def constraint_pen_mat_2d(self, c1="smooth", c2="smooth", y=None):
if y is None: y = self.y
descr_str = (
("s(1)", c1, 25, (1, 100), "equidistant"),
("s(2)", c2, 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
M.calc_LS_fit(X=x, y=y)
M.create_constraint_penalty_matrix(beta_test=M.coef_)
self.assertEqual(M.constraint_penalty_matrix.shape,
(2 * descr_str[0][2], 2 * descr_str[1][2]))
del M
class TestStarModel(unittest.TestCase):
def setUp(self):
self.x = np.linspace(0, 1, 100).reshape(100, 1)
self.y = np.exp(-(self.x - 0.4)**2 / 0.01)
self.n_params = 25
self.descr_str = (("s(1)", "smooth", self.n_params, (1, 100),
"equidistant"), )
self.M = StarModel(description=self.descr_str)
def tearDown(self):
del self.M
def test_starmodel_create_basis(self):
self.M.create_basis(X=self.x)
self.assertEqual(len(self.M.smooths), 1)
self.assertEqual(self.M.basis.shape, (len(self.x), self.n_params))
self.assertEqual(self.M.smoothness_penalty_matrix.shape,
(self.n_params, self.n_params))
def test_starmodel_create_constraint_penalty_matrix(self):
self.M.create_basis(X=self.x, y=self.y)
self.M.create_constraint_penalty_matrix(
beta_test=np.zeros(self.n_params))
self.assertEqual(self.M.constraint_penalty_matrix.shape,
(self.n_params, self.n_params))
def test_starmodel_create_constraint_penalty_matrix_2d(self):
self.constraint_pen_mat_2d(c1="smooth", c2="inc")
self.constraint_pen_mat_2d(c1="smooth", c2="dec")
self.constraint_pen_mat_2d(c1="peak", c2="smooth")
self.constraint_pen_mat_2d(c1="valley", c2="smooth", y=-1 * self.y)
self.constraint_pen_mat_2d(c1="inc", c2="dec")
self.constraint_pen_mat_2d(c1="peak", c2="inc")
self.constraint_pen_mat_2d(c1="valley", c2="inc", y=-1 * self.y)
self.constraint_pen_mat_2d(c1="peak", c2="dec")
self.constraint_pen_mat_2d(c1="valley", c2="dec", y=-1 * self.y)
def constraint_pen_mat_2d(self, c1="smooth", c2="smooth", y=None):
if y is None: y = self.y
descr_str = (
("s(1)", c1, 25, (1, 100), "equidistant"),
("s(2)", c2, 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
M.calc_LS_fit(X=x, y=y)
M.create_constraint_penalty_matrix(beta_test=M.coef_)
self.assertEqual(M.constraint_penalty_matrix.shape,
(2 * descr_str[0][2], 2 * descr_str[1][2]))
del M
def test_starmodel_calc_LS_fit(self):
self.M = self.M.calc_LS_fit(X=self.x, y=self.y)
self.assertEqual(len(self.M.coef_), self.M.basis.shape[1])
self.assertEqual(len(self.M.LS_coef_), self.M.basis.shape[1])
def test_starmodel_calc_LS_fit_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
y = np.sin(x[:, 0]) + x[:, 0] * x[:, 1]
M = StarModel(description=descr_str)
M.calc_LS_fit(X=x, y=y)
self.assertEqual(len(M.coef_), M.basis.shape[1])
self.assertEqual(len(M.LS_coef_), M.basis.shape[1])
self.assertEqual(len(M.coef_), 50)
self.assertEqual(len(M.LS_coef_), 50)
def test_starmodel_create_df_for_beta(self):
beta_test = np.random.random(self.n_params)
df = self.M.create_df_for_beta(beta_init=beta_test)
self.assertEqual(df.shape, (1, self.n_params))
self.assertTrue(np.allclose(df.values, beta_test))
def test_starmodel_fit(self):
self.M = self.M.fit(X=self.x, y=self.y, plot_=False)
self.assertEqual(len(self.M.coef_), self.M.basis.shape[1])
def test_starmodel_fit_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
("t(1,2)", "smooth", (5, 5), (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
y = np.sin(x[:, 0]) + x[:, 0] * x[:, 1]
M = StarModel(description=descr_str)
M = M.fit(X=x, y=y, plot_=False)
self.assertEqual(len(M.coef_), M.basis.shape[1])
def test_starmodel_predict_single_point(self):
self.descr_str = (("s(1)", "peak", self.n_params + 10, (1, 100),
"equidistant"), )
M = StarModel(description=self.descr_str)
M.fit(X=self.x, y=self.y, plot_=False)
y_pred = M.predict_single_point(x_sp=self.x[10])
self.assertEqual(type(y_pred), np.float64)
def test_starmodel_predict(self):
self.descr_str = (("s(1)", "peak", self.n_params + 10, (1, 100),
"equidistant"), )
M = StarModel(description=self.descr_str)
M.fit(X=self.x, y=self.y, plot_=False)
x_pred = np.linspace(0, 1, 7)
y_pred = M.predict(X_pred=x_pred)
self.assertEqual(type(y_pred), np.ndarray)
self.assertEqual(y_pred.shape[0], x_pred.shape[0])
def test_starmodel_calc_hat_matrix(self):
self.M.fit(X=self.x, y=self.y, plot_=False)
H = self.M.calc_hat_matrix()
self.assertEqual(H.shape, (self.x.shape[0], self.x.shape[0]))
self.assertTrue(np.allclose(H, H.T))
def test_starmodel_calc_hat_matrix_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
y = np.sin(x[:, 0]) + x[:, 0] * x[:, 1]
M = StarModel(description=descr_str)
M = M.fit(X=x, y=y, plot_=False)
H = M.calc_hat_matrix()
self.assertEqual(H.shape, (x.shape[0], x.shape[0]))
self.assertTrue(np.allclose(H, H.T))
def test_starmodel_calc_GCV_score(self):
self.M.fit(X=self.x, y=self.y, plot_=False)
gcv = self.M.calc_GCV_score(y=self.y)
self.assertEqual(type(gcv), np.float64)
def test_starmodel_calc_GCV_score_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
y = np.sin(x[:, 0]) + x[:, 0] * x[:, 1]
M = StarModel(description=descr_str)
M = M.fit(X=x, y=y, plot_=False)
gcv = M.calc_GCV_score(y=y)
self.assertEqual(type(gcv), np.float64)
def test_starmodel_generate_GCV_parameter_list(self):
n_grid = 5
p_min = 1e-4
grid = self.M.generate_GCV_parameter_list(n_grid=n_grid, p_min=p_min)
i = 0
c = np.inf
for _, g in enumerate(grid):
i += 1
c1, c2 = g["s(1)_constraint"], g["s(1)_smoothness"]
cc = np.min([c1, c2])
c = np.min([c, cc])
self.assertEqual(i, n_grid**2)
self.assertEqual(c, p_min)
def test_starmodel_generate_GCV_parameter_list_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
n_grid, p_min = 5, 1e-4
grid = M.generate_GCV_parameter_list(n_grid=n_grid, p_min=p_min)
i, c = 0, np.inf
for _, g in enumerate(grid):
i += 1
c1, c2 = g["s(1)_constraint"], g["s(1)_smoothness"]
cc = np.min([c1, c2])
c = np.min([c, cc])
self.assertEqual(i, n_grid**4)
self.assertEqual(c, p_min)
def test_starmodel_get_params(self):
d = self.M.get_params()
self.assertTrue(type(d) == dict)
self.assertTrue(type(d["s(1)"]) == dict)
self.assertTrue(len(d["s(1)"].keys()) == 4)
def test_starmodel_get_params_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
d = M.get_params()
self.assertTrue(type(d) == dict)
self.assertTrue(len(d.keys()) == 2)
self.assertTrue(type(d["s(1)"]) == dict)
self.assertTrue(len(d["s(1)"].keys()) == 4)
def test_starmodel_set_params(self):
self.M.create_basis(X=self.x)
d = self.M.get_params()
d["s(1)"]["constraint"] = "CHANGED"
d["s(1)"]["knot_type"] = "CHANGED"
d["s(1)"]["lam"] = {"smoothness": "CHANGED", "constraint": "CHANGED"}
d["s(1)"]["n_param"] = "CHANGED"
self.M.set_params(params=d)
self.assertEqual(self.M.description_dict["s(1)"]["constraint"],
"CHANGED")
self.assertEqual(self.M.description_dict["s(1)"]["knot_type"],
"CHANGED")
self.assertEqual(self.M.description_dict["s(1)"]["lam"], {
"smoothness": "CHANGED",
"constraint": "CHANGED"
})
self.assertEqual(self.M.description_dict["s(1)"]["n_param"], "CHANGED")
def test_starmodel_set_params_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
("t(1,2)", "smooth", (5, 5), (1, 100), "equidistant"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
M.create_basis(X=x)
d = M.get_params()
d["s(1)"]["constraint"] = "CHANGED"
d["s(1)"]["knot_type"] = "CHANGED"
d["s(1)"]["lam"] = {"smoothness": "CHANGED", "constraint": "CHANGED"}
d["s(1)"]["n_param"] = "CHANGED"
d["s(2)"]["constraint"] = "CHANGED"
d["s(2)"]["knot_type"] = "CHANGED"
d["s(2)"]["lam"] = {"smoothness": "CHANGED", "constraint": "CHANGED"}
d["s(2)"]["n_param"] = "CHANGED"
d["t(1,2)"]["constraint"] = "CHANGED"
d["t(1,2)"]["knot_type"] = "CHANGED"
d["t(1,2)"]["lam"] = {"smoothness": "CHANGED", "constraint": "CHANGED"}
d["t(1,2)"]["n_param"] = "CHANGED"
M.set_params(params=d)
self.assertEqual(M.description_dict["s(1)"]["constraint"], "CHANGED")
self.assertEqual(M.description_dict["s(1)"]["knot_type"], "CHANGED")
self.assertEqual(M.description_dict["s(1)"]["lam"], {
"smoothness": "CHANGED",
"constraint": "CHANGED"
})
self.assertEqual(M.description_dict["s(1)"]["n_param"], "CHANGED")
self.assertEqual(M.description_dict["s(2)"]["constraint"], "CHANGED")
self.assertEqual(M.description_dict["s(2)"]["knot_type"], "CHANGED")
self.assertEqual(M.description_dict["s(2)"]["lam"], {
"smoothness": "CHANGED",
"constraint": "CHANGED"
})
self.assertEqual(M.description_dict["s(2)"]["n_param"], "CHANGED")
self.assertEqual(M.description_dict["t(1,2)"]["constraint"], "CHANGED")
self.assertEqual(M.description_dict["t(1,2)"]["knot_type"], "CHANGED")
self.assertEqual(M.description_dict["t(1,2)"]["lam"], {
"smoothness": "CHANGED",
"constraint": "CHANGED"
})
self.assertEqual(M.description_dict["t(1,2)"]["n_param"], "CHANGED")
def test_starmodel_set_params_after_gcv(self):
d = dict()
d["s(1)_constraint"] = "CHANGED"
d["s(1)_smoothness"] = "CHANGED"
self.M.set_params_after_gcv(params=d)
self.assertEqual(self.M.description_dict["s(1)"]["lam"]["smoothness"],
"CHANGED")
self.assertEqual(self.M.description_dict["s(1)"]["lam"]["constraint"],
"CHANGED")
def test_starmodel_set_params_after_gcv_2d(self):
descr_str = (
("s(1)", "smooth", 25, (1, 100), "equidistant"),
("s(2)", "smooth", 25, (1, 100), "quantile"),
("t(1,2)", "smooth", (5, 5), (1, 100), "quantile"),
)
x = np.random.random(200).reshape(100, 2)
x.sort(axis=0)
M = StarModel(description=descr_str)
d = dict()
d["s(1)_constraint"] = "CHANGED"
d["s(1)_smoothness"] = "CHANGED"
d["s(2)_constraint"] = "CHANGED"
d["s(2)_smoothness"] = "CHANGED"
d["t(1,2)_constraint"] = "CHANGED"
d["t(1,2)_smoothness"] = "CHANGED"
M.set_params_after_gcv(params=d)
self.assertEqual(M.description_dict["s(1)"]["lam"]["smoothness"],
"CHANGED")
self.assertEqual(M.description_dict["s(1)"]["lam"]["constraint"],
"CHANGED")
self.assertEqual(M.description_dict["s(2)"]["lam"]["smoothness"],
"CHANGED")
self.assertEqual(M.description_dict["s(2)"]["lam"]["constraint"],
"CHANGED")
self.assertEqual(M.description_dict["t(1,2)"]["lam"]["smoothness"],
"CHANGED")
self.assertEqual(M.description_dict["t(1,2)"]["lam"]["constraint"],
"CHANGED")