Where does the randomness come from

[zonglunli7515]

Hi,

I think I fixed the input and reservoir weights but I still observed the difference in the internal state after running the following code multiple times. Where does the randomness come from? Thanks in advance.

from reservoirpy.nodes import Reservoir
import numpy as np
import matplotlib.pyplot as plt

def normal_w(n, m, **kwargs):
np.random.seed(42)
W_res = np.random.normal(0, 1, size=(n, m))
rhoW = max(abs(np.linalg.eig(W_res)[0]))
return rhoW*W_res/rhoW

def normal_win(n, m, **kwargs):
np.random.seed(43)
W_in = np.random.normal(0, 1, size=(n, m))
return W_in

X = np.sin(np.linspace(0, 6*np.pi, 100)).reshape(-1, 1)
reservoir = Reservoir(50, lr=1, Win=normal_win, W=normal_w)
states = reservoir.run(X)
plt.plot(states[:,1])
reservoir = reservoir.reset()

[PAUL-BERNARD]

Hi @zonglunli7515 ,

The randomness in your reservoir state comes from the input bias of the reservoir. If you want a deterministic behavior, you can set a value for bias, disable input bias by setting input_bias=False, or set a seed when creating your Reservoir.

Let me know if that helps

[zonglunli7515]

Thanks a lot! I overconcentrated on noise parameters and overlooked this!