Seed experiments


I would like to conduct some experiments on the run time of my algorithm, therefore I would like to know if it is possible to seed a script with concrete numpy ?
Does a np.random.seed(seed) is enough or is there a specific function for it in the library ?


Hello @tricycl3,

If you want to have consistent interaction with concrete-numpy (e.g., have the same exact tracing result each time), then you want to seed numpy and torch, and that should be enough.

However, if you want to have consistent FHE execution behavior (i.e.,, it’s not possible at the moment. Underlying cryptographic libraries use their own RNG implementation, which is not interactable at the moment.

Hope this answers your question!

Thanks that answers exactly my question !

Glad to help, let us know if you have more questions!