This first release contains the Microtune code and 2 dataset files (.pickle) used to train and test models.
Both render possible to reproduce all experiments (train and test Bandits and RL models) as described in the research paper Microtune which is still under review for publication (ask us if you want it).
The code is configured thanks to python Hydra framework, allowing reliability of results and experiments.
The dataset is obtained from metrics observed on a MariaDB database stressed by a Sysbench client with different OLTP workloads (in read/write mode & different profiles).
The complete dataset production required more than 4 weeks of intensive computing.
All is free, under MIT license for the code and Creative Commons for the dataset.