Out-of-sample Prediction with ReservoirPy about reservoirpy HOT 4 CLOSED

Hello,

The answer of your question relies on the training procedure of your RC network, and on the nature of your task.

1. Your RC network is trained to predict $X_{t+1}$ given $X[t]$. You can then compute a full month of predictions by entering a closed loop generative mode:

y = last_prediction
for i in range(one_month):
    y = reservoir(y)

2. Your RC network is trained to predict something else, like $X[t+n]$ given $X[t]$. If it happens that $n$ is a one month step, then you can just use run() using the $X$ data you have for the previous month to predict the next.

Did that answer your question ?

from reservoirpy.

So, regarding the second question, if one-month data consist of 750 steps and I train my model to predict X[t+750] given X[t] and X[-750:] is December, the output would be January?

from reservoirpy.

Well if you trained your model that way, then this is the expected behavior. Predicting n+750 given n is probably not that trivial though. If your timeseries is very chaotic, obtaining good results will be challenging.

from reservoirpy.

Sorry for the late reply. Thank you for the information you provided. I understand clearly now.

from reservoirpy.