on call function about reservoirpy

[Finebouche]

I understand that I am not using the library correctly.

I would like to :

• instantiate a reservoir with a defined internal matrix and a defined input matrix,
• feed the reservoir with input step by step (I thought call() was doing exactly that),
• observe the internal states
• and generate the output Wout matrix based on the internal states

However it doesn't seem possible to feed the the reservoir and get the internal states without setting a readout first ?

[nTrouvain]

Hello,
Your usage of node call seems good. What went wrong is the dimension of you input data. For some reason, X_train1[0] has a shape of (1, 1) (one timestep, 1 variable) but your matrix Win is expecting an input of shape (1, 4) (one timestep, 4 variables). You also probably let the input_bias parameter to True in your reservoir, which means that the row of your custom Win matrix will be considered as a bias vector. (this is a bit odd and will be changed in a future version), hence a Win matrix of dimension (5, 1). This also means that your reservoir is of size 1, which is probably nt what you want to do. Could you give a more complete code example ?

[Finebouche]

Thanks, I resolve the problem.
It was also linked to this bias parameter that I didn't get right.