Denumpyfy.jl

I hate translating numpy notebooks to Julia.

I then spent about 2 afternoons packing 20 regexes in to a trenchcoat and calling it a Julia package so that others may join me in not fiddling with indexing of nasty, nasty "vectorized" code.

My code is quite naive - it tries to be a bit smart about line by line regex replacing, and doesn't really handle global context or self rewriting - and likely won't. It was written with replacing tutorial numpy notebook tutorials in mind. If you are looking for something that is much more robust/heavy duty, please check out Py2Jl.jl -> py2many. You can compare Denumpyfy.jl vs py2many with puncture.jl vs tests/demopy2many.jl, respectively. At time of writing, their AST parser is an unfinished copy/paste from their Rust transpiler, so please help out by filing PRs/issues over there.

Use entirely at your own risk, and see some of the tests/runtests.jl to see what the package can help you out with.

For example, this code in test/puncture.py

for i in range(0, self.n_grid-1):
    for j in range(0, self.n_grid-1):
        for k in range(0, self.n_grid-1):
            delta_x = x[i, -1] ** 2 + y[j, -2] ** 2 + z[k, -3] ** 2

will be turned into this with denumpyfy("puncture.py")

for i in 1:self.n_grid
    for j in 1:self.n_grid
        for k in 1:self.n_grid
            Δ_x = x[i, end] ^ 2 + y[j, end-1] ^ 2 + z[k, end-2] ^ 2

in a new file puncture.jl.

There's replace rules for array indexing (with 1-index bumping), some ascii -> Unicode replacements, range(0,n-1): to 1:n, and the obvious arithmetic ** to ^.

TL;DR:

Sponsorship

If you want to see more of this work, consider sponsoring me via Github sponsors.

Acknowledgments

This work was done via nerdsnipe and support of Brenhin Keller.