Hello,

I am a student researcher who is new to the HPC domain.

I have recently taken a project where I am working on optimizing the parameter search space (both application level and hardware level parameters) for running HPC applications. In order to do this, I am trying to implement the Multi-arm bandit approach to find the optimal parameters by taking each parameter as an arm of the said algorithm and the performance metric(Accuracy, energy, execution time, etc.) as the reward for that particular arm.

But, I am facing difficulty in how implementing this approach experimentally. I was thinking about getting a dataset that can map a parameter (or a certain group of parameters) to the said performance metric. I have seen some previous works which deal with this kind of problem finding the best optimal parameters (Example). But, these works do not share the dataset that they worked with. Does anyone know where I can get access to the dataset which is befitting to my particular problem?