AlanMet / TicTacToe

0 stars 0 forks source link

README please? #1

Open manighahrmani opened 2 months ago

manighahrmani commented 2 months ago

Hi Allan. Can you please add a README to this project and describe how I can recreate your issue to suggest a fix?

AlanMet commented 2 months ago

yeah, what do you mean? because this was just an issue visible by reading the debug prints. I've since updated the backpropagation in the Networks.dart but still no luck at getting it to work.

manighahrmani commented 2 months ago

OK describe to me what u see in the debug prints (tell me step-by-step instructions on how I can reproduce it). If you are serious and want to learn AI, I would ask you first to complete this tutorial for me (it takes at most a couple of hours). We can revisit this repository afterwards.

AlanMet commented 2 months ago

ok will do. I've been studying it at a lower level for a couple months, watching videos from 3b1b and others. I've been wanting to implement a neural net from scratch for a while. Nad also told me to create an app for the coming first years and I overcomplicated it. I'll complete the tutorial as well.

AlanMet commented 2 months ago

I've read through it but all of this is for python. I've also been trying to figure out why neural networks are built the way they are rather than simply accepting information as being true. Maybe I don't understand it fully but why can't you just take the derivative of the cost function with respect to its weights and solve for when it = 0? wouldn't that give you the most accurate value for the updated weight? Or is there some issue with it overfitting a single set of data as a result?

manighahrmani commented 2 months ago

Regarding this:

ok will do. I've been studying it at a lower level for a couple months, watching videos from 3b1b and others. I've been wanting to implement a neural net from scratch for a while. Nad also told me to create an app for the coming first years and I overcomplicated it. I'll complete the tutorial as well.

It's great that you made a Flutter app but that is not the tool for this purpose (you need to reinvent the wheel as opposed to using pre-existing Python libraries. If you want I have several ideas for Flutter apps here which you are free to pick?

Regarding this:

I've read through it but all of this is for python. I've also been trying to figure out why neural networks are built the way they are rather than simply accepting information as being true. Maybe I don't understand it fully but why can't you just take the derivative of the cost function with respect to its weights and solve for when it = 0? wouldn't that give you the most accurate value for the updated weight? Or is there some issue with it overfitting a single set of data as a result?

You are hinting at the answer with overfitting. Essentially, the process that you are describing will find the best solution that satisfies a single instance or a type of instance. This means that if this approach was feasible (which generally is not, due to how resource-demanding it is on larger networks) it could only work on a few numbers of instances and is not suitable for real-world applications. Frankly, it kind of defeats the purpose of having it as an AI (you might as well write a hard-coded function that works on specified input).

What course do you do? I'd like you to take the AI module in your final year and do the AI-900 course (extra-curricular activity) with me in your second year.

AlanMet commented 2 months ago

I take computer science, yeah an AI course seems fun. And of course dart really isn't the tool to be doing this but Nadim set me a challenge to create a tictactoe algorithm and I've been wanting to learn the fundamentals of neural networks in practice so I included it. This is also the reason I'm not doing it in python even though it does have numpy I'm enjoying building everything from the ground up. Looking at the list you sent I've actually already built a few discord bots in python. Is that something the university server actually needs? because a quiz app could be built in a few hours.

manighahrmani commented 2 months ago

OK fair enough, I just wanted to make sure you get hands-on experience with real-life tools used in the industry (so you can add them to your CV for jobs or future degrees).

Regarding Discord, I do have some specific requirements for the quiz bot. If you are happy to collaborate on it and the product meets our specifications, we can use it in the next academic year. This is also something impressive that you can add to your CV/portfolio in the future.

AlanMet commented 1 month ago

well I've built a moderation bot in python which was paired with a html website and a SQL database as well as the paypal API in php. But the development of the python API has been turbulent. I had to swap to a fork because the creator got annoyed at the discord dev's direction. I had to then swap back when he came back and then discord stopped supporting prefix commands, opting for the more secure slash commands. so it does make it a bit difficult to get back into it but I can try.

AlanMet commented 1 month ago

In fact I have one of my old discord bots here: https://github.com/MysteryCrystal/Mayumi-Bot. It was meant to be collaborative and there is some data there that should be kept private such as the token or database sign ins but as everything is offline it shouldn't be an issue.

manighahrmani commented 1 month ago

well I've built a moderation bot in python which was paired with a html website and a SQL database as well as the paypal API in php. But the development of the python API has been turbulent. I had to swap to a fork because the creator got annoyed at the discord dev's direction. I had to then swap back when he came back and then discord stopped supporting prefix commands, opting for the more secure slash commands. so it does make it a bit difficult to get back into it but I can try.

I need to know about your career goals. If it wasn't related I won't bother you but it would be nice if you could help develop a quiz bot at first or maybe a role one. The moderation one can be more time-consuming because you'd have to interview Matt and Rich to see exactly what they want.

In fact I have one of my old discord bots here: https://github.com/MysteryCrystal/Mayumi-Bot. It was meant to be collaborative and there is some data there that should be kept private such as the token or database sign ins but as everything is offline it shouldn't be an issue.

Am I safe to assume that this bot essentially helps with moderation by providing commands but not scanning/flagging messages? My current students managed to hook it up to Azure or GCP AI resources so we get these scores to flag automatically.

manighahrmani commented 1 month ago

Can you message me on Discord pls? It's easier than here.