Group experiments according to hypotheses and CPU v. GPU

[ravivanpong]

We should have the following structures

• experiments
  • cpu: for algos that does not need GPU
    • parameters_default: experiments under default parameters
    • parameters_optimized: experiments under optimized parameters (best of best competition)
    • hyperparameter_optimization: document parameter optimization process
  • gpu: for algos that require GPU to work at all
    • analogous to cpu