Rewrites

  • Add more data set examples
  • Remove skew, Remove the broken CaCO3 analysis
  • Show distributions for increased sampling effort.
  • Does time interval make a difference for convergence of deuterium2 run?  Nope.  Fails in parallel, running in series.

Code:

  • get away from explicit parallelization and deeply nested functions.
  • really master optimizers, and avoid errors & failed convergence.
  • store finished data for plotting as data objects.
  • CaCO3 example -> MLE estimate of unstable model doesn’t converge, shouldn’t be used to simulate then! Add functions to check this convergence.
  • Examples: Could use further examples, not so convincing as is.
  • rescale time before fitting doesn’t change anything.
  • show fixed true pos, fixed false pos.
  • show thresholds, show observed
  • scripts that save must use rm list=ls()
  • G6, G10, ibm_short, deut2 running.