I can't speak to the complexity of the choices function but if you're proposing an alternative implementation for choices, it would be wise to show the benefits empirically. That is, benchmarks and an explanation of why your implementation would be better than just an example of your code among an example. Additionally, you need to expand your implementation to match all the functionality of the current function and make a pull request to the main Python repository.

Adding Raymond to nosy, the original author of random.choices