Hawkes processes, but make them pip-installable

What it does

tick is a Python 3 statistical learning library focused on time-dependent systems — point processes above all. You can simulate Hawkes processes with standard or exotic kernels, then infer them under various kernel assumptions (exponential, sums of exponentials, basis-kernel combinations, sparse interactions). It also covers linear, logistic, and Poisson regression, plus a generic optimization core of models, solvers, and proximal operators.

The interesting bit

The library isn’t just model classes — its heart is a reusable optimization toolbox with proximal regularization, backed by C++ extensions and Intel MKL. That’s the boring part, and it’s where the leverage is: the Hawkes inference sits on top of a general solver stack, not bespoke per-model code.

Key highlights

• Simulate and infer Hawkes processes with flexible kernel choices, including sparse-interaction variants
• Generalized linear models: linear, logistic, Poisson regression
• Optimization core with models, solvers, and proximal operators usable on their own
• C++ extensions linked against Intel MKL, efficient from desktops to HPC servers
• Real industrial use: pharmacovigilance on French health-care data, order-book modeling, Twitter information propagation during the 2017 French election

Caveats

• Windows support is explicitly experimental; Linux/macOS are the first-class platforms
• Installation compiles C++ extensions, so it can take a few minutes
• Requires Python 3.5 or newer — the project dates from 2016, so check how current the ecosystem around it feels

Verdict

If you model event streams — finance order books, social media cascades, adverse drug events — this is a rare, purpose-built tool with a published paper behind it. If you just need sklearn-style tabular ML, it’s overkill.