import torch
y = torch.tensor(data[4 * sr : 4 * sr + 1000])
x = torch.linspace(0.0, 1.0, steps=len(y))

w = torch.rand(100, requires_grad=True)
phi = torch.rand(100, requires_grad=True)

for t in range(200):
  y_pred = sum(w[i] * torch.sin((x + phi[i]) * i) for i in range(len(w)))

  loss = torch.mean((y_pred - y) ** 2)
  loss.backward()

  w.data = w.data - 0.05 * w.grad.data
  phi.data = phi.data - 0.05 * phi.grad.data

  w.grad.data.zero_()
  phi.grad.data.zero_()
  if t % 10 == 0:
    plt.plot(y_pred.data.numpy())
    plt.plot(y.numpy())
    plt.show()