energyflow.network

What It Is

EnergyFlow is an open-source Python toolkit for particle physics, especially the analysis of collider events and jet substructure. It is not a general-purpose developer tool, but a domain-specific library at the intersection of scientific computing and machine learning. The current documentation lists the version as 1.3.3. The source code is hosted on GitHub and released under the GPLv3 license.

Core Capabilities

In terms of functionality, EnergyFlow covers Energy Flow Polynomials (EFPs), Energy Flow Networks (EFNs), Particle Flow Networks (PFNs), Energy Mover's Distance (EMD), and Energy Flow Moments (EFMs). EFPs are used to compute IRC-safe jet substructure observables; EFN/PFN are Keras-based models for learning from unordered, variable-length sets of particles; EMD uses Python Optimal Transport to compute distances between events; and EFM emphasizes efficient computation in linear time with respect to the number of particles. It also provides capabilities for downloading, reading, and processing datasets such as CMS Open Data, Pythia, Herwig, and Delphes, and includes additional architectures such as CNN, DNN, and LinearClassifier.

Ecosystem, Documentation, and Ease of Use

From an ecosystem perspective, EnergyFlow relies heavily on the Python scientific stack, including NumPy, Keras/TensorFlow, scikit-learn, POT, SciPy, and matplotlib. The documentation is solid, covering installation, demos, examples, FAQ, release notes, architectures, datasets, EMD, EFP, EFM, and more. It also provides extensive parameter explanations and BibTeX references for related papers. Jupyter Notebook demos can be run without local installation via Binder, lowering the barrier for first-time testing; however, compute-intensive tasks such as training EFN/PFN models are still better suited to a local or dedicated computing environment.

Pricing, Pros, and Cons

The main documentation does not mention commercial pricing or paid plans. Given the GPLv3 license and its open-source availability on GitHub, it can be regarded as free and open source. Its strengths include strong academic focus, broad coverage of relevant methods, and complete entry points for examples and datasets. Its drawbacks are the high domain-specific learning curve, limited direct value for general developers, and relatively heavy deep-learning dependencies. The documentation does not indicate commercial support, SLAs, or enterprise services; support mainly appears to come from GitHub Issues and the research community.

Who It Is For and Access from China

EnergyFlow is suitable for researchers and practitioners working on high-energy physics, jet classification, particle-event representation, open data analysis, and paper reproduction. It is not intended as a general-purpose machine learning platform. Regarding access from China, the documentation only mentions external resources such as GitHub, Binder, Zenodo, and CMS Open Data, without providing availability guidance. In practice, downloading datasets and running Binder may be affected by local network conditions, so access is rated as unknown. If access is limited, users can consider building a local Python analysis workflow with components such as NumPy, SciPy, TensorFlow/Keras, scikit-learn, and POT.