I'm working on a hobby project that involves simulating emotional states for NPCs in a text-based RPG. I have a continuous 3D emotional vector represented as `E=(V,A,S)`, where both `V` and `S` range from -1 to 1, and `A` ranges from 0 to 1. My goal is to map this continuous vector onto 20 different emotional labels, such as Anger, Disgust, and Love. I've identified some reference points for these emotions: Anger at (-0.7,1.0,+0.7), Disgust at (-0.5,0.7,-0.9), and Love at (+1.0,0.6,+1.0). While I'm currently using a simple IF/ELSE approach to classify these emotions, it feels cumbersome. I'm looking for suggestions on more efficient and reliable classification methods, like whether K-Nearest Neighbors (KNN) could work better than a Radial Basis Function (RBF) Network. If KNN is the path to take, which distance metric would be most effective for handling the approach/avoid social dimensions of these emotions?
3 Answers
You might want to experiment with different methods and see how they perform. The IF/ELSE approach is definitely the cheapest, but it can become complex when dealing with multiple classes. The issue with using such a method is that it relies heavily on hand-tuning many boundaries, which can lead to inconsistencies between the emotional classifications. You should consider a classifier that can handle transitions smoothly across all 20 emotions. KNN could be a solid choice to help create this continuous mapping across your 3D space. For the distance metric, Euclidean distance should be a good starting point, but keep an eye out for performance as you develop further!
I'm a little confused; why are we discussing fishing techniques when it's about classifying emotional vectors? Can someone clarify the relevance here?
Yeah, I have no idea how that came up either. Seems a bit off-topic from the main discussion!
Just checked out your Github, and I'm really impressed with what you've got going on! The documentation is super polished, and it's clear a lot of thought went into your project. I absolutely love your idea! I'll dive deeper later, but I'm excited to see how it progresses. Keep it up!
Thanks a lot! I’m really excited about the potential of this project. The way each NPC interacts based on emotional responses is going to change how the game feels. Can’t wait for more feedback!
That definitely sounds reasonable! I'm also intrigued by how KNN adjusts to new data points and can provide a solid smooth mapping for you. Definitely worth testing.