[Epic] Future of AutoEmulate

[mastoffel]

some thoughts on future milestones, sorted by estimated effort to implement, starting with the most complex:

• true end-to-end automatic emulation, which needs:
  • automated input sampling / design of experiments (DoE) using latin hypercubes and similar techniques
  • automated evaluation of simulations using these inputs on an HPC
  • iteratively fitting emulators using increasingly large datasets until a condition is met (e.g. maximum computation time, minimum accuracy of the emulator)
• adding spatial / temporal data and simulations, which needs
  • possibly a new data interface more complex than the current X, y
  • new cross-validation, data-splitting methods that are aware of time and space
• adding more emulator applications, such as:
  • uncertainty quantification
  • optimisation

tbc