def test_get_set(self):
fit_obj = ns.BayesianRVFLRegressor(
n_hidden_features=10,
direct_link=False,
bias=False,
nodes_sim="sobol",
type_scaling=("std", "minmax", "std"),
activation_name="relu",
n_clusters=0,
)
fit_obj2 = ns.BayesianRVFL2Regressor(
n_hidden_features=9,
direct_link=False,
bias=True,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj.set_params(
n_hidden_features=5,
activation_name="relu",
a=0.01,
nodes_sim="sobol",
bias=True,
direct_link=True,
n_clusters=None,
type_clust="kmeans",
type_scaling=("std", "minmax", "std"),
seed=123,
s=0.1,
sigma=0.05,
#return_std=True,
)
fit_obj2.set_params(
n_hidden_features=4,
activation_name="relu",
a=0.01,
nodes_sim="sobol",
bias=True,
dropout=0.5,
direct_link=True,
n_clusters=None, # optim
type_clust="kmeans",
type_scaling=("std", "std", "minmax"),
seed=123,
s1=0.1,
s2=0.1,
sigma=0.05, # optim
#return_std=True,
)
def test_MTS(self):
np.random.seed(123)
X = np.random.rand(25, 3)
X[:, 0] = 100 * X[:, 0]
X[:, 2] = 25 * X[:, 2]
regr = linear_model.BayesianRidge()
regr2 = gaussian_process.GaussianProcessRegressor()
regr3 = ns.BayesianRVFLRegressor(s=0.01)
regr4 = ns.BayesianRVFL2Regressor(s1=0.01,
s2=0.01,
n_hidden_features=2)
fit_obj = ns.MTS(
regr,
n_hidden_features=10,
direct_link=False,
bias=False,
nodes_sim="sobol",
type_scaling=("std", "minmax", "std"),
activation_name="relu",
n_clusters=0,
)
fit_obj2 = ns.MTS(
regr,
n_hidden_features=9,
direct_link=False,
bias=True,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj3 = ns.MTS(
regr,
n_hidden_features=8,
direct_link=True,
bias=False,
nodes_sim="uniform",
type_scaling=("minmax", "minmax", "std"),
activation_name="tanh",
n_clusters=3,
)
fit_obj4 = ns.MTS(
regr,
n_hidden_features=7,
direct_link=True,
bias=True,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj5 = ns.MTS(
regr,
n_hidden_features=7,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj6 = ns.MTS(
regr2,
n_hidden_features=2,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj7 = ns.MTS(
regr3,
n_hidden_features=2,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj8 = ns.MTS(
regr4,
n_hidden_features=0,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj9 = ns.MTS(
regr4,
n_hidden_features=0,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
cluster_encode=False,
)
fit_obj10 = ns.MTS(
regr4,
n_hidden_features=0,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
cluster_encode=False,
)
index_train = range(20)
index_test = range(20, 25)
X_train = X[index_train, :]
X_test = X[index_test, :]
Xreg_train = np.reshape(range(0, 60), (20, 3))
Xreg_test = np.reshape(range(60, 75), (5, 3))
fit_obj.fit(X=X_train)
err = fit_obj.predict() - X_test
rmse = np.sqrt(np.mean(err**2))
fit_obj.fit(X_train, xreg=Xreg_train)
err_xreg = fit_obj.predict(new_xreg=Xreg_test) - X_test
rmse_xreg = np.sqrt(np.mean(err_xreg**2))
fit_obj2.fit(X_train)
err2 = fit_obj2.predict() - X_test
rmse2 = np.sqrt(np.mean(err2**2))
fit_obj3.fit(X_train)
err3 = fit_obj3.predict() - X_test
rmse3 = np.sqrt(np.mean(err3**2))
fit_obj4.fit(X_train)
err4 = fit_obj4.predict() - X_test
rmse4 = np.sqrt(np.mean(err4**2))
fit_obj5.fit(X_train)
err5 = fit_obj5.predict() - X_test
rmse5 = np.sqrt(np.mean(err5**2))
fit_obj6.fit(X_train)
preds = fit_obj6.predict(return_std=True)
fit_obj7.fit(X_train)
preds2 = fit_obj7.predict(return_std=True)
fit_obj8.fit(X_train)
preds3 = fit_obj8.predict(return_std=True)
fit_obj9.fit(X_train)
preds4 = fit_obj9.predict(return_std=True)
fit_obj9.fit(X_train[:, 0])
preds5 = fit_obj9.predict(return_std=True)
fit_obj10.fit(X_train)
preds6 = fit_obj10.predict(return_std=True)
self.assertTrue(np.allclose(np.round(rmse, 3), 10.396))
self.assertTrue(np.allclose(rmse_xreg, 10.39593866961476))
self.assertTrue(np.allclose(np.round(rmse2, 3), 10.395))
self.assertTrue(np.allclose(rmse3, 10.395986434438191))
self.assertTrue(np.allclose(rmse4, 10.677585029352571))
self.assertTrue(np.allclose(np.round(rmse5, 3), 10.360, atol=1e-3))
def test_rvfl(self):
np.random.seed(123)
X, y = datasets.make_regression(n_samples=25, n_features=3)
fit_obj = ns.BayesianRVFLRegressor(
n_hidden_features=10,
direct_link=False,
bias=False,
nodes_sim="sobol",
type_scaling=("std", "minmax", "std"),
activation_name="relu",
n_clusters=0,
)
fit_obj2 = ns.BayesianRVFL2Regressor(
n_hidden_features=9,
direct_link=False,
bias=True,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj3 = ns.BayesianRVFLRegressor(
n_hidden_features=8,
direct_link=True,
bias=False,
nodes_sim="uniform",
type_scaling=("minmax", "minmax", "std"),
activation_name="tanh",
n_clusters=3,
)
fit_obj4 = ns.BayesianRVFL2Regressor(
n_hidden_features=7,
direct_link=True,
bias=True,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
)
fit_obj5 = ns.BayesianRVFL2Regressor(
n_hidden_features=7,
direct_link=True,
bias=True,
nodes_sim="hammersley",
type_scaling=("minmax", "minmax", "minmax"),
activation_name="elu",
n_clusters=4,
cluster_encode=True,
)
index_train = range(20)
index_test = range(20, 25)
X_train = X[index_train, :]
y_train = y[index_train]
X_test = X[index_test, :]
y_test = y[index_test]
fit_obj.fit(X_train, y_train)
err = fit_obj.predict(X_test, return_std=True)[0] - y_test
rmse = np.sqrt(np.mean(err**2))
fit_obj2.fit(X_train, y_train)
err2 = fit_obj2.predict(X_test, return_std=True)[0] - y_test
rmse2 = np.sqrt(np.mean(err2**2))
fit_obj3.fit(X_train, y_train)
err3 = fit_obj3.predict(X_test, return_std=True)[0] - y_test
rmse3 = np.sqrt(np.mean(err3**2))
fit_obj4.fit(X_train, y_train)
err4 = fit_obj4.predict(X_test, return_std=True)[0] - y_test
rmse4 = np.sqrt(np.mean(err4**2))
fit_obj5.fit(X_train, y_train)
err5 = fit_obj5.predict(X_test, return_std=True)[0] - y_test
rmse5 = np.sqrt(np.mean(err5**2))
pred1 = fit_obj.predict(X_test[0, :], return_std=True)[0]
pred2 = fit_obj2.predict(X_test[0, :], return_std=True)[0]
pred3 = fit_obj.predict(X_test[0, :], return_std=False)
pred4 = fit_obj4.predict(X_test[0, :], return_std=False)
self.assertTrue(
np.allclose(rmse, 0.81893186154747988)
& np.allclose(rmse2, 17.278090150613096)
& np.allclose(rmse3, 32.582378601766486)
& np.allclose(rmse4, 53.052553013478331)
& np.allclose(rmse5, 53.052553013478295)
& np.allclose(pred1, 325.96545701774187)
& np.allclose(pred2, 299.56243221494879)
& np.allclose(pred3, 325.96545701774187)
& np.allclose(pred4, 234.76784640807193))
def test_score(self):
np.random.seed(123)
X, y = datasets.make_regression(n_samples=100, n_features=3)
fit_obj = ns.BayesianRVFLRegressor(
n_hidden_features=10,
direct_link=False,
bias=False,
nodes_sim="sobol",
type_scaling=("std", "minmax", "std"),
activation_name="relu",
n_clusters=0,
)
fit_obj2 = ns.BayesianRVFL2Regressor(
n_hidden_features=9,
direct_link=False,
bias=True,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj3 = ns.BayesianRVFL2Regressor(
n_hidden_features=9,
direct_link=True,
bias=True,
dropout=0.3,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj4 = ns.BayesianRVFL2Regressor(
n_hidden_features=9,
direct_link=True,
bias=True,
dropout=0.5,
nodes_sim="halton",
type_scaling=("std", "minmax", "minmax"),
activation_name="sigmoid",
n_clusters=2,
)
fit_obj.fit(X, y)
fit_obj2.fit(X, y)
fit_obj3.fit(X, y)
fit_obj4.fit(X, y)
#fit_obj4.set_params(return_std=True)
self.assertTrue(
np.allclose(
fit_obj.score(X, y, scoring="neg_mean_squared_error"),
0.023104115093245361,
)
& np.allclose(fit_obj.score(X, y), 0.023104115093245361)
& np.allclose(
fit_obj2.score(X, y, scoring="neg_mean_squared_error"),
51.485414634058536,
)
& np.allclose(fit_obj2.score(X, y), 51.485414634058536)
& np.allclose(fit_obj3.score(X, y), 0.20023262498412012)
& np.allclose(fit_obj4.score(X, y), 0.17517631177933579))