Feature Importance about reservoirpy

[PAUL-BERNARD]

Hello 👋
We are missing information here. What's your task, what kind of variable are you talking about ? What do you mean by "best way" ? What do you mean by important ?

If you want to optimize the input scaling of each feature, you can start a hyper-parameter search on the scaling of all your input features.
You can specify the input scaling of each feature like so:
input_scaling = [feature1_scaling, feature2_scaling, ...]
And then you can use an hyper-parameter optimization algorithm on the (feature1_scaling, feature2_scaling, ...) tuple.

Let us know if that helps, or if you have a different problem.

[FuzzyWuzzyIsABear]

What I mean is that how do I get a rank list of the most and least influential variables of a model? Say you input 10 variables I wanted to know which ones were more important than others.

[FuzzyWuzzyIsABear]

Sorry accidentally closed the issue.

[PAUL-BERNARD]

Sorry I had an issue opened but I accidentally closed it. So with many other AI techniques like transformers you have the ability to see which variables contribute the most to the ai model results. I was wondering if reservoirpy had a way of finding out how influential each input variable is.

this website below explains it more.

Feature importance

The feature permutation method proposed in the webpage you linked should work with an echo state network, as with any model.

Beyond this and optimizing your input scaling for each feature as hyper-parameters, there have been some research on the echo state network specifically, such as ESNigma, which are things I would like to have in ReservoirPy, but for now, you would have to re-implement the algorithm by hand.