def fit_steepestgd(self, X, y, iters=1000, verbose=False):
self.coefs = np.ones(X.shape[1])
cur_loss = self.mse_loss(y, self.predict(X))
i = 0
sk = 1
while i < iters and np.linalg.norm(sk) > 10**(-6):
func_for_opt = self.cost(X, y)
ad_obj = AD(func_for_opt, multivar=True)
if verbose:
print("=====\nIter {} Loss: {}".format(i, cur_loss))
print(self.coefs)
grads = ad_obj.get_der_numpy(self.coefs)
sk = -1 * grads
etak = line_search(ad_obj.get_val_numpy, ad_obj.get_der_numpy,
self.coefs, sk)[0]
self.coefs = self.coefs + etak * sk
cur_loss = self.mse_loss(y, self.predict(X))
i += 1
def test_numpy_wrapper():
functionxx_class = AD(functionxx)
val = functionxx_class.get_val_numpy(np.array([1, 2]))
der = functionxx_class.get_der_numpy(np.array([1, 2]))
assert list(val) == [2, 4]
assert list(der) == [2, 2]