New approximation methods: KNeighbors and RadiusNeighbors

.travis.yml

@@ -46,13 +46,13 @@ install:
         conda create --yes -n test python="3.6";
       fi
     - source activate test
-    - pip install numpy scipy matplotlib pip nose sphinx==1.4 gpy torch
+    - pip install numpy scipy matplotlib pip nose sphinx==1.4 gpy torch sklearn
     - pip install setuptools
     - pip install coveralls
     - pip install coverage
     - python setup.py install
 
-script: 
+script:
     - coverage run test.py
 
 after_success:

README.md

@@ -47,7 +47,7 @@
 See the [**Examples**](#examples) section below and the [**Tutorials**](tutorials/README.md) to have an idea of the potential of this package.
 
 ## Dependencies and installation
-**EZyRB** requires `numpy`, `scipy`, `matplotlib`, `vtk`, `nose` (for local
+**EZyRB** requires `numpy`, `scipy`, `sklearn`, `matplotlib`, `vtk`, `nose` (for local
 test) and `sphinx` (to generate the documentation).The coe has been tested with
 Python3.5 version, but it should be compatible with Python3. It can be
 installed using `pip` or directly from the source code.

ezyrb/__init__.py

@@ -1,7 +1,8 @@
 __all__ = [
     'database',
     'reduction', 'pod',
-    'approximation', 'rbf', 'linear', 'gpr', 'ann'
+    'approximation', 'rbf', 'linear', 'gpr', 'ann',
+    'kneighbors_regressor', 'radius_neighbors_regressor'
 ]
 
 from .meta import *
@@ -14,3 +15,5 @@
 from .gpr import GPR
 from .reducedordermodel import ReducedOrderModel
 from .ann import ANN
+from .kneighbors_regressor import KNeighborsRegressor
+from .radius_neighbors_regressor import RadiusNeighborsRegressor

ezyrb/kneighbors_regressor.py

@@ -0,0 +1,20 @@
+"""
+Wrapper for K-Neighbors Regressor.
+"""
+
+import numpy as np
+from sklearn.neighbors import KNeighborsRegressor as Regressor
+
+from .neighbors_regressor import NeighborsRegressor
+
+
+class KNeighborsRegressor(NeighborsRegressor):
+    """
+    K-Neighbors Regressor.
+
+    :param kwargs: arguments passed to the internal instance of
+        KNeighborsRegressor.
+    """
+
+    def __init__(self, **kwargs):
+        self.regressor = Regressor(**kwargs)

ezyrb/neighbors_regressor.py

@@ -0,0 +1,41 @@
+import numpy as np
+from .approximation import Approximation
+
+
+class NeighborsRegressor(Approximation):
+    """
+    A generic superclass for wrappers of *NeighborsRegressor from sklearn.
+
+    :param kwargs: arguments passed to the internal instance of
+        *NeighborsRegressor.
+    """
+
+    def __init__(self, **kwargs):
+        raise NotImplementedError("This class must be extended, not used.")
+
+    def fit(self, points, values):
+        """
+        Construct the interpolator given `points` and `values`.
+
+        :param array_like points: the coordinates of the points.
+        :param array_like values: the values in the points.
+        """
+        points = np.array(points).reshape(len(points), -1)
+        values = np.array(values)
+
+        self.regressor.fit(points, values)
+
+    def predict(self, new_point):
+        """
+        Evaluate interpolator at given `new_points`.
+
+        :param array_like new_points: the coordinates of the given points.
+        :return: the interpolated values.
+        :rtype: numpy.ndarray
+        """
+        if isinstance(new_point, np.ndarray) or isinstance(new_point, list):
+            new_point = np.array(new_point).reshape(len(new_point), -1)
+        else:
+            new_point = np.array([new_point])
+
+        return self.regressor.predict(new_point)

ezyrb/radius_neighbors_regressor.py

@@ -0,0 +1,20 @@
+"""
+Wrapper for RadiusNeighborsRegressor.
+"""
+
+import numpy as np
+from sklearn.neighbors import RadiusNeighborsRegressor as Regressor
+
+from .neighbors_regressor import NeighborsRegressor
+
+
+class RadiusNeighborsRegressor(NeighborsRegressor):
+    """
+    Radius Neighbors Regressor.
+
+    :param kwargs: arguments passed to the internal instance of
+        RadiusNeighborsRegressor.
+    """
+
+    def __init__(self, **kwargs):
+        self.regressor = Regressor(**kwargs)

setup.py

@@ -17,7 +17,7 @@
 KEYWORDS='pod interpolation reduced-basis model-order-reduction'
 
 REQUIRED = [
-    'future', 'numpy', 'scipy',	'matplotlib', 'GPy'
+    'future', 'numpy', 'scipy',	'matplotlib', 'GPy', 'sklearn'
 ]
 
 EXTRAS = {

tests/test_k_neighbors_regressor.py

@@ -0,0 +1,69 @@
+import numpy as np
+
+from unittest import TestCase
+from ezyrb import KNeighborsRegressor, Database, POD, ReducedOrderModel
+
+class TestKNeighbors(TestCase):
+    def test_params(self):
+        reg = KNeighborsRegressor(n_neighbors=20, algorithm='kd_tree')
+        assert reg.regressor.get_params()['n_neighbors'] == 20
+        assert reg.regressor.get_params()['algorithm'] == 'kd_tree'
+
+    def test_fit_onescalarparam_scalarfunc(self):
+        reg = KNeighborsRegressor()
+        reg.fit([1], [20])
+        assert reg.regressor.n_samples_fit_ == 1
+
+    def test_fit_scalarparam_scalarfunc(self):
+        reg = KNeighborsRegressor()
+        reg.fit([1, 2, 5, 7, 2], [2, 5, 7, 83, 3])
+        assert reg.regressor.n_samples_fit_ == 5
+
+    def test_fit_biparam_scalarfunc(self):
+        reg = KNeighborsRegressor()
+        reg.fit([[1, 2], [6, 7], [8, 9]], [1, 5, 6])
+        assert reg.regressor.n_samples_fit_ == 3
+
+    def test_fit_biparam_bifunc(self):
+        reg = KNeighborsRegressor()
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        assert reg.regressor.n_samples_fit_ == 3
+
+    def test_kneighbors(self):
+        reg = KNeighborsRegressor(n_neighbors=2)
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        neigh_idx = reg.regressor.kneighbors([[6, 6]], return_distance=False)[0]
+        assert neigh_idx[0] == 1
+        assert neigh_idx[1] == 2
+        assert len(neigh_idx) == 2
+
+    def test_predict(self):
+        reg = KNeighborsRegressor(n_neighbors=1)
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        neigh_idx = reg.regressor.predict([[1,2], [8,9], [6,7]])
+        assert (neigh_idx[0] == [1,0]).all()
+        assert (neigh_idx[1] == [8,6]).all()
+        assert (neigh_idx[2] == [20,5]).all()
+
+    def test_with_db_predict(self):
+        reg = KNeighborsRegressor(n_neighbors=1)
+        pod = POD()
+        db = Database(np.array([1, 2, 3])[:,None], np.array([1, 5, 3])[:,None])
+        rom = ReducedOrderModel(db, pod, reg)
+
+        rom.fit()
+        assert rom.predict([1]) == 1
+        assert rom.predict([2]) == 5
+        assert rom.predict([3]) == 3
+
+    def test_wrong1(self):
+        # wrong number of params
+        with self.assertRaises(Exception):
+            reg = KNeighborsRegressor()
+            reg.fit([[1, 2], [6,], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+
+    def test_wrong2(self):
+        # wrong number of values
+        with self.assertRaises(Exception):
+            reg = KNeighborsRegressor()
+            reg.fit([[1, 2], [6,], [8, 9]], [[20, 5], [8, 6]])

tests/test_radius_neighbors_regressor.py

@@ -0,0 +1,69 @@
+import numpy as np
+
+from unittest import TestCase
+from ezyrb import RadiusNeighborsRegressor, POD, Database, ReducedOrderModel
+
+class TestRadius(TestCase):
+    def test_params(self):
+        reg = RadiusNeighborsRegressor(radius=3.0, algorithm='kd_tree')
+        assert reg.regressor.get_params()['radius'] == 3.0
+        assert reg.regressor.get_params()['algorithm'] == 'kd_tree'
+
+    def test_fit_onescalarparam_scalarfunc(self):
+        reg = RadiusNeighborsRegressor()
+        reg.fit([1], [20])
+        assert reg.regressor.n_samples_fit_ == 1
+
+    def test_fit_scalarparam_scalarfunc(self):
+        reg = RadiusNeighborsRegressor()
+        reg.fit([1, 2, 5, 7, 2], [2, 5, 7, 83, 3])
+        assert reg.regressor.n_samples_fit_ == 5
+
+    def test_fit_biparam_scalarfunc(self):
+        reg = RadiusNeighborsRegressor()
+        reg.fit([[1, 2], [6, 7], [8, 9]], [1, 5, 6])
+        assert reg.regressor.n_samples_fit_ == 3
+
+    def test_fit_biparam_bifunc(self):
+        reg = RadiusNeighborsRegressor()
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        assert reg.regressor.n_samples_fit_ == 3
+
+    def test_radiusneighbors(self):
+        reg = RadiusNeighborsRegressor(radius=3.0)
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        neigh_idx = reg.regressor.radius_neighbors([[7,8]], return_distance=False)[0]
+        assert neigh_idx[0] == 1
+        assert neigh_idx[1] == 2
+        assert len(neigh_idx) == 2
+
+    def test_predict(self):
+        reg = RadiusNeighborsRegressor(radius=0.5)
+        reg.fit([[1, 2], [6, 7], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+        neigh_idx = reg.regressor.predict([[1,2], [8,9], [6,7]])
+        assert (neigh_idx[0] == [1,0]).all()
+        assert (neigh_idx[1] == [8,6]).all()
+        assert (neigh_idx[2] == [20, 5]).all()
+
+    def test_with_db_predict(self):
+        reg = RadiusNeighborsRegressor(radius=0.5)
+        pod = POD()
+        db = Database(np.array([1, 2, 3])[:,None], np.array([1, 5, 3])[:,None])
+        rom = ReducedOrderModel(db, pod, reg)
+
+        rom.fit()
+        assert rom.predict([1]) == 1
+        assert rom.predict([2]) == 5
+        assert rom.predict([3]) == 3
+
+    def test_wrong1(self):
+        # wrong number of params
+        with self.assertRaises(Exception):
+            reg = RadiusNeighborsRegressor()
+            reg.fit([[1, 2], [6,], [8, 9]], [[1, 0], [20, 5], [8, 6]])
+
+    def test_wrong2(self):
+        # wrong number of values
+        with self.assertRaises(Exception):
+            reg = RadiusNeighborsRegressor()
+            reg.fit([[1, 2], [6,], [8, 9]], [[20, 5], [8, 6]])