๐Ÿ–ฅ๏ธ POSTMORTEM FOR MY FIRST MACHINE LEARNING GAME! (4/4)

36 thoughts on “๐Ÿ–ฅ๏ธ POSTMORTEM FOR MY FIRST MACHINE LEARNING GAME! (4/4)”

  1. i love this topic of machine learning so fucking much because it touches on so many fundamental topics of humanity. You have to understand biological evolution, genetics, education and learning, psychology, statistics, data analysis, programming, and probably more, but i can't think of any right now.

  2. I think that the corner issue can be solved if you add the probability of going back… It may take more time but eventually it will be able to tackle both circles and corners…

  3. I believe forrest needs the same eyes as the human who plays, instead of those 5 lines, so that he can have a general overview of the map. Thanks for sharing your projects, that's awesome.

  4. I'm retroprogramming in BASIC developing an arkanoid clone…luckyly I still don't need machine learning…not yet! ๐Ÿ™‚

  5. The hex shapes constantly increase and decrease the distance to the walls next to him if running horizontally and even more confusing when running vertically because the distance to one wall gets large while the distance to another wall becomes small and the other way around.

  6. Have you considered that perhaps your 5 inputs don't provide enough information to complete the task. Would you be able to create a rules-based runner that only worked with those 5 inputs? Your bird's eye view provides you with data that "Forest" did not have.

  7. Next time you should use a simulation approach that is step wise basically give the neuronal network data each step wait for its output and calc the new step this way you don't need to have a graphical simulation running and it doesn't matter if the CPU is at 0 or 100%, also this way you can speed up the training more.

  8. Help me understand Machine Learning, please.
    Is it better than this my simple code?
    if it is better, why?
    if not, why suffer yourself training Forrest for hours?
    While(sensingTheEnvironment){
    runStraightForrest();
    recordTrack();
    If(aboutToTouchWall){
    Deviate();
    }
    if(trackIsFamiliar){
    dontTakeThatRouteAgain();
    }
    if(forrestIsGoingWrongWay){
    rotateForrest();
    }
    }
    I'm sure my Forrest equiped with this code will never crash into any wall, or what do you think?

  9. Probably spacing the rays of distance evenly every 45 degrees isn't enough/the best option. Try spacing them unevenly but symmetrically, e.g. 0, 25, 90, 155, 180 degrees. Also increasing the number of rays might help a lot. Good video series nevertheless. Kudos

  10. So I guess having forest adjust his vector in the direction with the longest ray-trace defeats the purpose of this exercise? Great series. Thank you for this!

  11. In many different NN setups; specifically more linear ones, a common theme with regards to learning is the fact that optimizations are made for general solutions to generally similar situations. Meaning: trying to make a NN learn how to deal with polar situations usually ends with less than lackluster results. Training for one thing can make you a master at specifically that thing. Training for another thing with a similarly structured dataset can make you above average at completing both. This pattern of a downward curve of percentage success over the number of things you are training for, is probably one of the main reasons animals are designed to be able to remember things, and create schemas based on those memories of different situations. This should ultimately achieve similar results to combining multiple simple NNs designed for only one situation in order to act as one NN.

  12. Hey! Wonderful content, really! On the topic of machine learning games, do you think the amiibo training in Smash uses real machine learning or it's just fake? I guess it could be that they made 50 AI with increasing skill levels and you just have to put time into the game to unlock a stronger AI, also because I'm not sure wether the time you spend leveling up the amiibo is really sufficient to train a neural network. I've been learning ANN for only a few days now so I figured I'd ask to someone who's more knowledgeable, maybe you can see something I can't!

  13. I know this is old, but I'd strongly suspect that there is no thought process of an allowed design that can handle every possible valid course. For the hex map design, particularly when allowing for both large open areas and freestanding columns, only the five wall-distance inputs just might not be enough of an input. It feels like it wouldn't be too hard to deliberately design courses that would break any otherwise successful thought processes. Beyond that, the measure of success was poorly chosen. Wall distance aids in measuring survivability, but survivability doesn't measure course progress when Forrest can freely run backwards and run infinite loops. As for the lap multiplier, it takes too long for it to trigger, particularly for larger maps. You really need some way to reward getting closer to the goal, either by implementing a GPS-route-style distance to the goal or by having the map generator automatically generate multiple checkpoints that can be used to mark progress.

  14. I've watched it all and i'm wondering if a simple rule could have get your buddy through those mazes. To complete a tour, your guy have to keep a wall on his left or on his right ( it doesn't matter wich one, it just have to be one or the other, but always the same one : always right or always left ). It will not get a fast way to complete a lap, but a logical and systematical way to complete it. You should have selected the ones that keep an even average value to the right or left wall ( not too close, not too far ). It might have solved any course if no lag or others issues. The logic is that a lap is in a way a circle, and to complet it, you have to turn the same side at every step. No matter how complex the maze is, from start to arrival, you always have the inside wall or the outside wall, makinge a loop on itself, otherwise, the maze won't return to his depart point ๐Ÿ˜‰ .

    I didn't read all the comments to see if anyone already had proposed this solution. Let me know if I'm the first one to have this idea ๐Ÿ˜‰ .

    Great job, and great fun while watching !

    ps : excuse my english, I'm french ๐Ÿ˜‰ .

  15. Couldn't Forest just run along the right side of any course and stick to that with a certain space and therefore never fail?

  16. First time I've seen you talk while moving your lips. Impressed. But real talk, your channel is amazing, and has inspired me to really start looking into machine learning. Keep up the great work!

  17. I just wanted to say how much I enjoyed following this AI project you put together! I'm thinking about learning python so I can start a similar process for mastering games by emulating AI that has solved it – like 80's retrocade games. This was really educational and I appreciate all the resources you included – especially being a visual learner! I subscribed and hope you wanna do more ML / AI game oriented videos in the future! Also, is that Eco Virtual you ended the video with?

  18. Couldn't he increase reward the closer he moves toward the other side of the course finish line along with what he currently put into place?

  19. You need to add a gradient for each of forrests "feelers" so that it becomes more negative to be too close to the wall in any one direction. Maybe use a logarithm. It's also better to use a predictive time series so you would actually have forrest running a frame or number of frames behind where he would run, and then anticipate where to go based on values and use a second, deeper network, in order to make decision trees based on which training sets to use in the next steps. And there are multiple ways to do this so you might want to look at making a neural network that tries them all randomly until the best combinations are found based on the same recursion. Then just add a lap marker at the starting point and make forrest drop markers for where he has already been and train him only on not hitting the walls and covering the least amount of distance to complete the tracks, and causing the first lap markers to be ignored at a certain point before the second lap or erase them up to a distance behind him, before they are in his field of vision, which is probably better except for in a few cases, which is a harder problem to case for randomly generating courses, which is another algorithm that is inversely co-variant with the difficulty in solving them, in some aspects. Being able to calculate that in the same means is of value as well, and you can see why now things are boring, and why it is of no value to explore things that don't make my possible as an independent conservative.

    Of course doing this in ways that affects people is a whole other domain and is in the realm of managing probability and risk. Since you wanted to get political and this is a matter of science involving politics and what we're not the ones doing..

  20. Overfitting was the first thing that came to my mind. When you trained, would you train on one course for a long time, and then move him on to the next one? Maybe it would be better to change courses after each weight update (multiple batches for an epoch). That way, weight updates that help him learn general rules would still be rewarded, but weight updates for memorizing the specific training course would be punished.

  21. Jabrils is there any chance that you'll notice my comment? How did you start on ml? You said that in 6 months. What did you do or learn in 6 months? Do you have suggestion or tips. Thankyou Jabrils, hoping that you soon build jarvis๐Ÿ˜‚

  22. I'm impressed by your videos. I am new to ML and you had shown me some interesting stuff to do. I am able to understand what's going on. Please share the code for the project. That would be really helpful.

  23. I was thinking that adjusting those input lines would make a big difference. I'd start by making them longer to "see" farther down the course. Part of Forrest's problem is he's a little near-sighted.

    And would it be cheating if Forrest was able to leave a "bread crumb" trail behind himself to identify where he's already been? Train him to not follow them unless it's the only way forward, and train him that if he does follow them, he changes direction if they continue for too long (ie- he's going the wrong way).

  24. When Forrest failed on level 4, it looked like he ran directly into the wall at a 90 degree angle. In every other level, he was moving around corners, so walls would always approach on one side and Forrest would know to turn the OTHER direction. If the wall approaches from a head-on angle, Forrest would not know which way to turn since the wall would activate his front-left and front-right equally. I believe including a similar scenario in your training would have corrected this, as you alluded to in the video, but I'm still pretty new to this topic. Love this content, you inject a lot of fun and personality into this topic! thanks for your work

  25. as not being any kind of pro or amything but from my own research and reading what others find out. neural networks work best with more and u made a very small brain so it couldnt work well, as there were very few options to choose from.

    nature has infinite possibilties which is why it has a 100% success rate so the closer to infinity the better of yr self working machine will be given it wont be perfect unless its allowed ininite ideas.

Leave a Reply

Your email address will not be published. Required fields are marked *