cospectrum/grast

Lazy automatic differentiation for Python

grast

Lazy automatic differentiation for Python

Install

Requires python >= 3.10

pip install grast

Usage

Create function R^n -> R

from grast import var

x = var('x')
y = var('y')
z = var('z').freeze()  # do not compute derivative

h = x/y + y**x
f = z * h + 3

Get gradient

df = f.grad()
df_dx = df['x']
df_dy = df['y']

Evaluate with specific arguments

args = dict(x=-3, y=5, z=2)
f(args)
df_dx(args)
df_dy(args)

View in symbolic format

print(str(f))
print(str(df_dx))
print(str(df_dy))

References

1. F. Krawiec, S. Peyton Jones, N. Krishnaswami, T. Ellis, R. A. Eisenberg, A. Fitzgibbon. 2022. Provably correct, asymptotically efficient, higher-order reverse-mode automatic differentiation. Proc. ACM Program. Lang., 6, POPL (2022), 1–30. https://doi.org/10.1145/3498710

2. Jerzy Karczmarczuk. 1998. Functional Differentiation of Computer Programs. In Proceedings of the Third ACM SIGPLAN International Conference on Functional Programming (Baltimore, Maryland, USA) (ICFP ’98). Association for Computing Machinery, New York, NY, USA, 195-203. https://doi.org/10.1145/289423.289442