The histograms have a name and a title right from the start, no predefined number of entries but a number of bins and a lower-upper range.
line 15: An entry is stored in the histogram through the TH1F::Fill
method.
line 19 and 22: The histogram can be drawn also normalised, ROOT automatically takes cares of the necessary rescaling.
line 25 to 31: This small snippet shows how easy it is to access the moments and associated errors of a histogram.
Quite a large number of operations can be carried out with histograms. The most useful are addition and division. In the following macro we will learn how to manage these procedures within ROOT.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
|
The plots that you will obtain are shown in Figures 5.2 and 5.3.
The sum of two histograms.
The ratio of two histograms.
Some lines now need a bit of clarification:
line 3: Cint, as we know, is also able to interpret more than one function per file. In this case the function simply sets up some parameters to conveniently set the line of histograms.
line 20 to 22: Some contracted C++ syntax for conditional statements is used to fill the histograms with different numbers of entries inside the loop.
line 27: This is a crucial step for the sum and ratio of histograms to handle errors properly. The method TH1::Sumw2
makes sure that the squares of weights are stored inside the histogram (equivalent to the number of entries per bin if weights of 1 are used). This information is needed to correctly calculate the errors of each bin entry when the methods TH1::Add
and TH1::Divide
are invoked.
line 33: The sum of two histograms. A weight can be assigned to the added histogram, for example to comfortably switch to subtraction.
line 44: The division of two histograms is rather straightforward.
line 53 to 63: When you draw two quantities and their ratios, it is much better if all the information is condensed in one single plot. These lines provide a skeleton to perform this operation.
Two-dimensional histograms are a very useful tool, for example to inspect correlations between variables. You can exploit the bi-dimensional histogram classes provided by ROOT in a very simple way. Let’s see how in the following macro:
// Draw a Bidimensional Histogram in many ways
// together with its profiles and projections
void macro7(){
gStyle->SetPalette(53);
gStyle->SetOptStat(0);
gStyle->SetOptTitle(0);
TH2F bidi_h("bidi_h","2D Histo;Guassian Vals;Exp. Vals",
30,-5,5, // X axis
30,0,10); // Y axis
TRandom3 rgen;
for (int i=0;i<500000;i++)
bidi_h.Fill(rgen.Gaus(0,2),10-rgen.Exp(4),.1);
TCanvas* c=new TCanvas("Canvas","Canvas",800,800);
c->Divide(2,2);
c->cd(1);bidi_h.DrawClone("Cont1");
c->cd(2);bidi_h.DrawClone("Colz");
c->cd(3);bidi_h.DrawClone("lego2");
c->cd(4);bidi_h.DrawClone("surf3");
// Profiles and Projections
TCanvas* c2=new TCanvas("Canvas2","Canvas2",800,800);
c2->Divide(2,2);
c2->cd(1);bidi_h.ProjectionX()->DrawClone();
c2->cd(2);bidi_h.ProjectionY()->DrawClone();
c2->cd(3);bidi_h.ProfileX()->DrawClone();
c2->cd(4);bidi_h.ProfileY()->DrawClone();
}
Two kinds of plots are provided within the code, the first one containing three-dimensional representations (Figure 5.4) and the second one projections and profiles (Figure 5.5) of the bi-dimensional histogram.
Different ways of representing bi-dimensional histograms.
The projections and profiles of bi-dimensional histograms.
When a projection is performed along the x (y) direction, for every bin along the x (y) axis, all bin contents along the y (x) axis are summed up (upper the plots of Figure 5.5). When a profile is performed along the x (y) direction, for every bin along the x (y) axis, the average of all the bin contents along the y (x) is calculated together with their RMS and displayed as a symbol with error bar (lower two plots of Figure 5.5).
Correlations between the variables are quantified by the methods Double_t GetCovariance()
and Double_t GetCorrelationFactor()
.
After going through the previous chapters, you already know how to use analytical functions (class TF1
), and you got some insight into the graph (TGraphErrors
) and histogram classes (TH1F
) for data visualisation. In this chapter we will add more detail to the previous approximate explanations to face the fundamental topic of parameter estimation by fitting functions to data. For graphs and histograms, ROOT offers an easy-to-use interface to perform fits - either the fit panel of the graphical interface, or the Fit
method. The class TFitResult
allows access to the detailed results.
Very often it is necessary to study the statistical properties of analysis procedures. This is most easily achieved by applying the analysis to many sets of simulated data (or “pseudo data”), each representing one possible version of the true experiment. If the simulation only deals with the final distributions observed in data, and does not perform a full simulation of the underlying physics and the experimental apparatus, the name “Toy Monte Carlo” is frequently used.4 Since the true values of all parameters are known in the pseudo-data, the differences between the parameter estimates from the analysis procedure w.r.t. the true values can be determined, and it is also possible to check that the analysis procedure provides correct error estimates.
In the example below, a pseudo-data set is produced and a model fitted to it.
ROOT offers various minimisation algorithms to minimise a chi2 or a negative log-likelihood function. The default minimiser is MINUIT, a package originally implemented in the FORTRAN programming language. A C++ version is also available, MINUIT2, as well as Fumili (Silin 1983) an algorithm optimised for fitting. Genetic algorithms and a stochastic minimiser based on simulated annealing are also available. The minimisation algorithms can be selected using the static functions of the ROOT::Math::MinimizerOptions
class. Steering options for the minimiser, such as the convergence tolerance or the maximum number of function calls, can also be set using the methods of this class. All currently implemented minimisers are documented in the reference documentation of ROOT: have a look for example to the ROOT::Math::Minimizer
class documentation.
The complication level of the code below is intentionally a little higher than in the previous examples. The graphical output of the macro is shown in Figure 6.1:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 |
|
Some step by step explanation is at this point necessary:
Lines 1-3: A simple function to ease the make-up of lines. Remember that the class TF1
inherits from TAttLine
.
Lines 5-7 : Definition of a customised function, namely a Gaussian (the “signal”) plus a parabolic function, the “background”.
Lines 10-12: Some make-up for the Canvas. In particular we want that the parameters of the fit appear very clearly and nicely on the plot.
Lines 20-25: Define and initialise an instance of TF1
.
Lines 27-32: Define and fill a histogram.
Lines 34-40: For convenience, the same function as for the generation of the pseudo-data is used in the fit; hence, we need to reset the function parameters. This part of the code is very important for each fit procedure, as it sets the initial values of the fit.
Line 43: A very simple command, well known by now: fit the function to the histogram.
Lines 45-49: Retrieve the output from the fit. Here, we simply print the fit result and access and print the covariance matrix of the parameters.
Lines 58-end: Plot the pseudo-data, the fitted function and the signal and background components at the best-fit values.
Fit of pseudo data: a signal shape over a background trend. This plot is another example of how making a plot “self-explanatory” can help you better displaying your results.
Let us look at a simple example of a toy experiment comparing two methods to fit a function to a histogram, the \(\chi^{2}\)
method and a method called “binned log-likelihood fit”, both available in ROOT.
As a very simple yet powerful quantity to check the quality of the fit results, we construct for each pseudo-data set the so-called “pull”, the difference of the estimated and the true value of a parameter, normalised to the estimated error on the parameter, \(\frac{(p_{estim} - p_{true})}{\sigma_{p}}\). If everything is OK, the distribution of the pull values is a standard normal distribution, i.e. a Gaussian distribution centred around zero with a standard deviation of one.
The macro performs a rather big number of toy experiments, where a histogram is repeatedly filled with Gaussian distributed numbers, representing the pseudo-data in this example. Each time, a fit is performed according to the selected method, and the pull is calculated and filled into a histogram. Here is the code:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 |
|
Your present knowledge of ROOT should be enough to understand all the technicalities behind the macro. Note that the variable pull
in line 59 is different from the definition above: instead of the parameter error on mean
, the fitted standard deviation of the distribution divided by the square root of the number of entries, sig/sqrt(n_tot_entries)
, is used.
What method exhibits the better performance with the default parameters ?
What happens if you increase the number of entries per histogram by a factor of ten ? Why ?
The answers to these questions are well beyond the scope of this guide. Basically all books about statistical methods provide a complete treatment of the aforementioned topics.
ROOT offers the possibility to write the instances of all the classes inheriting from the class TObject
(basically all classes in ROOT) on disk, into what is referred to as ROOT-file, a file created by the TFile
class. One says that the object is made “persistent” by storing it on disk. When reading the file back, the object can be restored to memory.
We can explore this functionality with histograms and two simple macros.
void write_to_file(){
// Instance of our histogram
TH1F h("my_histogram","My Title;X;# of entries",100,-5,5);
// Let's fill it randomly
h.FillRandom("gaus");
// Let's open a TFile
TFile out_file("my_rootfile.root","RECREATE");
// Write the histogram in the file
h.Write();
// Close the file
out_file.Close();
}
Not bad, eh ? Especially for a language that does not foresees persistency natively like C++. The RECREATE option forces ROOT to create a new file even if a file with the same name exists on disk.
Now, you may use the Cint command line to access information in the file and draw the previously written histogram:
>>> root my_rootfile.root
root [0]
Attaching file my_rootfile.root as _file0...
root [1] _file0.ls()
TFile** my_rootfile.root
TFile* my_rootfile.root
KEY: TH1F my_histogram;1 My Title
root [2] my_histogram.Draw()
Alternatively, you can use a simple macro to carry out the job:
void read_from_file(){
// Let's open the TFile
TFile* in_file= new TFile("my_rootfile.root");
// Get the Histogram out
TH1F* h = (TH1F*) in_file.GetObjectChecked("my_histogram",
"TH1F");
// Draw it
h->Draw();
}
Please note that the order of opening files for write access and creating objects determines whether the objects are stored or not. You can avoid this behaviour by using the Write()
function as shown in the previous example.
Although you could be tempted to access an object within a file also with the Get
function and a C++ type cast, it is advisable to always use GetObjectChecked
.
Up to now we have seen how to manipulate input read from ASCII files. ROOT offers the possibility to do much better than that, with its own n-tuple classes. Among the many advantages provided by these classes one could cite
Optimised disk I/O.
Possibility to store many n-tuple rows.
Write the n-tuples in ROOT files.
Interactive inspection with TBrowser
.
Store not only numbers, but also objects in the columns.
In this section we will discuss briefly the TNtuple
class, which is a simplified version of the TTree
class. A ROOT TNtuple
object can store rows of float entries. Let’s tackle the problem according to the usual strategy commenting a minimal example
// Fill an n-tuple and write it to a file simulating measurement of
// conductivity of a material in different conditions of pressure
// and temperature.
void write_ntuple_to_file(){
// Initialise the TNtuple
TNtuple cond_data("cond_data",
"Example N-Tuple",
"Potential:Current:Temperature:Pressure");
// Fill it randomly to fake the acquired data
float pot,cur,temp,pres;
for (int i=0;i<10000;++i){
pot=gRandom->Uniform(0.,10.); // get voltage
temp=gRandom->Uniform(250.,350.); // get temperature
pres=gRandom->Uniform(0.5,1.5); // get pressure
cur=pot/(10.+0.05*(temp-300.)-0.2*(pres-1.)); // current
// add some random smearing (measurement errors)
pot*=gRandom->Gaus(1.,0.01); // 1% error on voltage
temp+=gRandom->Gaus(0.,0.3); // 0.3 abs. error on temp.
pres*=gRandom->Gaus(1.,0.02);// 1% error on pressure
cur*=gRandom->Gaus(1.,0.01); // 1% error on current
// write to ntuple
cond_data.Fill(pot,cur,temp,pres);
}
// Open a file, save the ntuple and close the file
TFile ofile("conductivity_experiment.root","RECREATE");
cond_data.Write();
ofile.Close();
}
This data written to this example n-tuple represents, in the statistical sense, three independent variables (Potential or Voltage, Pressure and Temperature), and one variable (Current) which depends on the others according to very simple laws, and an additional Gaussian smearing. This set of variables mimics a measurement of an electrical resistance while varying pressure and temperature.
Imagine your task now consists in finding the relations among the variables – of course without knowing the code used to generate them. You will see that the possibilities of the NTuple
class enable you to perform this analysis task. Open the ROOT file (cond_data.root
) written by the macro above in an interactive session and use a TBrowser
to interactively inspect it:
root[0] new TBrowser()
You find the columns of your n-tuple written as leafs. Simply clicking on them you can obtain histograms of the variables!
Next, try the following commands at the shell prompt and in the interactive ROOT shell, respectively:
> root conductivity_experiment.root
Attaching file conductivity_experiment.root as _file0...
root [0] cond_data.Draw("Current:Potential")
You just produced a correlation plot with one single line of code!
Try to extend the syntax typing for example
root [1] cond_data.Draw("Current:Potential","Temperature<270")
What do you obtain ?
Now try
root [2] cond_data.Draw("Current/Potential:Temperature")
It should have become clear from these examples how to navigate in such a multi-dimensional space of variables and unveil relations between variables using n-tuples.
For completeness, you find here a small macro to read the data back from a ROOT n-tuple
// Read the previously produced N-Tuple and print on screen
// its content
void read_ntuple_from_file(){
// Open a file, save the ntuple and close the file
TFile in_file("conductivity_experiment.root");
TNtuple* my_tuple = (TNtuple*) in_file.GetObjectChecked(
"cond_data",
"TNtuple");
float pot,cur,temp,pres; float* row_content;
cout << "Potential\tCurrent\tTemperature\tPressure\n";
for (int irow=0;irow<my_tuple->GetEntries();++irow){
my_tuple->GetEntry(irow);
row_content = my_tuple->GetArgs();
pot = row_content[0];
cur = row_content[1];
temp = row_content[2];
pres = row_content[3];
cout << pot << "\t" << cur << "\t" << temp
<< "\t" << pres << endl;
}
}
The macro shows the easiest way of accessing the content of a n-tuple: after loading the n-tuple, its branches are assigned to variables and GetEntry(long)
automatically fills them with the content for a specific row. By doing so, the logic for reading the n-tuple and the code to process it can be split and the source code remains clear.
It is also possible to write n-tuples of arbitrary type by using ROOT’s TBranch
class. This is especially important as TNtuple::Fill()
accepts only floats. The following macro creates the same n-tuple as before but the branches are booked directly. The Fill()
function then fills the current values of the connected variables to the tree.
// Fill an n-tuple and write it to a file simulating measurement of
// conductivity of a material in different conditions of pressure
// and temperature using branches.
void write_ntuple_to_file_advanced(
const std::string& outputFileName="conductivity_experiment.root"
,unsigned int numDataPoints=1000000){
// Initialise the TNtuple
TTree cond_data("cond_data", "Example N-Tuple");
// define the variables and book them for the ntuple
float pot,cur,temp,pres;
cond_data.Branch("Potential", &pot, "Potential/F");
cond_data.Branch("Current", &cur, "Current/F");
cond_data.Branch("Temperature", &temp, "Temperature/F");
cond_data.Branch("Pressure", &pres, "Pressure/F");
for (int i=0;i<numDataPoints;++i){
// Fill it randomly to fake the acquired data
pot=gRandom->Uniform(0.,10.)*gRandom->Gaus(1.,0.01);
temp=gRandom->Uniform(250.,350.)+gRandom->Gaus(0.,0.3);
pres=gRandom->Uniform(0.5,1.5)*gRandom->Gaus(1.,0.02);
cur=pot/(10.+0.05*(temp-300.)-0.2*(pres-1.))*
gRandom->Gaus(1.,0.01);
// write to ntuple
cond_data.Fill();}
// Open a file, save the ntuple and close the file
TFile ofile(outputFileName.c_str(),"RECREATE");
cond_data.Write();
ofile.Close();
}
The Branch()
function requires a pointer to a variable and a definition of the variable type. The following table lists some of the possible values. Please note that ROOT is not checking the input and mistakes are likely to result in serious problems. This holds especially if values are read as another type than they have been written, e.g. when storing a variable as float and reading it as double.
List of variable types that can be used to define the type of a branch in ROOT:
type | size | C++ | identifier |
---|---|---|---|
signed integer | 32 bit | int | I |
64 bit | long | L | |
unsigned integer | 32 bit | unsigned int | i |
64 bit | unsigned long | l | |
floating point | 32 bit | float | F |
64 bit | double | D | |
boolean | - | bool | O |
Usually n-tuples or trees span over many files and it would be difficult to add them manually. ROOT thus kindly provides a helper class in the form of TChain
. Its usage is shown in the following macro which is very similar to the previous example. The constructor of a TChain
takes the name of the TTree
(or TNuple
) as an argument. The files are added with the function Add(fileName)
, where one can also use wild-cards as shown in the example.
// Read several previously produced N-Tuples and print on screen its
// content.
//
// you can easily create some files with the following statement:
//
// for i in 0 1 2 3 4 5; \\
// do root -l -x -b -q \\
// "write_ntuple_to_file.cxx \\
// (\"conductivity_experiment_${i}.root\", 100)"; \\
// done
void read_ntuple_with_chain(){
// initiate a TChain with the name of the TTree to be processed
TChain in_chain("cond_data");
in_chain.Add("conductivity_experiment*.root"); // add files,
// wildcards work
// define variables and assign them to the corresponding branches
float pot, cur, temp, pres;
my_tuple->SetBranchAddress("Potential", &pot);
my_tuple->SetBranchAddress("Current", &cur);
my_tuple->SetBranchAddress("Temperature", &temp);
my_tuple->SetBranchAddress("Pressure", &pres);
cout << "Potential\tCurrent\tTemperature\tPressure\n";
for (size_t irow=0; irow<in_chain.GetEntries(); ++irow){
in_chain.GetEntry(irow); // loads all variables that have
// been connected to branches
cout << pot << "\t" << cur << "\t" << temp <<
"\t" << pres << endl;
}
}
Another very general and powerful way of processing a TChain
is provided via the method TChain::Process()
. This method takes as arguments an instance of a – user-implemented– class of type TSelector
, and – optionally – the number of entries and the first entry to be processed. A template for the class TSelector
is provided by the method TTree::MakeSelector
, as is shown in the little macro makeSelector.C
below.
It opens the n-tuple conductivity_experiment.root
from the example above and creates from it the header file MySelector.h
and a template to insert your own analysis code, MySelector.C
.
{
// create template class for Selector to run on a tree
//////////////////////////////////////////////////////
//
// open root file containing the Tree
TFile *f = TFile::Open("conductivity_experiment.root");
// create TTree object from it
TTree *t = (TTree *) f->Get("cond_data");
// this generates the files MySelector.h and MySelector.C
t->MakeSelector("MySelector");
}
The template contains the entry points Begin()
and SlaveBegin()
called before processing of the TChain
starts, Process()
called for every entry of the chain, and SlaveTerminate()
and Terminate()
called after the last entry has been processed. Typically, initialization like booking of histograms is performed in SlaveBegin()
, the analysis, i.e. the selection of entries, calculations and filling of histograms, is done in Process()
, and final operations like plotting and storing of results happen in SlaveTerminate()
or Terminate()
.
The entry points SlaveBegin()
and SlaveTerminate()
are called on so-called slave nodes only if parallel processing via PROOF
or PROOF lite
is enabled, as will be explained below.
A simple example of a selector class is shown in the macro MySelector.C
. The example is executed with the following sequence of commands:
> TChain *ch=new TChain("cond_data", "Chain for Example N-Tuple");
> ch->Add("conductivity_experiment*.root");
> ch->Process("MySelector.C+");
As usual, the “+
” appended to the name of the macro to be executed initiates the compilation of the MySelector.C
with the system compiler in order to improve performance.
The code in MySelector.C
, shown in the listing below, books some histograms in SlaveBegin()
and adds them to the instance fOutput
, which is of the class TList
.5 The final processing in Terminate()
allows to access histograms and store, display or save them as pictures. This is shown in the example via the TList
fOutput
. See the commented listing below for more details; most of the text is actually comments generated automatically by TTree::MakeSelector
.
#define MySelector_cxx
// The class definition in MySelector.h has been generated
// automatically by the ROOT utility TTree::MakeSelector().
// This class is derived from the ROOT class TSelector. For
// more information on the TSelectorframework see
// $ROOTSYS/README/README.SELECTOR or the ROOT User Manual.
// The following methods are defined in this file:
// Begin(): called every time a loop on the tree starts,
// a convenient place to create your histograms.
// SlaveBegin(): called after Begin(), when on PROOF called
// only on the slave servers.
// Process(): called for each event, in this function you
// decide what to read and fill your histograms.
// SlaveTerminate: called at the end of the loop on the tree,
// when on PROOF called only on the slave
// servers.
// Terminate(): called at the end of the loop on the tree, a
// convenient place to draw/fit your histograms.
//
// To use this file, try the following session on your Tree T:
//
// root> T->Process("MySelector.C")
// root> T->Process("MySelector.C","some options")
// root> T->Process("MySelector.C+")
//
#include "MySelector.h"
#include <TH2.h>
#include <TStyle.h>
#include <TCanvas.h>
// user defined variables may come here:
UInt_t fNumberOfEvents; TDatime tBegin, tNow;
TH1F *h_pot,*h_cur,*h_temp,*h_pres,*h_resistance;
void MySelector::Begin(TTree * /*tree*/)
{
// The Begin() function is called at the start of the query.
// When running with PROOF Begin() is only called on the client.
// The tree argument is deprecated (on PROOF 0 is passed).
TString option = GetOption();
// some time measurement
tBegin.Set(); printf("*==* ---------- Begin of Job ----------");
tBegin.Print();
}
void MySelector::SlaveBegin(TTree * /*tree*/)
{
// The SlaveBegin() function is called after the Begin()
// function. When running with PROOF SlaveBegin() is called on
// each slave server. The tree argument is deprecated
// (on PROOF 0 is passed).
TString option = GetOption();
//book some histograms
h_pot=new TH1F("pot","potential",100,-0.5,10.5);
h_cur=new TH1F("cur","current",100,-0.1,1.5);
h_temp=new TH1F("temp","temperature",100,200.,400.);
h_pres=new TH1F("pres","pressure",100,-0.,2.);
h_resistance=new TH1F("resistance","resistance",100,5.,15.);
// add all booked histograms to output list
// (only really needed for PROOF)
fOutput->AddAll(gDirectory->GetList());
}
Bool_t MySelector::Process(Long64_t entry)
{
// The Process() function is called for each entry in the tree
// (or possibly keyed object in the case of PROOF) to be
// processed. The entry argument specifies which entry in the
// currently loaded tree is to be processed. It can be passed to
// either MySelector::GetEntry() or TBranch::GetEntry()
// to read either all or the required parts of the data. When
// processing // keyed objects with PROOF, the object is already
// loaded and is available via the fObject pointer.
//
// This function should contain the "body" of the analysis. It
// can contain simple or elaborate selection criteria, run
// algorithms on the data // of the event and typically fill
// histograms.
//
// The processing can be stopped by calling Abort().
//
// Use fStatus to set the return value of TTree::Process().
//
// The return value is currently not used.
// - - - - - - - - - begin processing
GetEntry(entry);
// count number of entries (=events) ...
++fNumberOfEvents;
// analsiys code comes here - fill histograms
h_pot->Fill(Potential);
h_cur->Fill(Current);
h_temp->Fill(Temperature);
h_pres->Fill(Pressure);
h_resistance->Fill(Potential/Current);
return kTRUE; //kFALSE would abort processing
}
void MySelector::SlaveTerminate()
{
// The SlaveTerminate() function is called after all entries or
// objects have been processed. When running with PROOF
// SlaveTerminate() is called on each slave server.
// some statistics at end of job
printf("\n *==* ---------- End of Slave Job ---------- ");
tNow.Set(); tNow.Print();
printf(
"Number of Events: %i, elapsed time: %i sec, rate: %g evts/sec\n"
,fNumberOfEvents,
tNow.Convert()-tBegin.Convert(),
float(fNumberOfEvents)/(tNow.Convert()-tBegin.Convert()) );
}
void MySelector::Terminate()
{
// The Terminate() function is the last function to be called
// during a query. It always runs on the client, it can be used
// to present the results graphically or save the results to
// file.
// finally, store all output
TFile hfile("MySelector_Result.root","RECREATE","MuonResults");
fOutput->Write();
//Example to retrieve output from output list
h_resistance=
dynamic_cast<TH1F *>(fOutput->FindObject("resistance"));
TCanvas c_result("cresult","Resistance",100,100,300,300);
h_resistance->Draw();
c_result.SaveAs("ResistanceDistribution.png");
tNow.Set(); printf("*==* ---------- End of Job ---------- ");
tNow.Print();
}
PROOF lite
The processing of n-tuples via a selector function of type TSelector
through TChain::Process()
, as described at the end of the previous section, offers an additional advantage in particular for very large data sets: on distributed systems or multi-core architectures, portions of data can be processed in parallel, thus significantly reducing the execution time. On modern computers with multi-core CPUs or hyper-threading enabled, this allows a much faster turnaround of analyses, since all the available CPU power is used.
On distributed systems, a PROOF server and worker nodes have to be set up, as described in detail in the ROOT documentation. On a single computer with multiple cores, PROOF lite
can be used instead. Try the following little macro, RunMySelector.C
, which contains two extra lines compared to the example above (adjust the number of workers according to the number of CPU cores):
{// set up a TChain
TChain *ch=new TChain("cond_data", "My Chain for Example N-Tuple");
ch->Add("conductivity_experiment*.root");
// eventually, start Proof Lite on cores
TProof::Open("workers=4");
ch->SetProof();
ch->Process("MySelector.C+");}
The first command, TProof::Open(const char*)
starts a local PROOF server (if no arguments are specified, all cores will be used), and the command ch->SetProof();
enables processing of the chain using PROOF. Now, when issuing the command ch->Process("MySelector.C+);
, the code in MySelector.C
is compiled and executed on each slave node. The methods Begin()
and Terminate()
are executed on the master only. The list of n-tuple files is analysed, and portions of the data are assigned to the available slave processes. Histograms booked in SlaveBegin()
exist in the processes on the slave nodes, and are filled accordingly. Upon termination, the PROOF master collects the histograms from the slaves and merges them. In Terminate()
all merged histograms are available and can be inspected, analysed or stored. The histograms are handled via the instances fOutput
of class TList
in each slave process, and can be retrieved from this list after merging in Terminate
.
To explore the power of this mechanism, generate some very large n-tuples using the script from the section Storing Arbitrary N-tuples - you could try 10 000 000 events (this results in a large n-tuple of about 160 MByte in size). You could also generate a large number of files and use wildcards to add the to the TChain
. Now execute: > root -l RunMySelector.C
and watch what happens:
Processing RunMySelector.C...
+++ Starting PROOF-Lite with 4 workers +++
Opening connections to workers: OK (4 workers)
Setting up worker servers: OK (4 workers)
PROOF set to parallel mode (4 workers)
Info in <TProofLite::SetQueryRunning>: starting query: 1
Info in <TProofQueryResult::SetRunning>: nwrks: 4
Info in <TUnixSystem::ACLiC>: creating shared library
~/DivingROOT/macros/MySelector_C.so
*==* ----- Begin of Job ----- Date/Time = Wed Feb 15 23:00:04 2012
Looking up for exact location of files: OK (4 files)
Looking up for exact location of files: OK (4 files)
Info in <TPacketizerAdaptive::TPacketizerAdaptive>:
Setting max number of workers per node to 4
Validating files: OK (4 files)
Info in <TPacketizerAdaptive::InitStats>:
fraction of remote files 1.000000
Info in <TCanvas::Print>:
file ResistanceDistribution.png has been created
*==* ----- End of Job ----- Date/Time = Wed Feb 15 23:00:08 2012
Lite-0: all output objects have been merged
Log files of the whole processing chain are kept in the directory ~.proof
for each worker node. This is very helpful for debugging or if something goes wrong. As the method described here also works without using PROOF, the development work on an analysis script can be done in the standard way on a small subset of the data, and only for the full processing one would use parallelism via PROOF.
It is worth to remind the reader that the speed of typical data analysis programs limited by the I/O speed (for example the latencies implied by reading data from a hard drive). It is therefore expected that this limitation cannot be eliminated with the usage of any parallel analysis toolkit.
ROOT automatically applies compression algorithms on n-tuples to reduce the memory consumption. A value that is in most cases only zero will consume only small space on your disk (but it has to be deflated on reading). Nevertheless, you should think about the design of your n-tuples and your analyses as soon as the processing time exceeds some minutes.
Try to keep your n-tuples simple and use appropriate variable types. If your measurement has only a limited precision, it is needless to store it with double precision.
Experimental conditions that do not change with every single measurement should be stored in a separate tree. Although the compression can handle redundant values, the processing time increase with every variable that has to be filled.
The function SetCacheSize(long)
specifies the size of the cache for reading a TTree
object from a file. The default value is 30MB. A manual increase may help in certain situations. Please note that the caching mechanism can cover only one TTree
object per TFile
object.
You can select the branches to be covered by the caching algorithm with AddBranchToCache
and deactivate unneeded branches with SetBranchStatus
. This mechanism can result in a significant speed-up for simple operations on trees with many branches.
You can measure the performance easily with TTreePerfStats
. The ROOT documentation on this class also includes an introductory example. For example, TTreePerfStats
can show you that it is beneficial to store meta data and payload data separately, i.e. write the meta data tree in a bulk to a file at the end of your job instead of writing both trees interleaved.
This is the end of our guided tour for beginners through ROOT. There is still a lot coming to mind to be said, but by now you are experienced enough to use the ROOT documentation, most importantly the ROOT home page and the ROOT reference guide with the documentation of all ROOT classes, or the ROOT users guide.
A very useful way for you to continue exploring ROOT is to study the examples in the sub-directory tutorials/
of any ROOT installation.
There are some powerful features of ROOT which were not treated in this document, e.g. packages named RooFit and RooStats providing an advanced framework for model building, fitting and statistical analysis. The ROOT class TMVA
offers multi-variate analysis tools including an artificial neural network and many other advanced tools for classification problems. The remarkable ability of ROOT to handle large data volumes was already mentioned in this guide, implemented through the class TTree
. But there is still much more for you to explore!
End of this guide … but hopefully not of your interaction with ROOT !
Brun, René and Rademakers, Fons. 1997. “ROOT - An Object Oriented Data Analysis Framework.” Nucl. Inst. and Meth. in Phys. Res. A 389: 81–86.
Goto, Masaharu. 2005. “Cint Users Guide.” http://root.cern.ch/drupal/content/cint.
Matsumoto, Makoto. 1997. “Mersenne Twister Home Page.” http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html.
Silin, I. N. 1983. “FUMILI.” CERN Program Library D510.
The ROOT Team. 2013a. The ROOT Users Guide. http://root.cern.ch/drupal/content/users-guide.
———. 2013b. The ROOT Reference Guide. http://root.cern.ch/drupal/content/reference-guide.
This guide was prepared for the ROOT IRMM Tutorial adapting “A ROOT Guide for Students” http://www-ekp.physik.uni-karlsruhe.de/~quast, a document by D. Piparo, G. Quast and M. Zeise.↩
All ROOT classes start with the letter T.↩
To optimise the memory usage you might go for one byte (TH1C), short (TH1S), integer (TH1I) or double-precision (TH1D) bin-content.↩
“Monte Carlo” simulation means that random numbers play a role here which is as crucial as in games of pure chance in the Casino of Monte Carlo.↩
The usage of fOutput
is not really needed for this simple example, but it allows re-usage of the exact code in parallel processing with PROOF
(see next section).↩