I trainned a computer to cross a mountain cliff by using q-learning. This was very exciting for me since this is my first working reinforcement learning project. Furtheremore, knowledge from this project will help me build my previous project successfully(hopefully).
Devlog 11
Hello, everyone!! I have finally made a computer learn to cross a mountain cliff!!!!!!
Log in to leave a comment
Devlog 10
Hello, everyone! I finally got the ai agent trained correcctly!!!! Even though the photo seems to indicate it’s doing horrible but, it’s just a small piece of the entire 10,000 episodes. Also it gets -1 for every step, -100 for falling into the cliff and 100 for reaching the end.
Log in to leave a comment
Devlog 01
Hello everyone! I finally made the enviroment for trainning the agent!
Log in to leave a comment
Devlog 04
Hello again, Everyone! So far, I have finally started training the agent! Initially, I wanted to make to 2 agents and have them learn ping pong on their own but, switched to training only one since it was getting too difficult. I am still having difficulties with even one agent. Luckily, I found out that I had made a tiny error in understanding RL, that the neural network should have output nodes equal to the number of actions possible to be taken in that enviroment. So hopefully, after fixing this issue, it will work properly. Also, I have partially written the code to test the model out after traning. (Total reward for agent 2 in the photo was something I forgot to delete so, it isnt anything useful)
Log in to leave a comment
Devlog 03
Hello Everyone! so far I have fixed bugs as always, and codded the “agents”. but, now I have a issue while updating their weights. I’ll have to look through the documentation or something for the solution. But, overalll almost all the structure for this project is done except the reward function.
Log in to leave a comment
Devlog 02
Hello everyone! It took me a while but, I finally made the enviroment for trainning the agent!!!
Even though it looks slow but, that is just because of timing in pygame for fps. In trainning it will be much faster.
Log in to leave a comment
Devlog 01
Today I didnt do much. Just made how the enviroment will look. This is important since from this Ican get the agents starting positions, balls position and border positions. Even though I wont make a gui game for now. Since I will make a function to do it internally without gui for training the agent. So it can be faster.
Log in to leave a comment
I made a terminal based program for communication between 2 computers!!
This project mainly uses the the socket library but it also uses threading and requests library.
I mainly learned how to send data from one computer to another and also taking input from the terminal. I wanted to make this for a long time. Anyway It was a very fun expirience building this.
Hello, Everyone! Today I have finally finished making this project. It honestly went smoother than I thought. I also figured out why it wasn’t printing the recived messages before it was because I forgot start the thread for them. Opsie. But anyways after getting that fixed I just tidied up the Code and added some some text for help with -h and I am finally done!
Log in to leave a comment
Hello again everyone! This time I was able to impliment most the code for sending data across 2 computers. I still though have a problem with sendng data. This is mostly a threading or blocking issue. But I will work on this next.
Log in to leave a comment
Hello, everybody! Today wrote the base code. I encountered a few problems while working with new libraries. Mainly with argparse due to me using a 6 year old tutorial. But I now have a good working code for accepting user arguments from command line.
Log in to leave a comment
