def test_piecewise_regressor_issue(self):
X, y = make_regression(10000, n_features=1, n_informative=1, # pylint: disable=W0632
n_targets=1)
y = y.reshape((-1, 1))
model = PiecewiseRegressor(
binner=DecisionTreeRegressor(min_samples_leaf=300))
model.fit(X, y)
vvc = model.predict(X)
self.assertEqual(vvc.shape, (X.shape[0], ))
def test_piecewise_regressor_pandas(self):
X = pandas.DataFrame(numpy.array([[0.1, 0.2], [0.2, 0.3]]))
Y = numpy.array([1., 1.1])
clr = LinearRegression(fit_intercept=False)
clr.fit(X, Y)
clq = PiecewiseRegressor()
clq.fit(X, Y)
pred1 = clr.predict(X)
pred2 = clq.predict(X)
self.assertEqual(pred1, pred2)
def test_piecewise_regressor_intercept_weights3(self):
X = numpy.array([[0.1, 0.2], [0.2, 0.3], [0.3, 0.3]])
Y = numpy.array([1., 1.1, 1.2])
W = numpy.array([1., 1., 1.])
clr = LinearRegression(fit_intercept=True)
clr.fit(X, Y, W)
clq = PiecewiseRegressor(verbose=False)
clq.fit(X, Y, W)
pred1 = clr.predict(X)
pred2 = clq.predict(X)
self.assertNotEqual(pred2.min(), pred2.max())
self.assertEqual(pred1, pred2)
def test_piecewise_regressor_diff(self):
X = numpy.array([[0.1], [0.2], [0.3], [0.4], [0.5]])
Y = numpy.array([1., 1.1, 1.2, 10, 1.4])
clr = LinearRegression()
clr.fit(X, Y)
clq = PiecewiseRegressor(verbose=False)
clq.fit(X, Y)
pred1 = clr.predict(X)
self.assertNotEmpty(pred1)
pred2 = clq.predict(X)
self.assertEqual(len(clq.estimators_), 2)
p1 = clq.estimators_[0].predict(X[:3, :])
p2 = clq.estimators_[1].predict(X[3:, :])
self.assertEqual(pred2[:3], p1)
self.assertEqual(pred2[-2:], p2)
sc = clq.score(X, Y)
self.assertEqual(sc, 1)
def test_piecewise_regressor_no_intercept_bins(self):
X = numpy.array([[0.1, 0.2], [0.2, 0.3], [0.2, 0.35], [0.2, 0.36]])
Y = numpy.array([1., 1.1, 1.15, 1.2])
clr = LinearRegression(fit_intercept=False)
clr.fit(X, Y)
clq = PiecewiseRegressor(binner="bins")
clq.fit(X, Y)
pred1 = clr.predict(X)
pred2 = clq.predict(X)
self.assertEqual(pred1.shape, (4, ))
self.assertEqual(pred2.shape, (4, ))
sc1 = clr.score(X, Y)
sc2 = clq.score(X, Y)
self.assertIsInstance(sc1, float)
self.assertIsInstance(sc2, float)
paths = clq.binner_.transform(X)
self.assertIn(paths.shape, ((4, 7), (4, 8), (4, 9), (4, 10)))
self.assertNotEqual(pred2.min(), pred2.max())
def test_piecewise_regressor_no_intercept(self):
X = numpy.array([[0.1, 0.2], [0.2, 0.3], [0.2, 0.35], [0.2, 0.36]])
Y = numpy.array([1., 1.1, 1.15, 1.2])
clr = LinearRegression(fit_intercept=False)
clr.fit(X, Y)
clq = PiecewiseRegressor()
clq.fit(X, Y)
pred1 = clr.predict(X)
pred2 = clq.predict(X)
self.assertEqual(pred1.shape, (4, ))
self.assertEqual(pred2.shape, (4, ))
sc1 = clr.score(X, Y)
sc2 = clq.score(X, Y)
sc3 = clq.binner_.score(X, Y)
self.assertIsInstance(sc1, float)
self.assertIsInstance(sc2, float)
self.assertIsInstance(sc3, float)
paths = clq.binner_.decision_path(X)
s = paths.sum()
self.assertEqual(s, 8)
self.assertNotEqual(pred2.min(), pred2.max())
self.assertGreater(clq.n_estimators_, 1)
def test_piecewise_regressor_list(self):
X = [[0.1, 0.2], [0.2, 0.3]]
Y = numpy.array([1., 1.1])
clq = PiecewiseRegressor()
self.assertRaise(lambda: clq.fit(X, Y), TypeError)
def test_piecewise_regressor_raise(self):
X, y = make_regression(10000, n_features=2, n_informative=2, # pylint: disable=W0632
n_targets=2)
model = PiecewiseRegressor(
binner=DecisionTreeRegressor(min_samples_leaf=300))
self.assertRaise(lambda: model.fit(X, y), RuntimeError)
verbose=True)
else:
model = DecisionTreeRegressor(splitter="best",
max_leaf_nodes=40,
criterion="friedman_mse")
# model = LinearRegression()
dataset_features = [f for f in features if f in data.columns]
input_features = dataset_features + ["IThreshold", "CThreshold"
] + c + parameters
x = data[input_features]
y = data[measures[0]]
d = x.copy()
X = squish(d)
pred = model.fit(np.asarray(X), transform(y))
print()
print(method)
y_pred = untransform(pred.predict(X.values))
score = (((y - y_pred)**2).sum()**0.5 / len(y_pred))
if score < best_model.get(method, (None, 1e12))[1]:
best_model[method] = pred, score
else:
pred, _ = best_model[method]
y_pred = untransform(pred.predict(X.values))
print(((y - y_pred)**2).sum()**0.5 / len(y_pred))
print(np.quantile(np.abs(y - y_pred), 0.5))
print(np.quantile(np.abs(y - y_pred), 0.67))
print(np.quantile(np.abs(y - y_pred), 0.9))
print(np.quantile(np.abs(y - y_pred), 0.95))