Vectorisation like `np.vectorize` or `jax.vmap`

Open
#1,111 0 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
25/100
Issue type
Feature
Clarity
Needs clarification
Activity status
Stale
Tech stack
numpy, python, rust
Domain
data

Research direction

The issue mentions no files, tests, or entry points in the ndarray repository. Start by surveying the existing array and function APIs, then clarify the proposed vectorisation interface, supported argument patterns, parallelisation expectations, and tests needed to define when the feature is complete.

Written by the indexing model from the issue text.

Description

enhancement help wanted

This is a feature suggestion.

One of the important feature of numpy and jax in Python is to provide a vectorisation scheme for function.

Here is a mini example in Python:

import jax.numpy as np

# This function takes two float arguments and output a float
def func(x: float, y: float) -> float:
    return x - y

# This vmap will give a vectorised func, now this vectorised_func
# can take an array as input and produce an array output.
# Here[0, None] means that we vectorise func's first argument only.
vectorised_func = jax.vmap(func, in_axes=[0, None])

# Test arrays
a = jnp.array([1., 2., 3])
b = 2.

vectorised_func(a, b)  # This will output an array([-1, 0, 1])

# You can even do more to vectorise the y argument of func also
even_more_vectorised_func = jax.vmap(vectorised_func, in_axes=[None, 0])

# 
c = jnp.array([3., 2., 1.])
even_more_vectorised_func(a, c) # This will output a matrix ([[-2, -1, 2], [-1, 0, 2], [0, 1, 2]])

In numpy, this is similar to np.vectorize, but np.vectorize is actually not using any parallelisation scheme unlike jax.vmap.

This feature, in my opinion, is very important for scientific computing. For example, if you have two arrays, you can use this to compute their pairwise distances. As an another example, in multidimensional numerical quadrature, computing \int f(x) dx requires to evaluting the function f by a number of node points, this could be significantly improved by vectorisation.

I am a researcher in signal processing and machine learning, and I can definetly say this feature is valuable.

Dominant language
Rust
Stars
4.3k
Forks
391
No merged PRs in 30d

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from rust-ndarray/ndarray

All issues in rust-ndarray/ndarray

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.