Multi-Objective Optimisation - Writing your own Genetic Algorithm Part 6 

I'm glad I could help!

Thanks I really appreciate that! I don’t make any money off this channel, but if you think others in your class might benefit from the content id appreciate a recommendation :)

I’m glad you found it helpful!

That’s awesome. If you haven’t watched the earlier videos in the series I would strongly recommend it. In the next video I’m going through the implementation of multi-objective for our TSP. You can grab the code early here: github.com/Gary-The-Cat/MultiObjective it’s well documented, but I will go over all the changes in my next video.

Thanks!

I’m glad it helped!

Thanks!

Thanks! I’m really glad you liked it

Hey thanks for the feedback! I used after effects, is a very slow process

Thanks! Painstakingly slowly in after effects 😂

Thank you! I don’t have any other courses sorry, but it’s something I’d love to do.

I can’t say I have, I have used GA’s for a lot of random problems, but mostly in experimentation.

Thanks, I used AfterEffects for all animation

@@TutorialsWithGary thanks

I don’t know if your problem sounds like a good fit for a GA, but I can provide information anyway :) If your two functions truly were (positively) correlated, then summing them and using a single objective should yield good results (if a GA is an appropriate option for this problem). Implementing it as a multi objective problem should likewise improve the result over time.

Hey, Imagine if you have three solutions A = (1,5) B = (5,1) C = (2,2) By your logic, C would be the best solution, however it is worse in the first objective than solution A, and is worse in the second objective than solution B. Additionally say you only had A and B, your logic would not be able to tell them apart. The purpose of the Pareto front is a way for us to trade objectives against one another - specifically when we have more than one objective because direct comparisons are not possible without giving up some information.

@@TutorialsWithGary Alright it makes sense, thank you !

Please keep making videos Gary!

Thank you, I’m glad you liked it. I’ve been struggling to find time to make videos recently but I really want to make more

Hey Roy, I'm glad you're getting into Genetic Algorithms. In my next video in this series I go through the code required for a C# implementation of this

Yes, i'm using the code provided there but can't find the restrictions code in it. Like how do you decide that a solution is infeasible?

The way the spawning, crossover and mutation operator have been designed, every solution is always feasible. There are ways to generate these operators in a way that they produce infeasible solutions, where you can then ‘repair’ them, but if I was designing it, I would prefer to modify my operators to guarantee the results generated are feasible

Thank you very much!!😄

I’m glad it was helpful for you :) Are you learning about GA’s for work of fun?

Actually i'm phd student, and my topic has relation with ga, and i need help, if you could to help me?

No worries, if you are interested in a chat, you can always join my discord server :)

That is part of the beauty of having each objective in its own axis, the scale of different objectives will not interact with one another, always providing the best trade off between objectives.

Hey, in a later video I added in the concept of a camera that allows you to zoom out & use any screen size. It would be worth grabbing that code

Thank you for replying quickly... The tutorial is super amazing for new comers like me. However, I could not find out how to zoom since the screen freezes. Is there a way, I can run it and see output on console or in form without seeing those graphical computations? that could remove the barriers for those who wants to experiment it without concept of gaming

Certainly! The next video in the series has code attached in the description.

That’s great, thank you

Hey! I have been working hard on a new project that I will be using to continue the series (It's taking longer than expected.. what doesn't haha). The same techniques we have used here can be used for more than 2 objectives, however, they start to become slow and less meaningful must past 3. There is another approach for that (If you're interested in reading, give 'NSGA III' a google). If you go to three objectives, all of the same logic works, however instead of being a line, the Pareto front will be a 3D surface. Selection is similarly scaled, however, it sounds like you really might be better suited to something like NSGA III. The selection and ranking approach is unfortunately entirely different, but the reasoning logic is all the same as we've covered in these videos. I'm really keen to make a series on NSGA III, but I don't think it will be ready by the time you need it. I've read the original NSGA III paper though and they walk through all of the techniques and wouldn't be too hard to follow and implement. I hope this has helped!

You’re in luck, this series implements NSGA II 😀

@@TutorialsWithGary that's great Help me a lot with my thesis

@@TutorialsWithGary but actually I'm working on three objectives problems so I dont know how to rank them when they have three obj func

The ranking process is the same as the method explained in the last video in this series. The same logic applies to three objectives.