[Feature Request] Random numbers generation functions

[SemyonSinchenko]

What we have

Currently PySpark provide only two methods of generating random numbers:

1. rand which generates numbers from $\sim U(0, 1)$
2. randn which generates numbers from $\sim N(0, 1)$

What I want to have?

• Random integers generation
• Samples from $\sim U(\alpha, \beta)$
• Samples from $\sim N(\alpha, \beta)$
• Samples from $\sim L(\alpha, \beta)$ (Laplace distribution)
• Samples from $\sim \Gamma (\alpha, \beta)$ (Gamma distribution)

Motivation

1. Random integers is important for a lot of reasons and re-partitioning is one of them.
2. Gaussian, Laplace and Gamma distributions are a key part to implement Differential Privacy Driven aggregations (for example, Additive noise mechanism and especially Laplace mechanism).

How to do it?

All distribution can be generated from the uniform. The idea is to provide a top level functions.

[MrPowers]

@SemyonSinchenko - yep, I like this idea. Can you please propose the function names => expected interfaces at a high level? e.g. "I think we should have a quinn.laplace(col: Column) function that returns a Leplace number". These should be useful in a variety of contexts.

[SemyonSinchenko]

My proposal is quinn.math.laplace(a: float, b: float) -> Column
I can make a PR with this and we will discuss it.

[MrPowers]

@SemyonSinchenko - sounds good, going to assign you this issue now.