Jakob Jordan
Jakob Jordan
Previously we used github pages to host the documentation (https://happy-algorithms-league.github.io/hal-cgp/), but now we use readthedocs (https://readthedocs.org/projects/hal-cgp/). We should remove github pages to avoid confusing, shouldn't we @mschmidt87 ? If so,...
It seems to me we have reached the point where some of the awesome advanced features of hal-cgp may be too hidden from the user. We should enrich the README...
Currently, no `__repr__` is implemented for `Population`: ```python >>> print(pop) ``` We should support easily printing a summary of the a population instance, for example including information about parameters, champions,...
This method is currently used in the beginning of deterministic objectives to avoid reevaluating individuals which already have a fitness value assigned: ```python def objective(individual): if not individual.fitness_is_None(): return individual...
If a decorated objective function changes individuals, for example by setting new member variables which influence the fitness in subsequent objective function calls, the consistency check can fail. This should...
If one is using hurdles/multiple objective functions, there is currently no way to set the termination fitness for each individually. Rather the fitness of all objectives is combined and compared...
Since #294 there seems to be a difference between how the torch module is called and the other "compiled" callables are used. This should be unified by adjusting the generated...
Currently we set the fitness of an individual like `individual.fitness = ` in the objective function. Internally, this triggers the setter `self._fitness[self._objective_idx] = v` where `_objective_idx` is set by the...
Instead of using a multi-start scheme in which multiple instance of the evolutionary algorithm run in parallel without communicating, one may consider collections of populations, e.g., "island models/archipelagos" (https://link.springer.com/chapter/10.1007%2F978-3-642-28789-3_7), which...