While it is currently possible to restart a run from a known snapshot, the method
is quite involved and error-prone (needs to happen at a certain point in the flow,
many values need to be set, the optimizers might need patching, ...).
This is a useful feature for time-limited environments, allows for switching optimizers,
and enables tweaking hyper-parameters.
Therefore, please make a one-line 'restart from here' method, preferable with these
features
either a single individual or a full population can be given (the latter is only sensible
in population-based algorithms)
this allows for scaling the populations (for GA) up and down
or switching the optimizer GD <-> GA
generation count is kept (by default) or reset (if asked)
While it is currently possible to restart a run from a known snapshot, the method is quite involved and error-prone (needs to happen at a certain point in the flow, many values need to be set, the optimizers might need patching, ...).
This is a useful feature for time-limited environments, allows for switching optimizers, and enables tweaking hyper-parameters.
Therefore, please make a one-line 'restart from here' method, preferable with these features