search

SuperFluffy / rust-recless

Recursive least squares algorithm in Rust
Apache License 2.0
2 stars 0 forks source link

dumb question: how to use the model parameters #3

Closed jonathanstrong closed 4 years ago

jonathanstrong commented 4 years ago

this might be a dumb question, but would it be possible to explain how to utilize the model's parameters in the documentation? what I mean is, what operations are needed to perform filtering / predicting on an input row? I'm getting useful results from doing model.weight_ref().dot(&xs) where xs is a ArrayBase<S, Ix1>, but am not sure how gain comes into it. also - on the data I tested it on, something like 1e-3 for initialization_factor and 1.0 for forgetting_factor is working best. what are very broad guidelines for what to set those variables to?

thank you!

SuperFluffy commented 4 years ago

Heya! I think your best chance is to jump to Haykin's Adaptive Filter Theory (see the link in the README) and go from there. If I recall, I have pretty much used their terminology. I have written this library to implement the FORCE algorithm for learning in a chaotic neural network, but I haven't really pursued it since then, so I don't have a good feeling for recursive least squares to give you suggestions.