Question - Lua and torch

[nightsta69]

With Lua being a programming language in mq now, does this mean that you could include torch and Theoretically add machine learning to mqnext?


Has anyone thought about doing such a thing?

sorry if this seems like a dumb question, but this sorta stuff interests me.

 

[Knightly]

Torch Lua development hasn’t been active for a few years, but you can use Lua libraries.

 

[dannuic]

I mean, you don't generally do machine learning in realtime. You'd probably write a script/plugin to dump data and then run whatever you are trying to find out in asynchronous jobs, where it won't need to integrate with the EQ client at all.

 

[tweeb]

You could always do this in a plugin. I'm having a hard time imagining a situation where you would want to, though.

 

[nightsta69]

bazaar market prediction/trending items
auto reply/learning how people talk and predicting automated responses
mana efficiency/clerics learn mob dps, group "tankability", and predict which heals are better suited for the immediate situation
mana efficiency/DPS casters, ability to better make use of your damage spells, or conserve mana.
better pathing/ more human like reaction to situations (IE instead of /stick !front at 98% health on a mob, one character can learn that it can start attacking earlier on certain mobs because the tank can hold agro)
better pathing/ machine learning could help predict better pathing for MQ2Nav, even i have problems with it getting stuck on a log in the middle of no where or something with a weird hit box every now and then.


again I don't know how feasible any of these ideas are. I was just curious if that was something that could be done. Appreciate the Responses.

 

[dannuic]

In all of those circumstances, you would usually collect data, train an algorithm, and then apply it at runtime outside of the ML framework. ML isn't really a real-time apparatus

 

[Tone]

In all of those circumstances, you would usually collect data, train an algorithm, and then apply it at runtime outside of the ML framework. ML isn't really a real-time apparatus

I agree with dannuic 100% you'd want to train a model, and then once it is trained you could MAYBE get a model producing outputs with real time data.

The best example of using ML real time w/ video games I got for you is take an example like a Fortnite aimbot. You train a modelthat predicts the probability that the pixels on the screen = an enemy player.
That thing has to be tuned / updated for new models / retrained to avoid drift... You need a large data set of still image screenshots to train this thing to get the accuracy up.

Once that model exist, you basically write code that is taking screenshots of your screen every x seconds / milliseconds, and looking for the area with the highest probability of an enemy player.
If the probability returned is > your threshold mouse move to the x,y coords and click mouse.

I did ^^ project in Python awhile ago just to see if it was possible and gave me a decent use case to mess with ML.