Samples from convex combination space.

In theory this module "only" provides a bit of math. In practice, it's a useful piece of abstraction.

This module provides a function that pickes a convex combination of a desired dimension (= number of items), uniformly distributed among all convex combinations. It is possible supply your own RNG, or a hand-picked point from a hypercube (uniform distribution in the hypercube guarantees uniform distribution in "convex combination"-space).

• Install
• Usage
• TODOs
• Contribute

Just pip install sample_convex_combinations. (As soon as I have uploaded it to PyPI.)

Or just copy sample_convex_combinations/_impl.py into your project as sample_convex_combinations.py.

There are no dependencies outside stdlib.

Let's say you want to interpolate between three things, and wish the sum of the weights to be 1:

>>> import sample_convex_combinations
>>> sample_convex_combinations.sample(3)
[0.6847422669970515, 0.6847422669970515, 0.6847422669970515]  # FIXME

Or maybe you want the sum of the weights to be between 0 and 1:

>>> sample_convex_combinations.sample(3, span_origin=True)
[0.49931160565103394, 0.49931160565103394, 0.49931160565103394]  # FIXME

The default random.random is not enough for your needs? Sure thing! You can substitute your own RNG, for example from secrets:

>>> import secrets
>>> sample_convex_combinations.sample(3, rng=secrets.randbits(64) / (2 ** 64))
[0.49931160565103394, 0.49931160565103394, 0.49931160565103394]  # FIXME

Isn't that nice? :D

FIXME

• Implement
• Document
• Do WeirdStuff™ with it
• Advertise

Feel free to dive in! Open an issue or submit PRs.