#matrix #vector #nalgebra #normal-distribution #parallel-processing #random-normal

randn

A package to populate nalgebra's vector, matrix and vector of matrices with standard normal random variables in parallel

4 releases

0.1.12 Nov 8, 2023
0.1.11 Nov 4, 2023
0.1.1 Nov 3, 2023
0.1.0 Oct 28, 2023

#255 in Finance

25 downloads per month

MIT/Apache

7KB
82 lines

# Randn Rust Library

The Randn Rust library provides utilities for generating random vectors, matrices, and a list of matrices with values sampled from the standard normal distribution. It leverages the power of parallel processing to efficiently generate large sets of data.

## Dependencies

This library is built upon a set of Rust crates:
- `rand` and `rand_distr` for random number generation.
- `nalgebra` for array operations and representations. (Switched from `ndarray`)
- `rayon` for parallel computing capabilities.

## Features

1. **Parallel Random Data Generation**
   
   The library uses parallel processing to generate random numbers faster:
   ```rust
   fn parallel_randn(size: usize) -> Vec<f64> {...}
  1. Random Vector Generation

    Generates a random vector of the specified size using ndarray's Array1:

    pub fn randn_vector(size: usize) -> Array1<f64> {...}
    
  2. Random Matrix Generation

    Generates a random matrix of the specified rows and columns sizes using ndarray's Array2:

    pub fn randn_matrix(rows: usize, cols: usize) -> Array2<f64> {...}
    
  3. Generation of Multiple Random Matrices

    Generates a vector of random matrices of specified rows, columns, and number of simulations using ndarray's Array2:

    pub fn randn_matrices(rows: usize, cols: usize, sims: usize) -> Vec<Array2<f64>> {...}
    

Usage

To use the RandN library in your project, include it in your Cargo.toml dependencies and then import the necessary functions in your Rust code.

Contribution

Contributions to the RandN library are welcomed! Please open an issue or submit a pull request on the GitHub repository.

License

This library is licensed under the MIT License.

Dependencies

~11–22MB
~323K SLoC