def test_kneighbors(self):
"""Test k neighbor searches for all k-neighbors estimators"""
nn = NearestNeighbors()
nn.fit(self.X)
lof = LocalOutlierFactor(n_neighbors=5)
lof.fit(self.X)
knn = KNeighborsClassifier()
knn.fit(self.X, self.y)
knnr = KNeighborsRegressor()
knnr.fit(self.X, self.modes_location)
for neigh in [nn, knn, knnr, lof]:
dist, links = neigh.kneighbors(self.X[:4])
np.testing.assert_array_equal(links, [[0, 7, 21, 23, 15],
[1, 12, 19, 18, 17],
[2, 17, 22, 27, 26],
[3, 4, 9, 5, 25]])
graph = neigh.kneighbors_graph(self.X[:4])
dist_kneigh = lp_distance(self.X[0], self.X[7])
np.testing.assert_array_almost_equal(dist[0, 1], dist_kneigh)
for i in range(30):
self.assertEqual(graph[0, i] == 1.0, i in links[0])
self.assertEqual(graph[0, i] == 0.0, i not in links[0])
def test_score_scalar_response(self):
neigh = KNeighborsRegressor()
neigh.fit(self.X, self.modes_location)
r = neigh.score(self.X, self.modes_location)
np.testing.assert_almost_equal(r, 0.9975889963743335)
def test_functional_response_basis(self):
knnr = KNeighborsRegressor(weights='distance', n_neighbors=5)
response = self.X.to_basis(Fourier(domain_range=(-1, 1), n_basis=10))
knnr.fit(self.X, response)
res = knnr.predict(self.X)
np.testing.assert_array_almost_equal(res.coefficients,
response.coefficients)
def test_knn_functional_response(self):
knnr = KNeighborsRegressor(n_neighbors=1)
knnr.fit(self.X, self.X)
res = knnr.predict(self.X)
np.testing.assert_array_almost_equal(res.data_matrix,
self.X.data_matrix)
def test_knn_functional_response_sklearn(self):
# Check sklearn metric
knnr = KNeighborsRegressor(n_neighbors=1, metric='euclidean',
multivariate_metric=True)
knnr.fit(self.X, self.X)
res = knnr.predict(self.X)
np.testing.assert_array_almost_equal(res.data_matrix,
self.X.data_matrix)
def test_knn_functional_response_precomputed(self):
knnr = KNeighborsRegressor(n_neighbors=4, weights='distance',
metric='precomputed')
d = pairwise_distance(lp_distance)
distances = d(self.X[:4], self.X[:4])
knnr.fit(distances, self.X[:4])
res = knnr.predict(distances)
np.testing.assert_array_almost_equal(res.data_matrix,
self.X[:4].data_matrix)
def test_functional_response_custom_weights(self):
def weights(weights):
return np.array([w == 0 for w in weights], dtype=float)
knnr = KNeighborsRegressor(weights=weights, n_neighbors=5)
response = self.X.to_basis(Fourier(domain_range=(-1, 1), n_basis=10))
knnr.fit(self.X, response)
res = knnr.predict(self.X)
np.testing.assert_array_almost_equal(res.coefficients,
response.coefficients)
def test_predict_regressor(self):
"""Test scalar regression, predics mode location"""
# Dummy test, with weight = distance, only the sample with distance 0
# will be returned, obtaining the exact location
knnr = KNeighborsRegressor(weights='distance')
rnnr = RadiusNeighborsRegressor(weights='distance', radius=.1)
knnr.fit(self.X, self.modes_location)
rnnr.fit(self.X, self.modes_location)
np.testing.assert_array_almost_equal(knnr.predict(self.X),
self.modes_location)
np.testing.assert_array_almost_equal(rnnr.predict(self.X),
self.modes_location)
def test_functional_regression_distance_weights(self):
knnr = KNeighborsRegressor(
weights='distance', n_neighbors=10)
knnr.fit(self.X[:10], self.X[:10])
res = knnr.predict(self.X[11])
d = pairwise_distance(lp_distance)
distances = d(self.X[:10], self.X[11]).flatten()
weights = 1 / distances
weights /= weights.sum()
response = self.X[:10].mean(weights=weights)
np.testing.assert_array_almost_equal(res.data_matrix,
response.data_matrix)
def test_score_functional_response(self):
neigh = KNeighborsRegressor()
y = 5 * self.X + 1
neigh.fit(self.X, y)
r = neigh.score(self.X, y)
np.testing.assert_almost_equal(r, 0.962651178452408)
# Weighted case and basis form
y = y.to_basis(Fourier(domain_range=y.domain_range[0], n_basis=5))
neigh.fit(self.X, y)
r = neigh.score(self.X[:7], y[:7],
sample_weight=4 * [1. / 5] + 3 * [1. / 15])
np.testing.assert_almost_equal(r, 0.9982527586114364)
#
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=.1,
random_state=28)
##############################################################################
#
# We will try make a prediction using 5 neighbors and the :math:`\mathbb{L}^2`
# distance. In this case, to calculate
# the response we will use a mean of the response, weighted by their distance
# to the test sample.
#
knn = KNeighborsRegressor(n_neighbors=5, weights='distance')
knn.fit(X_train, y_train)
##############################################################################
#
# We can predict values for the test partition using
# :meth:`~skfda.ml.regression.KNeighborsFunctionalRegressor.predict`. The
# following figure shows the real precipitation curves, in dashed line, and
# the predicted ones.
#
y_pred = knn.predict(X_test)
# Plot prediction
fig = y_pred.plot()
fig.axes[0].set_prop_cycle(None) # Reset colors
random_state=7)
##############################################################################
#
# Firstly we will try make a prediction with the default values of the
# estimator, using 5 neighbors and the :math:`\mathbb{L}^2` distance.
#
# We can fit the :class:`~skfda.ml.regression.KNeighborsRegressor` in the
# same way than the
# sklearn estimators. This estimator is an extension of the sklearn
# :class:`~sklearn.neighbors.KNeighborsRegressor`, but accepting a
# :class:`~skfda.representation.grid.FDataGrid` as input instead of an array with
# multivariate data.
#
knn = KNeighborsRegressor(weights='distance')
knn.fit(X_train, y_train)
##############################################################################
#
# We can predict values for the test partition using
# :meth:`~skfda.ml.regression.KNeighborsScalarRegressor.predict`.
#
pred = knn.predict(X_test)
print(pred)
##############################################################################
#
# The following figure compares the real precipitations with the predicted
# values.