Python Interface

Batch Generation
Simplified Interface
Installation
Examples
Interface Generation

Python is a powerful, high-level interpreted language that is widely used within the particle physics community. It is particularly useful as an interactive language that provides fast proto-typing of code. A Python interface to PYTHIA is available. This interface is automatically generated with Binder using the PyBind11 template library. Please note that this interface has not been extensively tested and consequently issues may arise. If bugs are found or additional features are required, as usual, please report them to the PYTHIA authors. This interface is a significant departure from the previous interface generated with SWIG, which does not support C++11.

A complete interface ships with PYTHIA. However, in some cases this interface might be more than needed by users and so a simplified lightweight interface can also be generated. Note, not all PYTHIA classes are available through this simplified interface. It is also possible for users to automatically generate their own interface. This is particularly useful if users modify the PYTHIA source code and need to update the interface. When generating the interface there are three options: the simplified interface, a full interface, and a user defined interface. The process of generating these interfaces is described in more detail below.

There are some idiosyncrasies when working with standard template library classes. One of the notable examples are maps with keys as strings. In many cases, it is useful to be able to access the settings maps from the Settings class of a Pythia instance. Rather than calling something like getFlagMap(), getFlagMap("") must be called so the interface knows how to map the keys. A full working example is given below.

 
import pythia8 
pythia = pythia8.Pythia("", False) 
settings = pythia.settings 
flags = settings.getFlagMap("") 
for key, val in settings.getFlagMap("").items(): print(key)

The following points highlight some of the features of the interface.

Documentation through the built-in help function in Python provides the relevant mappings between the Python interface and the C++ code. For documentation on the purpose of the code, see the user HTML manual and Doxygen.
Templated C++ types are returned as native Python types, e.g. vectors are returned as lists, etc. Similarly, native Python types can be passed as arguments and are converted to their equivalent C++ templates.
No difference is made between passing by reference or pointer; methods in C++ which take a pointer as an argument in Python simply take the object, e.g. foo(Vec4*) and foo(Vec4&) are the same in the Python interface.
All operators defined in C++, e.g. Vec4*double are available. Note that reverse operators are not, e.g. double*Vec4.
Classes with defined [] operators are iterable, using standard Python iteration, e.g. for prt in pythia.event.
Classes with a << operator can be printed via the built-in print function in Python. Note this means that a string representation via str is also available for these classes in Python.
Derived classes in Python can be passed back to PYTHIA, and should be available for all classes. See main293.py for a direct Python translation of the C++ main242.cc example which uses a derived class from the UserHooks class to veto events.
Events can be generated in parallel using PythiaParallel and providing an analysis function written in Python. This is demonstrated in main292.py.
Events can be generated in batch using Awkward arrays, see below for more details as well as main247.py and main248.py.
The constant Pythia::info member is available as Pythia::infoPython. Note, this creates a new instance of the Info class, and so needs to be called each time the information might be updated.

In general, most code using PYTHIA implemented through C++ is also possible in the Python interface. There are a number of issues, such as passing streams, which cannot be handled in the Python interface. Additionally, protected members of classes are exposed as fully public members in the Python interface.

Batch Generation

Individually generating events and then accessing each particle by the Python interface can slow things down considerably with respect to running directly in C++. This is because the Python interface must translate the C++ PYTHIA event record into Python as each particle is accessed by the user. Similarly, using PythiaParallel where a Python function is passed back to the Pythia C++ code to perform an analysis can result in a slow down. Another option is to batch translate events to Python using the Awkward array library. In this way, the event translation is all done in a single C++ call for a group of events, rather than on a per particle basis. An example on how to generate events with the Awkward interface are given in main297.py.

For this batch interface to work, your Python version must be greater than 3.5 and the Awkward module must be available in Python. This can be installed via pip or the Python package manager of your choice.

 
python -m pip install awkward

It should then be possible to use the nextBatch method of the Pythia class which is called with two arguments nextBatch(nEvents, errorMode), where the second argument is optional. In the following code, 100 events are generated and returned via the Awkward array named events.

 
pythia = pythia.Pythia() 
pythia.readString("HardProcess:all = on") 
pythia.init() 
events = pythia.nextBatch(100)

For a single Pythia instance, failed events can be treated in different ways using the optional errorMode argument that can be passed to nextBatch.

"none" - events that fail to generate are included in the Awkward array as None objects.
"skip" - this is the default behavior, and is similar to what PythiaParallel also does. Events that fail to generate are skipped, and events are generated until the requested nEvents is reached.
"fail" - if an event fails to generate, then an exception is thrown.
fac - here, fac is a real number greater than 1. This is the same as "skip" but event generation will terminate after fac*nEvents attempts have been made to generate an event and an exception will be thrown.

Some examples are given in main298.py.

There are two other features for batch generation. First, nextBatch can also be called from a PythiaParallel instance, but without the optional errorMode as PythiaParallel handles failed events internally. Second, the momentum and production vertex four-vectors can automatically be translated into a four-vector class using the scikit-hep/vector library.

 
python -m pip install vector

Translating automatically to this library can then be accomplished with the following code.

 
import vector 
vector.register_awkward()

This allows for analyses to be performed on the event record with slicing. Some examples can be found in the Awkward array documentation, as well as in main297.py.

Simplified Interface

An attempt has been made to provide the everyday functionality of PYTHIA through the simplified interface. However, if classes or methods appear to be missing, please contact the PYTHIA authors; additional functionality can be included upon request. The following list highlights the available classes and methods in this interface, roughly categorized by function. Note that help(pythia8) will return all the available classes and methods in the interface.

top level: Pythia
event access: Event, Particle, Vec4, RotBstMatrix
particle data: ParticleData, ParticleDataEntry, DecayChannel
settings and information: Info, HIInfo, Settings, Flag, Mode, Parm, Word, FVec, MVec, PVec, WVec
user interface pointers: Rndm, PDF, DecayHandler, RndmEngine, UserHooks, MergingHooks, BeamShape, SigmaProcess, TimeShower, SpaceShower, HeavyIons, PartonVertex
analysis tools: Hist, HistPlot, CellJet, ClusterJet, SlowJet, SlowJetHook, Sphericity, Thrust
kinematic functions: m, m2, dot3, cross3, cross4, theta, costheta, phi, cosphi, RRapPhi, REtaPhi

Installation

To install the Python interface, the Python system header Python.h must be available. By default when configuring using --with-python the system python-config script will be used to determine the location of Python.h. In some cases, users might wish to use a different version of Python, and so --with-python-config can be used to set the Python configuration script used to pick up paths. Alternatively, the directory containing Python.h can be set manually with the option --with-python-include. Some example configurations could be as follows,

 
    ./configure --with-python 
    ./configure --with-python-config=python3-config 
    ./configure --with-python-include=/usr/include/python2.7

where the Python configuration script and path must be changed accordingly for the local system.

After configuring the Python interface for PYTHIA to be built and running make as usual, the following files should be generated in the directory lib.

pythia8.so: compiled Python module.
libpythia8.[so,dylib]: the standard shared PYTHIA library.

To ensure that the pythia8.so module is available to Python, the system variable PYTHONPATH should be set similar to

 
    export PYTHONPATH=$(PREFIX_LIB):$PYTHONPATH

where PREFIX_LIB is the directory lib which contains pythia8.so. Generally, the library paths should be set correctly, but it also may be necessary to set

 
    export LD_LIBRARY_PATH=$(PREFIX_LIB):$LD_LIBRARY_PATH

where DYLD should be substituted for LD in OS X. Alternatively, it is also possible to define the Python path from within Python, as is done within the provided examples. Note that the Python module is always compiled as pythia8.so, even in OS X. This is because older versions of Python in OS X do not correctly recognize modules ending with the dylib suffix.

Examples

To use the Python interface for PYTHIA, start Python and import pythia8. The provided examples can be run by python mainNNN.py where NNN is the number of the example.

Interface Generation

A script for automatically generating the Python interface, generate, is available in plugins/python/ and should be run from this directory. This script requires that the user has Docker installed. A small Docker image of roughly 80 MB will be pulled to run the necessary generation code. There are a number of command line arguments which can be passed.

--help prints the help message for the script.
--part generates the simplified interface. When generate is run without arguments, then this is the default behaviour. Note that this requires running the interface generation twice. The first time the required classes are interfaced, the second time additional classes are removed.
--full generates the full interface.
--user=FILE generates a user defined interface using the Binder configuration given by FILE.
--debug can be specified with any of the options above and stops the temporary files used in generation from being removed.

Whenever PYTHIA headers are modified, the Python interface is invalidated, and so this automatic generation is particularly useful in such a situation. An example of generating the full Python interface is as follows.

 
    cd plugins/python 
    ./generate --full