This repository contains code and scripts for evaluating and benchmarking the performance and quality of various Pseudo-Random Number Generators (PRNGs) implemented in both C and modern Machine Learning (ML) frameworks.
The purpose of this project is to analyze the efficiency and statistical quality of PRNGs in different environments, such as PyTorch, TensorFlow, and NumPy, and compare them against the original C implementations of Mersenne Twister (MT), PCG, Philox, and Mrg32k3a.
The folder integerValues contains the tests made on 32 bits integer pseudo random numbers.
The folder realValues contains the tests made on 64 bits double pseudo random numbers.
Link of the paper : https://arxiv.org/pdf/2401.17345
In the two folders, you should open Jupyter notebook. Experiments are described inside it.