numsci

num[ba]sci[py] - A Numba-Compatible Port for Popular Scipy Functions

Most Scipy Functions do not work in jit-compiled Numba functions. This project aims to make this possible.

How to use

Model functions must abide by the following convention

from numsci.optimize import model_sig

@model_sig
def model_function(x: float64, params: ndarray(dtype=float64)) -> float64:
    ***do computation***
    return result

Where the @model_sig decorator indicates that the function is a Numba Cfunc of signature float64(float64, *float64).

Outside of the jit-compiled function, the caller must also obtain the address to the model function

model_function_address = model_function.address

This can then be used to call Scipy-like functions like curve_fit() in a @njit decorated function

from numsci.optimize import curve_fit

@njit
def njit_function():
    fvec, pcov = curve_fit(model_function_address, xdata, ydata)

Setup

Install dependencies and compile

pip install -r requirements.txt
bash build.sh