The interface between FastJet and NumPy

cern, cython, hep, numpy, python, scikit-hep
pip install pyjet==1.9.0


⚠️pyjet is deprecated, largely unmaintained, and superseded⚠️

  • pyjet has been maintained minimalistically over the past couple of years. In the meantime a new project started, to provide a modern Pythonic jet-finding package in the Scikit-HEP ecosystem.
  • fastjet provides official FastJet bindings to Python and Awkward Array. Refer to the GitHub repository for details.

pyjet: the interface between FastJet and NumPy

PyPI version Zenodo link Test status Wheel builds Binder

pyjet allows you to perform jet clustering with FastJet on NumPy arrays. By default pyjet only depends on NumPy and internally uses FastJet's standalone fjcore release. The interface code is written in Cython that then becomes compiled C++, so it's fast. Remember that if you use pyjet then you are using FastJet and should cite the papers listed here.

Strict dependencies

Getting started

pyjet provides the cluster() function that takes a NumPy array as input and returns a ClusterSequence from which you can access the jets:

from pyjet import cluster
from pyjet.testdata import get_event

vectors = get_event()
sequence = cluster(vectors, R=1.0, p=-1)
jets = sequence.inclusive_jets()  # list of PseudoJets
exclusivejets = sequence.exclusive_jets(3)  # Find the cluster history when there are 3 jets

The input is given in the form of a structured array in numpy. The first four fields of the input array vectors must be either:

np.dtype([('pT', 'f8'), ('eta', 'f8'), ('phi', 'f8'), ('mass', 'f8')])

or if cluster(..., ep=True):

np.dtype([('E', 'f8'), ('px', 'f8'), ('py', 'f8'), ('pz', 'f8')])

Note that the field names of the input array need not match 'pT', 'eta', 'phi', 'mass' etc. pyjet only assumes that the first four fields are those quantities. This array may also have additional fields of any type. Additional fields will then become attributes of the PseudoJet objects.

See the examples to get started:

Standalone Installation

To simply use the built-in FastJet source, from your virtual environment, run:

python -m pip install pyjet

And you're good to go! If you have a old version of pip (<10), you will need to have Cython and Numpy already installed to build from source - however on most systems, you should get a binary wheel.

Get and run it:

    curl -O
    jet#          pT        eta        phi       mass  #constit.
    1        983.280     -0.868      2.905     36.457         34
    2        901.745      0.221     -0.252     51.850         34
    3         67.994     -1.194     -0.200     11.984         32
    4         12.465      0.433      0.673      5.461         13
    5          6.568     -2.629      1.133      2.099          9
    6          6.498     -1.828     -2.248      3.309          6

    The 6th jet has the following constituents:
    PseudoJet(pt=0.096, eta=-2.166, phi=-2.271, mass=0.000)
    PseudoJet(pt=2.200, eta=-1.747, phi=-1.972, mass=0.140)
    PseudoJet(pt=1.713, eta=-2.037, phi=-2.469, mass=0.940)
    PseudoJet(pt=0.263, eta=-1.682, phi=-2.564, mass=0.140)
    PseudoJet(pt=1.478, eta=-1.738, phi=-2.343, mass=0.940)
    PseudoJet(pt=0.894, eta=-1.527, phi=-2.250, mass=0.140)

    Get the constituents as an array (pT, eta, phi, mass):
    [( 0.09551261, -2.16560157, -2.27109083,   4.89091390e-06)
     ( 2.19975694, -1.74672746, -1.97178728,   1.39570000e-01)
     ( 1.71301882, -2.03656511, -2.46861524,   9.39570000e-01)
     ( 0.26339374, -1.68243005, -2.56397904,   1.39570000e-01)
     ( 1.47781519, -1.7378898 , -2.34304346,   9.39570000e-01)
     ( 0.89353864, -1.52729244, -2.24973202,   1.39570000e-01)]

    or (E, px, py, pz):
    [( 0.42190436, -0.06155242, -0.07303395, -0.41095089)
     ( 6.50193926, -0.85863306, -2.02526044, -6.11692764)
     ( 6.74203628, -1.33952806, -1.06775374, -6.45273802)
     ( 0.74600384, -0.22066287, -0.1438199 , -0.68386087)
     ( 4.43164941, -1.0311407 , -1.05862485, -4.07096881)
     ( 2.15920027, -0.56111108, -0.69538886, -1.96067711)]

Reclustering the constituents of the hardest jet with the kt algorithm
[PseudoJet(pt=983.280, eta=-0.868, phi=2.905, mass=36.457)]

Go back in the clustering sequence to when there were two jets
PseudoJet(pt=946.493, eta=-0.870, phi=2.908, mass=20.117)
PseudoJet(pt=36.921, eta=-0.800, phi=2.821, mass=4.119)

Ask how many jets there are with a given dcut
There are 9 jets with a dcut of 0.5

Get the jets with the given dcut
1 PseudoJet(pt=308.478, eta=-0.865, phi=2.908, mass=2.119)
2 PseudoJet(pt=256.731, eta=-0.868, phi=2.906, mass=0.140)
3 PseudoJet(pt=142.326, eta=-0.886, phi=2.912, mass=0.829)
4 PseudoJet(pt=135.971, eta=-0.870, phi=2.910, mass=0.140)
5 PseudoJet(pt=91.084, eta=-0.864, phi=2.899, mass=1.530)
6 PseudoJet(pt=30.970, eta=-0.831, phi=2.822, mass=2.124)
7 PseudoJet(pt=7.123, eta=-0.954, phi=2.939, mass=1.017)
8 PseudoJet(pt=5.951, eta=-0.626, phi=2.818, mass=0.748)
9 PseudoJet(pt=4.829, eta=-0.812, phi=3.037, mass=0.384)

Using an External FastJet Installation

To take advantage of the full FastJet library, including the jet area calculations and the optimized O(NlnN) kt and anti-kt algorithms, you can first build and install FastJet and then install pyjet with the --external-fastjet flag. Before building FastJet you will need to install CGAL and GMP.

On a Debian-based system (Ubuntu):

sudo apt-get install libcgal-dev libcgal11v5 libgmp-dev libgmp10

On an RPM-based system (Fedora):

sudo dnf install gmp.x86_64 gmp-devel.x86_64 CGAL.x86_64 CGAL-devel.x86_64

On Mac OS:

brew install cgal gmp wget

Then run pyjet's script:

curl -O
chmod +x
sudo ./

Now install pyjet like:

python -m pip install numpy Cython
python install --external-fastjet

pyjet will now use the external FastJet installation on your system.

Note on units

The package is indifferent to particular units, which are merely "propagated" through the code. We do recommend that the HEP units be used, as defined in the units module of the hepunits package.

It is worth noting that the azimuthal angle phi is expressed in radians and varies from pi to pi.


If you want to setup for development:

python3 -m venv .env
source .env/bin/activate
pip install -e .[dev]