- Implementation of a new learning rate scheduler: two-way backtracking line search.
- Added in the C++ model classes: pre-calculation now leverage the Markov property of the exponential kernel.
- The
LearnerHawkesExpis now compatible with scikit-learn routines for variable selection, enabling an easy-to-use pipeline for selecting the decay hyperparameter when it is unknown. - Providing
end_timewith data is now optional.
-
Initial release of Sparklen with core features:
-
A efficient cluster-based simulation method for generating events.
-
A highly versatile and flexible framework for performing inference of multivariate Hawkes process.
-
Novel approaches to address the challenge of multiclass classification within the supervised learning framework.
-