def test_on_trivial_input(inp):
"""Test that with one cluster, and one point, we always get one cluster,
regardless of its location."""
n_points_per_cluster, n_clusters, dim, pts = inp
fs = FirstSimpleGap()
fs = fs.fit(pts)
assert fs.n_clusters_ == n_clusters
fh = FirstHistogramGap()
fh = fh.fit(pts)
assert fh.n_clusters_ == n_clusters
def test_max_fraction_clusters(inp, max_frac):
""" Check that ``FirstSimpleGap`` and ``FirstHistogramGap`` respect the
``max_num_clusters`` constraint, if it is set."""
n_points_per_cluster, n_clusters, _, pts = inp
max_num_clusters = max_frac * n_points_per_cluster * n_clusters
fs = FirstSimpleGap(max_fraction=max_frac)
_ = fs.fit_predict(pts)
assert fs.n_clusters_ <= np.floor(max_num_clusters)
fh = FirstHistogramGap(max_fraction=max_frac)
_ = fh.fit_predict(pts)
assert fh.n_clusters_ <= np.floor(max_num_clusters)
def test_firsthistogramgap(inp):
"""For a multimodal distribution, check that the ``FirstHistogramGap`` with
appropriate parameters finds the right number of clusters, and that each
has the correct number of points ``n_points_per_cluster``."""
n_points_per_cluster, n_clusters, _, pts = inp
fh = FirstHistogramGap(freq_threshold=0,
max_fraction=1.,
n_bins_start=5,
affinity='euclidean',
memory=None,
linkage='single')
preds = fh.fit_predict(pts)
unique, counts = np.unique(preds, return_counts=True)
# check that the nb of clusters corresponds to the nb of synth. clusters
assert unique.shape[0] == n_clusters
# check that the nb of pts in a cluster corresponds to what we expect
assert_almost_equal(counts, n_points_per_cluster)
def test_precomputed_distances(inp):
"""Verify that the clustering based on a distance matrix is the same as
the clustering on points used to calculate that distance matrix."""
n_points_per_cluster, n_clusters, _, pts = inp
dist_matrix = distance_matrix(pts, pts, p=2)
fh_matrix = FirstHistogramGap(freq_threshold=0,
max_fraction=1.,
n_bins_start=5,
affinity='precomputed',
memory=None,
linkage='single')
preds_mat = fh_matrix.fit_predict(dist_matrix)
fh = FirstHistogramGap(freq_threshold=0,
max_fraction=1.,
n_bins_start=5,
affinity='euclidean',
memory=None,
linkage='single')
preds = fh.fit_predict(pts)
indices_cluster = set(preds)
def get_partition_from_preds(preds):
"""From a vector of predictions (labels), get a set of frozensets,
where each frozenset represents a cluster, and has the indices of rows
of its elements."""
return set(
[frozenset(np.where(preds == c)[0]) for c in indices_cluster])
assert get_partition_from_preds(preds) == \
get_partition_from_preds(preds_mat)
def test_precomputed_distances(inp):
"""Verify that the clustering based on ``distance_matrix`` is the same
as the clustering on points, that were used to calculate
that distance matrix."""
n_points_per_cluster, n_clusters, _, pts = inp
dist_matrix = distance_matrix(pts, pts, p=2)
fh_matrix = FirstHistogramGap(freq_threshold=0,
max_fraction=None,
n_bins_start=5,
affinity='precomputed',
memory=None,
linkage='single')
preds_mat = fh_matrix.fit_predict(dist_matrix)
fh = FirstHistogramGap(freq_threshold=0,
max_fraction=None,
n_bins_start=5,
affinity='euclidean',
memory=None,
linkage='single')
preds = fh.fit_predict(pts)
assert_almost_equal(preds, preds_mat)