Feeding in solutions from a previous optimisation run

[trbedwards]

I have a previous optimisation run that I would like to feed in solutions and function values from, to give CMA-ES a warm-start. This previous optimisation might not have necessarily been run using pycma.

Is there a way to do this? I've tried to use inject(x_best) from my best solution so far but after calling a few asks and tells, it throws the following error:
ValueError: Found unused injected direction/solutions

[nikohansen]

Generally, this is what the arguments X0 and sigma are for. If we assume X0 is already a pretty good guess, we use a small value for sigma.

When a full run from CMA-ES should be continued, the feed_for_resume method is supposed to "re-create" a previous run from solutions and f-values, however see also #240, which is also "directly" possible using the ask-and-tell interface. These depend on the population size popsize of the current CMA-ES instance to be the same as the one how the data were generated. If the solutions to be fed were not sampled during a run from CMA-ES, this is bound to fail, because the algorithm assumes a particular sample distribution. In this case passing the found best guess as initial solution with a small initial step-size seems the best approach.

Specifically, it's hard to understand the above error without additional information. Having the calling code to be used would be a first step. Generally, it should be possible to inject one or a few solutions in each iteration with the inject method without getting an error (depending on popsize).