Skip to content

Optimisation

Kalix’s in-built optimisation tools supports calibration, and general optimisation tasks.

A range of optimisation algorithms are available. Each makes different trade-offs between robustness and optimisation efficiency. SCE is a good choice for difficult problems where a high number of model evaluations is acceptable. DE is a more efficient at finding good solutions quickly, but is not as thorough as SCE.

Algorithm
SCE-UA Shuffled Complex Evolution (SCE)
DE Differential Evolution (DE)
Not implemented yet (below)
SC-SAHEL Shuffled Complex Self Adaptive Hybrid Evolution (SC-SAHEL)
DREAM Dream

Optimisable parameters

Optimisations are defined against parameter addresses — model constants (const.*) and node parameters (node.<name>.<param>). The full reference of what can be tuned, with typical search ranges, is at Optimisable parameters.

Objective functions

Check them out at Objective functions

Pages

ReparameterizationsObjective functionsShuffled Complex Evolution (SCE)Shuffled Complex Self Adaptive Hybrid Evolution (SC-SAHEL)Differential Evolution (DE)DreamCovariance Matrix Adaptation Evolution Strategy (CMA-ES)