Skip to content

Latest commit

 

History

History
24 lines (23 loc) · 1.05 KB

TODO.md

File metadata and controls

24 lines (23 loc) · 1.05 KB

TODO

  • Add more examples
  • Using the NamedTupleAdaptor, and remove the convert_to_ntp kwargs in component running
  • Write more error messages
  • macro building
  • The package is too heavy, which needs to be splitted into small packages
  • Flux output need constraint
  • Add NoneMeanFlux
  • GPU support (limited by scalar index)
    • DataInterpolations.jl is not support for now
  • meta should store variables and parameters
  • compact problem
  • ensemble model
  • base merge for component meta
  • Neural Flux make the gradient computation slow a lot
    • @btime gradient计算: NeuralFlux (84 us) 略大于 常规计算 (69 us)
    • @btime gradient计算: TotalModel (90s) 显著大于 常规计算 (23 s)
    • 原因:
    1. 多个bucket遍历插值
  • 针对multiple nodes模型参数的最佳设置方式,需要考虑计算效率, 考虑采用 ComponentVector(params=(p1=[2,3,4], p2=[3,3,4]))
  • 添加gradient的test
  • 添加参数检查, 参数默认设置的功能
  • 添加说明文档