def test_func_w_mult_params_single_var():
fn = lambda x, y: x**2
ad_fn = AD(fn)
der = ad_fn.get_der([1, 2])
val = ad_fn.get_val([1, 2])
assert der == [2, 0]
assert val == 1
def test_mul_array():
ad_mul_array = AD(mul_array, ndim=3)
val = ad_mul_array.get_val([[1, 2], [1, 2], [1, 2]])
der = ad_mul_array.get_der([[1, 2], [1, 2], [1, 2]])
assert val == [[4, 3, 2], [4, 3, 2], [4, 3, 2]]
assert der == [[[8, 4], [1, 1], [2, 0]], [[8, 4], [1, 1], [2, 0]],
[[8, 4], [1, 1], [2, 0]]]
def test_nested_1():
nested_fn = AD(my_fn_nested_1)
xs = [-10, 2, 5, 10]
for x in xs:
assert nested_fn.get_der(x) == pytest.approx(
-10 * x * (x * np.sin(x) - 2 * np.cos(x)))
assert nested_fn.get_val(x) == pytest.approx(5 * x**2 * 2 * np.cos(x))
def test_square_fn():
ad_square = AD(square_fn)
der = ad_square.get_der(3)
val = ad_square.get_val(3)
assert der == 6
assert val == 9
der1 = ad_square.get_der([1, 2, 3, 4])
val1 = ad_square.get_val([1, 2, 3, 4])
assert der1 == [2, 4, 6, 8]
assert val1 == [1, 4, 9, 16]
def fit_gd(self, X, y, iters=1000, verbose=False):
self.coefs = np.random.rand(X.shape[1])
i = 0
while i < iters:
func_for_opt = self.cost(X, y)
ad_obj = AD(func_for_opt, multivar=True)
if verbose:
print("=====\nIter {}".format(i))
print(self.coefs)
grads = ad_obj.get_der(list(self.coefs))
for idx, grad in enumerate(grads):
if verbose:
print('coef{}: {}'.format(idx, self.coefs[idx]))
self.coefs[idx] = self.coefs[idx] - self.lr * grad
i += 1
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_exception_der():
fn = AD(my_fn_2d, ndim=2)
with pytest.raises(Exception):
fn.get_der([1, 2, 3])
def test_get_val_lenlist():
a = AD(lambda x, y: 3 * x**2 + 2 * y**3)
with pytest.raises(Exception):
a.get_val(1, 2, 3)
with pytest.raises(Exception):
a.get_val([1, 2, 3], [3, 4, 5], [1, 3, 4])
def test_get_val_types():
with pytest.raises(TypeError):
AD.get_val('string')
with pytest.raises(TypeError):
AD.get_val(dict[1:'a', 2:'b'])
def test_2d_fn():
fn = AD(my_fn_2d, ndim=2)
der = fn.get_der([1, 2])
val = fn.get_val([1, 2])
assert der == [[2, 4], [1, 1]]
assert val == [5, 5]
def test_list_lists():
fn = AD(my_fn_1d)
der = fn.get_der([[1, 2], [3, 4], [5, 6]])
val = fn.get_val([[1, 2], [3, 4], [5, 6]])
assert der == [[2, 4], [6, 8], [10, 12]], [[1, 1], [1, 1], [1, 1]]
assert val == [5, 25, 61]
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]
def test_cos():
cos_fn = AD(my_fn_cos)
assert cos_fn.get_der(5) == -5 * np.sin(5)
assert cos_fn.get_val(5) == 5 * np.cos(5)
def test_list():
fn = AD(my_fn_1d)
der = fn.get_der([1, 2])
val = fn.get_val([1, 2])
assert der == [2, 4]
assert val == 5