Is is possible to manually initialize the population?

[niaorg]

Is is possible to manually initialize the population?

[kb2623]

Hi @jhmenke

I am working on update that will have your requested functionality. You can have a look at my repository of NiaPy. There is still a lot of work to be done on old algorithms. So the fix won't reach release so soon.

[firefly-cpp]

Hi @jhmenke,

Thank you very much for a very interesting suggestion. We will try to add this
feature in the following days.

[GregaVrbancic]

Hi @jhmenke, thank you again for showing the interest in our micro-framework. @kb2623 is working actively on this functionality. As soon as we reach the release phase, we will let you know.

[jhmenke]

i will monitor your repository, thanks!

[jhmenke]

May i ask what the status of the issue is?
I have seen promising commit messages: kb2623@bfee01b

[kb2623]

Now you can pass a function for population initialization to all algorithms. You have to pass InitPopFunc parameter to setParameters function or to algorithms constructor. Parameter should be a function that takes int, Task and mtrand.RandomState as parameters. In Algorithm class you can see the example of two predefined functions for population initialization for additional help.

[jhmenke]

Will try asap! Thank you for your time!

[GregaVrbancic]

@jhmenke the pull request addressing the issues is now merged into the master branch.

Please, try it out and give us some feedback.

[FlorianShepherd]

@jhmenke the pull request addressing the issues is now merged into the master branch.

Please, try it out and give us some feedback.

Dear @GregaVrbancic,
I tested the code in behalf of @jhmenke . Thank you so much for adding this feature. It seems to work great :)

However, I wondered why some algorithms didn't support the initPopFunc. Either I did something wrong when creating the task and calling algorithm.run(task) or it is really not supported (yet). I noticed this for PSO and MFO for example. Maybe others as well.

I fixed this for the PSO here:
FlorianShepherd@9cbe071

All I had to do was to add "InitPopFunc" to "setParameters".

Also I added the option "binary" for the PSO, which rounds the velocity V to integers. This way I can use the PSO as a binary PSO like here: https://pyswarms.readthedocs.io/en/latest/examples/feature_subset_selection.html#using-binary-pso

[GregaVrbancic]

Dear @FlorianShepherd,

thank you very much for taking an effort and testing out the new feature 😃 Regarding the problem with some algorithms, we will take a look at it as soon as possible.

The "binary" option for the PSO also looks promising, we might also integrate it into NiaPy.

[FlorianShepherd]

Probably it is just my mistake. I (have to) run the optimization like this, otherwise my initFunc() is not called:

from NiaPy.algorithms.basic.pso import ParticleSwarmAlgorithm

    task = Mytask(max_evals=max_evals, D=D, NP=NP, nGEN=nGEN,
                  benchmark=benchmark_class(), InitPopFunc=initfunc, binary=True)
    alg = ParticleSwarmAlgorithm(max_evals=max_evals, D=D, NP=NP, nGEN=nGEN,
                  benchmark=benchmark_class(), InitPopFunc=initfunc, binary=True)
    alg.setParameters(max_evals=max_evals, D=D, NP=NP, nGEN=nGEN,
                  benchmark=benchmark_class(), InitPopFunc=initfunc, binary=True)
    result, _ = alg.run(task=task)

It seems like I have to call each function with the same parameters, otherwise an inconsistent default value is used.

[jhmenke]

thanks for the updates! i have tried PSO and it works well with the initlization!

there are some problems with other algorithms with the new initialization, e.g. the artificial bee colony. due to the Foods now being a numpy array, the algorithm is broken.

newSolution.x[param2change] = Foods[i].x[param2change] + (-1 + 2 * self.rand()) * (Foods[i].x[param2change] - Foods[neighbor].x[param2change])

[GregaVrbancic]

@jhmenke thanks for the feedback!

@kb2623 could you take a look at this?

[kb2623]

Hi.

I have added missing code to algorithms that where not using passed function for population initialization.

If you want to write a problem specific algorithm, you can rewrite initPopulation method form algorithm you are trying to solve a specific problem. For examples on how to rewrite initPopulationmethod you can check other algorithms. For example I suggest that you check CamelAlgorithm or FishSchoolSearch algorithms.

[GregaVrbancic]

@all-contributors please add @jhmenke for question

[GregaVrbancic]

The example of custom initialization of population is available here.

[allcontributors]

@GregaVrbancic

I've put up a pull request to add @jhmenke! 🎉