Skip to content

Latest commit

 

History

History
27 lines (20 loc) · 599 Bytes

roadmap.md

File metadata and controls

27 lines (20 loc) · 599 Bytes

BenchOpt

  • API

    • Easy to adapt new algorithms
    • Think about f/f_prime API?
    • Independent of python implementation?
    • Multiple datasets
    • Easy install
    • Easy command line interface
  • Support

    • TPU/GPU/parallelism
    • PR + unit test
    • Can run in local
    • Local run with a callback API?
  • complicated part:

    • Same definition of the objective (mean vs sum)

Current TODO:

  • Adapt n_iter spacing to time (while loop)
  • Compute optimal solution
  • Argument in client to pass solvers to run in benchmark
  • Make CI run the benchmark and check install