# inputting LEAK = 0.9 ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss0 = np.mean((ys - ys_target)**2) print('loss before training', loss0) # training LA = 0.01 esn.update(ys_target.reshape(-1, 1), la=LA) # inputting again ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss1 = np.mean((ys - ys_target)**2) print('loss after training', loss1) # plot plt.plot(ys, label='prediction', color='r', marker='o') plt.plot(ys_target, label='target', color='k', marker='o')
# inputting LEAK = 0.9 ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss0 = np.mean((ys[DELAY:] - ys_target[DELAY:])**2) print('loss before training', loss0) # training LA = 0.01 esn.update(ys_target[DELAY:].reshape(-1, 1), t_start_at=DELAY, la=LA) # inputting again ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss1 = np.mean((ys[DELAY:] - ys_target[DELAY:])**2) print('loss after training', loss1) # plot plt.plot(ys[DELAY:], label='prediction', color='r', marker='o') plt.plot(ys_target[DELAY:], label='target', color='k', marker='o')
# inputting LEAK = 0.9 ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss0 = np.mean((ys[BIT - 1:] - ys_target[BIT - 1:])**2) print('loss before training', loss0) # training LA = 0.01 esn.update(ys_target[BIT - 1:].reshape(-1, 1), t_start_at=BIT - 1, la=LA) # inputting again ys = [] for u in us: y = esn(u.reshape(1, 1), leak=LEAK) ys.append(y[0][0]) # compute loss ys = np.asarray(ys) loss1 = np.mean((ys[BIT - 1:] - ys_target[BIT - 1:])**2) print('loss after training', loss1) # plot plt.plot(ys[BIT - 1:], label='prediction', color='r', marker='o') plt.plot(ys_target[BIT - 1:], label='target', color='k', marker='o')