def test_returning_value(): func = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), -1 * np.linspace(1, 10, 11), 'test_1') assert func.get_value(10.45) == -10.0 assert func.get_value(0.45) == -1.0 assert func.get_value(2.0) == -2.8
def test_func_add_fails_with_different_features(): func1 = ShapeFunction(np.linspace(1, 3, 3), np.linspace(1, 3, 3), 'test_1') func2 = ShapeFunction(np.linspace(1.5, 10.5, 5), -1 * np.linspace(1, 10, 5), 'test_2') assert func1 != func2 with pytest.raises(AssertionError): func1.add(func2)
def test_func_add_with_equal_splits(): func1 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), np.linspace(1, 10, 11), 'test_1') func2 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), -1 * np.linspace(1, 10, 11), 'test_1') func3 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), np.zeros(11), 'test_1') assert func1.add(func2).equals(func3)
def test_func_add_2(): func1 = ShapeFunction([np.PINF], [0], 'test_1') func2 = ShapeFunction([0, 1, 2], [-1, 1, -2], 'test_1') assert func1 != func2 func3 = ShapeFunction([0.0, 1.0, 2.0, np.PINF], [-1.0, 1.0, -2.0, 0.0], 'test_1') assert func1.add(func2).equals(func3)
def test_func_add(): func1 = ShapeFunction(np.linspace(1, 3, 3), np.linspace(1, 3, 3), 'test_1') func2 = ShapeFunction(np.linspace(1.5, 10.5, 5), -1 * np.linspace(1, 10, 5), 'test_1') assert func1 != func2 func3 = ShapeFunction([1.0, 1.5, 2.0, 3.0, 3.75, 6.0, 8.25, 10.5], [1.0, -1.0, 2.0, 3.0, -3.25, -5.5, -7.75, -10.0], 'test_1') assert func1.add(func2).equals(func3)
def test_correct_scoring(): func1 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), np.linspace(1, 10, 11), 'attr_1') func2 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), -1 * np.linspace(1, 10, 11), 'attr_2') gam = GAM(max_depth=3, max_leaf_nodes=5, random_state=42) gam.shapes = {'attr_1': func1, 'attr_2': func2} gam.feature_names = ['attr_1', 'attr_2'] data = np.asarray([[1, 1], [-1, -1]]) for vec in data: # the shape function cancel each other in their effect # hence the exponent is zero and the probability 0.5 assert gam.score(vec) == [0.5, 0.5]
def _create_shape_functions(self, data): shapes = dict() for feat in self.feature_names: dim_idx = self._get_index_for_feature(feat) splits = np.unique(data[:, dim_idx].flatten()).tolist() vals = self._get_shape(dim_idx, splits) shapes[feat] = ShapeFunction(splits, vals, feat) self.shapes = shapes
def test_pseudo_response(): func1 = ShapeFunction(_create_partition([0]), [np.log(3), np.log(100)], 'attr_1') gam = GAM(max_depth=3, max_leaf_nodes=5, random_state=42) gam.shapes = {'attr_1': func1} gam.feature_names = ['attr_1'] data = np.asarray([[-1], [1]]) pseudo_resp = gam._get_pseudo_responses(data, [1, -1]) np.testing.assert_almost_equal(pseudo_resp, [0.2, -1.9998], decimal=6, verbose=True, err_msg="Pseudo Response doesn't match")
def test_correct_shape_addition(): func1 = ShapeFunction(_create_partition(np.linspace(1, 10, 10)), np.linspace(1, 10, 11), 'attr_1') func2 = func1.add(func1).multiply(0.5) assert func1.equals(func2)
def test_func_multiply(): func1 = ShapeFunction(np.linspace(1, 3, 3), np.linspace(1, 3, 3), 'test_1') func2 = ShapeFunction(np.linspace(1, 3, 3), 0.5 * np.linspace(1, 3, 3), 'test_2') assert func1.multiply(0.5).equals(func2)