def test_gstat_optics():
"""Test the GapStatClustering using OPTICS as the
base clusterer.
"""
# OPTICS is incompatble with GapStatClustering
# because it doesn't allow number of clusters to be specified
with pytest.raises(AttributeError):
gapstat.GapStatClustering(base_clusterer=OPTICS())
def test_gstat_meanshift():
"""Test the GapStatClustering using MeanShift as the
base clusterer.
"""
# meanshift is incompatble with GapStatClustering
# because it doesn't allow number of clusters to be specified
with pytest.raises(AttributeError):
gapstat.GapStatClustering(base_clusterer=MeanShift())
def test_gstat_dbscan():
"""Test the GapStatClustering using DBSCAN as the
base clusterer.
"""
# dbscan is incompatble with GapStatClustering
# because it doesn't allow number of clusters to be specified
with pytest.raises(AttributeError):
gapstat.GapStatClustering(base_clusterer=DBSCAN())
def test_gstat_affinitypropagation():
"""Test the GapStatClustering using AffinityPropagation as the
base clusterer.
"""
# affinitypropagation is incompatble with GapStatClustering
# because it doesn't allow number of clusters to be specified
with pytest.raises(AttributeError):
gapstat.GapStatClustering(base_clusterer=AffinityPropagation())
def _run_transform_test(base_clusterer=None):
"""Test GapStatClustering.transform() and GapStatClustering.fit_transform()
methods using the specified base clusterer.
"""
# construct test data
T, _, K = _quad4(center=[0, 0], offset1=[0.5, 0.5], offset2=[2, 2])
n_samples = T.shape[0]
# create gapstat clusterer with base clusterer
gstat = gapstat.GapStatClustering(base_clusterer=base_clusterer)
base_clusterer = gstat.base_clusterer
# test transform() before fit()
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'transform')):
# test transform() before fit()
with pytest.raises(NotFittedError):
gstat.transform(T)
else:
# test unsupported transform()
with pytest.raises(AttributeError):
gstat.transform(T)
# test when max_k is too small
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'fit_transform')):
# test when max_k is too small
gstat.set_params(max_k=K-1)
with pytest.raises(NotFittedError):
gstat.fit_transform(T)
else:
# test unsupported fit_transform()
with pytest.raises(AttributeError):
gstat.fit_transform(T)
# determine expected number of columns in transformed data
# -- in most cases this is the number of clusters
# -- for birch this is the number of subclusters
if (isinstance(base_clusterer, Birch)):
n_features = len(base_clusterer.subcluster_centers_)
else:
n_features = K
# test fit_transform()
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'fit_transform')):
gstat.set_params(max_k=K)
transformed_T = gstat.fit_transform(T)
_check_transformed(transformed_T, n_samples, n_features)
# test transform()
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'transform')):
transformed_T = gstat.transform(T)
_check_transformed(transformed_T, n_samples, n_features)
def test_gstat_featureagglomeration():
"""Test the GapStatClustering using FeatureAgglomeration as the
base clusterer.
"""
# FeatureAgglomeration is incompatble with GapStatClustering
# because it transforms the data set before clusering
# which breaks some of the gapstat logic
# Instead, use AgglomerativeClustering and transform the
# data set before fitting the data
with pytest.raises(AttributeError):
gapstat.GapStatClustering(base_clusterer=FeatureAgglomeration())
def _run_fit_test(base_clusterer=None):
"""Test GapStatClustering.fit() method using the specified base clusterer.
"""
# construct test data
T, _, K = _quad4(center=[0, 0], offset1=[0.5, 0.5], offset2=[2, 2])
n_samples = T.shape[0]
# create gapstat clusterer with base clusterer
gstat = gapstat.GapStatClustering(base_clusterer=base_clusterer)
base_clusterer = gstat.base_clusterer
# test when max_k is too small
gstat.set_params(max_k=K-1)
with pytest.raises(NotFittedError):
gstat.fit(T)
# test fit()
gstat.set_params(max_k=K)
gstat = gstat.fit(T)
_check_labels(gstat.labels_, gstat.n_clusters_, n_samples, K)
def _run_predict_test(base_clusterer=None):
"""Test GapStatClustering.predict() and GapStatClustering.fit_predict()
methods using the specified base clusterer.
"""
# construct test data
T, _, K = _quad4(center=[0, 0], offset1=[0.5, 0.5], offset2=[2, 2])
n_samples = T.shape[0]
# create gapstat clusterer with base clusterer
gstat = gapstat.GapStatClustering(base_clusterer=base_clusterer)
base_clusterer = gstat.base_clusterer
# test predict() before fit()
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'predict')):
# test predict() before fit()
with pytest.raises(NotFittedError):
gstat.predict(T)
else:
# test unsupported predict()
with pytest.raises(AttributeError):
gstat.predict(T)
# test when max_k is too small
gstat.set_params(max_k=K-1)
with pytest.raises(NotFittedError):
gstat.fit_predict(T)
# test fit_predict()
gstat.set_params(max_k=K)
predicted_labels = gstat.fit_predict(T)
_check_labels(predicted_labels, gstat.n_clusters_, n_samples, K)
_check_labels(gstat.labels_, gstat.n_clusters_, n_samples, K)
# test predict()
# adapt test case to the base_cluster capabilities
if (hasattr(base_clusterer, 'predict')):
predicted_labels = gstat.predict(T)
_check_labels(predicted_labels, gstat.n_clusters_, n_samples, K)
_check_labels(gstat.labels_, gstat.n_clusters_, n_samples, K)