Skip to main content

Table 1 Covariates

From: Guidelines for benchmarking of optimization-based approaches for fitting mathematical models

AbbreviationCovariateTypical possible choices
C1Application problemModel equations and the data set(s)
C2Primary performance criteriaConvergence per computation time, iteration steps
C3Secondary performance criteriaDocumentation, user-friendliness, code quality
C4Parameter scaleLinear vs. log scale
C5Global search strategyMultiple initial guesses, scatter search algorithms, stochastic search
C6Initial guess strategyFixed vs. random, normally distributed vs. uniform vs. latin-hypercube
C7Parameter constraintsUpper and lower bounds
C8Prior knowledgeNone vs. (weakly) informative priors
C9ODE integration implementationSUNDIALS, Matlab, R
C10ODE integration algorithmStiff vs. non-stiff approaches, Adams-Moulton vs. BDF
C11Integration accuracyODE integrator tolerances
C12Derivative calculationFinite differences, sensitivity equations, adjoint sensitivities
C13Stopping ruleOptimization termination criteria
C14Handling of non-converging ODE integrationTermination of optimization vs. infinite loss
C15Algorithm-specific configurationsCross-over rate, annealing temperature, number of particles
  1. The performance of an optimization approaches depend on many decisions and configurations C1–C15. For the comparison of several approaches, these attributes appear as covariates. Performance benefits for individual choices do not necessarily indicate a general advantage because benefits might merely originate from the chosen configurations