def test_fps_consistency(self):
"""
Checks if different FPS methods give consistent results.
"""
ref = fps(self.x, 100, 0, method="simple")
r_fps = fps(self.x, 100, 0, method="voronoi")
self.fps_return_match(r_fps, ref)
def test_simple_fps(self):
"""
Just runs a small FPS run and see if the results make sense.
"""
r_fps = fps(self.x, 100, 0, method="simple")
self.assertTrue("fps_indices" in r_fps)
self.assertTrue("fps_minmax_d2" in r_fps)
self.assertTrue("fps_hausdorff_d2" in r_fps)
self.assertTrue(
r_fps["fps_hausdorff_d2"].max() <= r_fps["fps_minmax_d2"][-2])
self.assertTrue(is_sorted(r_fps["fps_minmax_d2"]))
def test_fps_restart(self):
"""
Checks if FPS restart works as intended.
"""
# This is the reference selection
ref = fps(self.x, 100, 0, method="simple")
# now we do this in two stages, by restarting
# I first select 50
r_fps = fps(self.x, 50, 0, method="simple")
# I ask for 100 but provide the restart tuple.
# Selection will continue from point 51
r_fps = fps(self.x, 100, 0, method="simple", restart=r_fps)
self.fps_return_match(r_fps, ref)
# check if this works also when recomputing the distances
r_fps = fps(self.x, 50, 0, method="simple")
r_fps = fps(
self.x,
100,
0,
method="simple",
restart={"fps_indices": r_fps["fps_indices"]},
)
self.fps_return_match(r_fps, ref)
# You can also add new points (in this case we reuse
# the same feature matrix) and continue the selection
r_fps = fps(self.x, 50, 0, method="simple")
xx = np.concatenate((self.x[r_fps["fps_indices"]], self.x))
r_fps = fps(
xx,
100,
0,
method="simple",
restart={"fps_indices": np.asarray(range(50))},
)
self.assertTrue(
np.array_equal(r_fps["fps_indices"][:50], np.asarray(range(50))))
self.assertTrue(
np.array_equal(r_fps["fps_indices"][50:] - 50,
ref["fps_indices"][50:]))