deln

GPU-accelerated kernels for exotic / scientific math — 174 operators across 34 domains (special functions, information theory, point processes, copulas, Bayesian & GP methods, PDE solvers, computational geometry, materials, medical imaging, and more), exposed as plain functions you call on torch tensors.

pip install deln

Quick start

import torch
import deln

deln.init("you@company.com")          # REQUIRED once, before any algorithm

from deln.specfun import bessel_i0
x = torch.linspace(0.0, 3.0, 1024, device="cuda")
y = bessel_i0(x)                       # runs on the GPU, returns on x's device

If you call any algorithm before deln.init(...), it raises deln.DelnNotInitializedError telling you to initialize first.

How it works

• Every algorithm is a function grouped under its domain submodule: deln.specfun.bessel_i0, deln.quant-style domains, deln.pointproc.hawkes_log_likelihood, deln.medimg.radon_backprojection, …
• Pass torch tensors (CUDA or CPU); the result comes back on the same device as the inputs — no host round-trips.
• Each function is documented (help(bessel_i0)), with its parameters and the initialization requirement.

Discovering algorithms

import deln
deln.init("you@company.com")
deln.list_algorithms()                 # -> every available operator name
help(deln.specfun.bessel_i0)           # -> docstring, params, notes

Domains

bayesian_inference, bayesmc, compgeo, control, copula, csense, exactalg, funcanal, gametheory, geomproc, harmonics, hjb, homotopy, infogeo, infotheory, interval, ksg, materials, medimg, modred, npstats, numde, optimization, pointproc, prob_programming, qmc, randla, riemann, seqbio, specfun, stats, tdavec, tensordec, uq.

License

Apache-2.0.