How can I use wash_nr_time_step correctly in generative task?

[mustafakucuk0]

Hello. I want to use this library to compare my from scratch code's results but when I try to use your washout parameter, I come across with this error. I know the reason behind this error but could not find the solution without changing the actual source code. Am I missing something about this repo?

Thank you for this wonderful repo.

[mustafakucuk0]

Here is my data and training pipeline:

[neuronalX]

Hi @mustafakucuk0, thank you for your question.
I assume you are using the v0.2 version, right?

Could you give us an example of a simple training with a reservoir (without going through the hyperparameter search)?

[mustafakucuk0]

In this case, yes, I was using 0.2. I modified your hyperopt tutorial code to generate Lorenz63 time series. With 6000 steps given, I wanted to model to create the last 2000 steps and optimize the model throughout the parameter pool.

The main aim of this study was to compare my from-scratch code results and this optimized library's results. But in your code, firstly, I have to give X_train and y_train to the model but I designed my RC code to take only one training array for training phase and one test array for test phase.

I could not understand why we have to give y_train again? FYI, I am not 100% sure about my code. As you can see in the image, I start to collect states after number of transient steps.

But for example, in your code, I try to discard transient but I come across this error:

It's about ridge regression part but I do not know how to deal with it. I hope I could tell my story.