def test_MultipleHiddenLayers(data_reg):
X, Y = data_reg
elm = ELMRegressor(n_neurons=[2, 3],
ufunc=['tanh', 'sigm'],
density=[None, None],
pairwise_metric=[None, None])
elm.fit(X, Y)
assert len(elm.hidden_layers_) == 2
def test_MultipleHiddenLayers(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[2, 3],
ufunc=['tanh', 'sigm'],
density=[None, None],
pairwise_metric=[None, None])
elm.fit(X, Y)
self.assertEqual(len(elm.SLFNs_), 2)
def test_MultipleHiddenLayers_MoreCombinations(data_reg):
X, Y = data_reg
elm = ELMRegressor(n_neurons=[1, 1, 1, 1, 1],
ufunc=['relu', 'sigm', np.sin, None, None],
density=[None, 0.5, 0.8, None, None],
pairwise_metric=[None, None, None, 'l1', 'chebyshev'])
elm.fit(X, Y)
assert len(elm.hidden_layers_) == 5
def test_ELMClassifier_MultilabelClassification_Works(data_ml):
X, Y = data_ml
elm_c = ELMClassifier(random_state=0).fit(X, Y)
elm_r = ELMRegressor(random_state=0).fit(X, Y)
Yc_hat = elm_c.predict(X)
Yr_hat = (elm_r.predict(X) >= 0.5).astype(np.int)
assert Yc_hat == approx(Yr_hat)
def test_ELMClassifier_MultilabelClassification_Works(self):
X, Y = self.data_ml
elm_c = ELMClassifier(random_state=0).fit(X, Y)
elm_r = ELMRegressor(random_state=0).fit(X, Y)
Yc_hat = elm_c.predict(X)
Yr_hat = (elm_r.predict(X) >= 0.5).astype(int)
assert_allclose(Yc_hat, Yr_hat)
def test_MultipleHiddenLayers_MoreCombinations(self):
X, Y = self.data_reg
elm = ELMRegressor(
n_neurons=[1, 1, 1, 1, 1],
ufunc=['relu', 'sigm', np.sin, None, None],
density=[None, 0.5, 0.8, None, None],
pairwise_metric=[None, None, None, 'l1', 'chebyshev'])
elm.fit(X, Y)
self.assertEqual(len(elm.SLFNs_), 5)
def test_ELMClassifier_ReportedScore_ActuallyIsClassificationScore(data_class):
X, Y = data_class
Yr = np.vstack((Y == 0, Y == 1, Y == 2)).T
elm_c = ELMClassifier(random_state=0).fit(X, Y)
elm_r = ELMRegressor(random_state=0).fit(X, Yr)
Yc_hat = elm_c.predict(X)
Yr_hat = elm_r.predict(X).argmax(1)
assert Yc_hat == approx(Yr_hat)
def test_SineWave_Solves():
"""A highly non-linear regression problem, with added strong noise.
"""
X = np.linspace(-1, 1, num=1000)[:, None]
Y = np.sin(16 * X) * X + 0.2 * np.random.randn(1000)[:, None]
elm = ELMRegressor(random_state=0)
elm.fit(X, Y)
Yt = elm.predict(X)
MSE = np.mean((Y - Yt)**2)
assert MSE < 0.3
def test_Default_SetNumberOfNeurons(self):
X, y = self.data_reg
elm5 = ELMRegressor(n_neurons=5, random_state=0).fit(X, y)
elm50 = ELMRegressor(n_neurons=50, random_state=0).fit(X, y)
score5 = elm5.score(X, y)
score50 = elm50.score(X, y)
self.assertGreater(score50, score5)
self.assertGreater(score50, 0.33)
def test_Serialize_Solver(data_reg):
X, Y = data_reg
elm = ELMRegressor(random_state=0)
elm.fit(X, Y)
Yh1 = elm.predict(X)
solver_data = pickle.dumps(elm.solver_, pickle.HIGHEST_PROTOCOL)
del elm.solver_
elm.solver_ = pickle.loads(solver_data)
Yh2 = elm.predict(X)
assert Yh1 == pytest.approx(Yh2)
def test_MultipleHL_WrongDimensions_Raises(data_reg):
X, Y = data_reg
elm = ELMRegressor(n_neurons=[1, 2, 3, 4], ufunc=['relu', 'sigm'])
with pytest.raises(ValueError):
elm.fit(X, Y)
def test_Regressor_WrongNumberOfFeatures_RaisesError(self):
X, T = self.data_reg
elm = ELMRegressor()
elm.fit(X, T)
with self.assertRaises(ValueError):
elm.predict(X[:, 1:])
def test_MultipleHL_WrongDimensions_Raises(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[1, 2, 3, 4], ufunc=['relu', 'sigm'])
with self.assertRaises(ValueError):
elm.fit(X, Y)
def test_MultipleHL_Pairwise_SingleValue(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[2, 3], pairwise_metric='l2')
elm.fit(X, Y)
def test_MultipleHL_Density_SingleValue(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[2, 3], density=0.7)
elm.fit(X, Y)
def test_Regressor_WrongNumberOfFeatures_RaisesError(data_reg):
X, T = data_reg
elm = ELMRegressor().fit(X, T)
with pytest.raises(ValueError):
elm.predict(X[:, 1:])
""" =========================== Plotting Template Estimator =========================== An example plot of :class:`skelm.template.TemplateEstimator` """ import numpy as np from matplotlib import pyplot as plt from skelm import ELMRegressor X = np.arange(100).reshape(100, 1) y = np.zeros((100, )) estimator = ELMRegressor() estimator.fit(X, y) plt.plot(estimator.predict(X)) plt.show()
def test_RegularizationL2_DifferentValue_ChangesPrediction(self):
X, Y = self.data_reg
Yh_1 = ELMRegressor(alpha=1e-7, random_state=0).fit(X, Y).predict(X)
Yh_2 = ELMRegressor(alpha=1e+3, random_state=0).fit(X, Y).predict(X)
self.assertFalse(np.allclose(Yh_1, Yh_2))
def test_MultipleHL_DefaultValues(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[2, 3])
elm.fit(X, Y)
def test_MultipleHL_Ufunc_SingleValue(self):
X, Y = self.data_reg
elm = ELMRegressor(n_neurons=[2, 3], ufunc='sigm')
elm.fit(X, Y)
def test_RegularizationL2_DifferentValue_ChangesPrediction(data_reg):
X, Y = data_reg
Yh_1 = ELMRegressor(alpha=1e-7, random_state=0).fit(X, Y).predict(X)
Yh_2 = ELMRegressor(alpha=1e+3, random_state=0).fit(X, Y).predict(X)
assert Yh_1 != approx(Yh_2)
def test_Regression_Boston_BetterThanNaive(data_reg):
elm = ELMRegressor(random_state=0)
elm.fit(*data_reg)
r2score = elm.score(*data_reg)
assert r2score > 0.3