Plan

[M4a1x]

This issue tracks our overall progress. Tick stuff that has been done and add new jobs here as they come up. Possibly link to other issues containing more info on the subject.

• [x] Figure out how to implement shuffling with grey code with the Mga [Hjalti]
• [x] Set up a class skeleton with a random start and calculate fitness functionality
• [x] Create fitness function
• [x] Create an evaluate function that runs one iteration and shuffles with the mga[Hjalti]
• [ ] Create a vizualuzation for fitness output i.e [Max]
• [ ] Setup a library to have as a benchmark [Max]
  • [x] Evaluate existing libraries
  • [ ] Setup MWE
• [ ] LONG TERM Look into how to create a pareto front fitness function [Lukas + Max + Hjalti]
• [x] Look into other representations for shuffling other than gray code [Lukas]
• [x] Find out how to set weights, like normalization ? [Lukas]

[HjaltiThrastarson]

This issue tracks our overall progress. Tick stuff that has been done and add new jobs here as they come up. Possibly link to other issues containing more info on the subject.

• [x] Figure out how to implement shuffling with grey code with the Mga [Hjalti]
• [x] Set up a class skeleton with a random start and calculate fitness functionality
• [x] Create fitness function
• [x] Create an evaluate function that runs one iteration and shuffles with the mga[Hjalti]
• [ ] Create a vizualuzation for fitness output i.e [Max]
• [ ] Setup a library to have as a benchmark [Max]
  • [x] Evaluate existing libraries
  • [ ] Setup MWE
• [ ] LONG TERM Look into how to create a pareto front fitness function [Lukas + Max + Hjalti]
• [x] Look into other representations for shuffling other than gray code [Lukas]
• [x] Find out how to set weights, like normalization ? [Lukas]

[HjaltiThrastarson]

This issue tracks our overall progress. Tick stuff that has been done and add new jobs here as they come up. Possibly link to other issues containing more info on the subject.

Hjalti

• [x] Fix the kinks in the mga optimizer, such as constraints on the fitness function
• [ ] Implement a pareto frontier search algorithm that banks the points on the frontier
  • [x] Check into randomly initilazing weigths in the utility (fitness) function each random restart to get a whole idea of the frontier
  • [x] There is a mathematical proof (linear scalarization problem) that any weights will guarantee that any minima will land on the pareto frontier

Max

• [x] Create a vizualuzation for fitness output i.e
• [x] Setup a library to have as a benchmark
  • [x] Setup MWE
• [x] Continue looking into benchmark (nsga-ii, pymoo library) etc.

Lukas

• [ ] Figure out how to get the output to be compatible with pymoo vizulaization
• [ ] Create a presentation slide skeleton and start on intro/motivation, creating a timeline etc.

Team milestones

• [ ] LONG TERM Look into how to create a pareto front fitness function
• [ ] Figure out scalar/binary benchmark/comparison to compare different pareto front
• [x] After a working prototype, send an email to michal/smajic and ask if we are going in the right direction

[HjaltiThrastarson]

This issue tracks our overall progress. Tick stuff that has been done and add new jobs here as they come up. Possibly link to other issues containing more info on the subject.

• [x] Implement a function class that does constraint and function evaluation, so we don't have to do n + m functions

Hjalti

• [x] Implement ranges for each variable in the mga class, should be like np.array([-20, 20])
• [x] Looks into "intelligent" gray code shuffling, i.e try shuffling until it doesn't break constraints, and if that doesnt work generate new agent
• [x] Clean up the gray code part of the code, maybe put it into a new .py file or a class

Max

• [x] Create a function to compare two points and if one is dominating the other, return should be True/1 if dominating, False/0 if not
• [x] Find out how to create a ranking based on the dominating function, i.e how to sort agents with the domination function and therefore a fitness function

Lukas

• [ ] Figure out how to get the output to be compatible with pymoo vizulaization
• [ ] Create a presentation slide skeleton and start on intro/motivation, creating a timeline etc.
• [ ] Vizualize the the design space i.e constraint space with respect to optimal solutions
• [ ] Create a vizualazation of how the agents change over the iteration, can try to do a graphic with agents_history (check benchmark_bnh.py for how to do a video

Team milestones

• [ ] Add some nice documentation
• [ ] LONG TERM Look into how to create a pareto front fitness function
• [ ] Figure out scalar/binary benchmark/comparison to compare different pareto front

[HjaltiThrastarson]

This issue tracks our overall progress. Tick stuff that has been done and add new jobs here as they come up. Possibly link to other issues containing more info on the subject.

Hjalti

• [x] Add a selector class that basically does evaluate_agents and calc_fitness_func
• [x] Create tournament based selector class, bot using weight fitness and the non dominated ranking from ranking.py

Max

• [ ] Check out the code and confirm it works
• [ ] Implement more problems
• [ ] Go into git history and take Hjaltis old weight function and see how it changes the outcome
• [ ] Look into just using regular binary instead of gray code

Lukas

• [ ] Create some cool graphs for the presentation
• [ ] Write some text for presentation

Team milestones

• [ ] Add some nice documentation
• [x] LONG TERM Look into how to create a pareto front fitness function
• [x] Figure out scalar/binary benchmark/comparison to compare different pareto front