def test_load(self):
"""
Check viability and correctness of an LST saved then loaded from file.
"""
tree = dcl.construct_tree(self.dataset, self.k, prune_threshold=self.gamma)
with tempfile.NamedTemporaryFile() as f:
tree.save(f.name)
tree2 = dcl.load_tree(f.name)
self._check_tree_viability(tree2)
self._check_tree_correctness(tree)
def test_load(self):
"""
Check viability and correctness of an LST saved then loaded from file.
"""
tree = dcl.construct_tree(self.dataset,
self.k,
prune_threshold=self.gamma)
with tempfile.NamedTemporaryFile() as f:
tree.save(f.name)
tree2 = dcl.load_tree(f.name)
self._check_tree_viability(tree2)
self._check_tree_correctness(tree)
k = int(0.05 * n)
knn_graph, radii = dcl.utils.knn_graph(X, k, method='kd-tree')
density = dcl.utils.knn_density(radii, n, p, k)
## Build the level set tree.
gamma = int(0.1 * n)
tree = dcl.construct_tree_from_graph(knn_graph,
density,
prune_threshold=gamma,
verbose=True)
print tree
## Save the tree to disk and load it back (just for demo purposes).
tree.save('crater_tree.lst')
tree2 = dcl.load_tree('crater_tree.lst')
print tree2
## Get leaf clusters and leaf node indices.
leaves = tree.get_leaf_nodes()
labels = tree.get_clusters(method='leaf')
## Plot the tree, coloring the nodes used to cluster.
fig = tree.plot(form='mass', color_nodes=leaves, colormap=plt.cm.Spectral)[0]
fig.show()
## Plot the clusters in feature space.
X_clusters = X[labels[:, 0], :]
fig, ax = plt.subplots()
ax.set_title("High-density cluster labels")
def test_old_trees(self):
"""
Make sure models from previous versions of DeBaCl still load in the
current version.
"""
old_tree_dir = os.path.join(os.path.dirname(__file__), 'saved_trees')
print("old tree dir: {}".format(old_tree_dir))
for f in os.listdir(old_tree_dir):
## Load the tree.
tree = dcl.load_tree(os.path.join(old_tree_dir, f))
## Is it a level set tree?
self.assertIsInstance(tree, dcl.LevelSetTree)
## Make sure all methods run ok.
# printing
print(tree)
for c in ('start_level', 'end_level', 'start_mass', 'end_mass',
'size', 'parent', 'children'):
self.assertTrue(c in tree.__str__())
# pruning
old_gamma = tree.prune_threshold
new_gamma = 2 * tree.prune_threshold
pruned_tree = tree.prune(new_gamma)
self.assertEqual(tree.prune_threshold, old_gamma)
self.assertEqual(pruned_tree.prune_threshold, new_gamma)
# plotting
# plot = tree.plot() # Travis CI doesn't like this.
# retrieve clusters.
labels = tree.get_clusters()
n = len(tree.density)
self.assertTrue(labels.dtype is np.dtype('int64'))
self.assertLessEqual(len(labels), n)
self.assertEqual(len(labels[:, 0]), len(np.unique(labels[:, 0])))
self.assertGreaterEqual(np.min(labels), 0)
max_row_index = np.max(labels[:, 0])
self.assertLessEqual(max_row_index, n)
# retrieve leaf indices
leaves = tree.get_leaf_nodes()
self.assertIsInstance(leaves, list)
self.assertGreater(len(leaves), 0)
# full partition
partition = tree.branch_partition()
self.assertTrue(partition.dtype is np.dtype('int64'))
self.assertEqual(len(partition), n)
self.assertEqual(len(partition[:, 0]), len(np.unique(partition[:, 0])))
self.assertEqual(np.min(partition[:, 0]), 0)
self.assertItemsEqual(np.unique(partition[:, 1]),
tree.nodes.keys())
# save and load
with tempfile.NamedTemporaryFile() as t:
tree.save(t.name)
tree2 = dcl.load_tree(t.name)
self.assertItemsEqual([x.start_level for x in tree.nodes.values()],
[y.start_level for y in tree2.nodes.values()])
def test_old_trees(self):
"""
Make sure models from previous versions of DeBaCl still load in the
current version.
"""
old_tree_dir = os.path.join(os.path.dirname(__file__), 'saved_trees')
print("old tree dir: {}".format(old_tree_dir))
for f in os.listdir(old_tree_dir):
## Load the tree.
tree = dcl.load_tree(os.path.join(old_tree_dir, f))
## Is it a level set tree?
self.assertIsInstance(tree, dcl.LevelSetTree)
## Make sure all methods run ok.
# printing
print(tree)
for c in ('start_level', 'end_level', 'start_mass', 'end_mass',
'size', 'parent', 'children'):
self.assertTrue(c in tree.__str__())
# pruning
old_gamma = tree.prune_threshold
new_gamma = 2 * tree.prune_threshold
pruned_tree = tree.prune(new_gamma)
self.assertEqual(tree.prune_threshold, old_gamma)
self.assertEqual(pruned_tree.prune_threshold, new_gamma)
# plotting
# plot = tree.plot() # Travis CI doesn't like this.
# retrieve clusters.
labels = tree.get_clusters()
n = len(tree.density)
self.assertTrue(labels.dtype is np.dtype('int64'))
self.assertLessEqual(len(labels), n)
self.assertEqual(len(labels[:, 0]), len(np.unique(labels[:, 0])))
self.assertGreaterEqual(np.min(labels), 0)
max_row_index = np.max(labels[:, 0])
self.assertLessEqual(max_row_index, n)
# retrieve leaf indices
leaves = tree.get_leaf_nodes()
self.assertIsInstance(leaves, list)
self.assertGreater(len(leaves), 0)
# full partition
partition = tree.branch_partition()
self.assertTrue(partition.dtype is np.dtype('int64'))
self.assertEqual(len(partition), n)
self.assertEqual(len(partition[:, 0]),
len(np.unique(partition[:, 0])))
self.assertEqual(np.min(partition[:, 0]), 0)
self.assertItemsEqual(np.unique(partition[:, 1]),
tree.nodes.keys())
# save and load
with tempfile.NamedTemporaryFile() as t:
tree.save(t.name)
tree2 = dcl.load_tree(t.name)
self.assertItemsEqual(
[x.start_level for x in tree.nodes.values()],
[y.start_level for y in tree2.nodes.values()])
k = int(0.05 * n)
knn_graph, radii = dcl.utils.knn_graph(X, k, method='kd-tree')
density = dcl.utils.knn_density(radii, n, p, k)
## Build the level set tree.
gamma = int(0.1 * n)
tree = dcl.construct_tree_from_graph(knn_graph, density, prune_threshold=gamma,
verbose=True)
print tree
## Save the tree to disk and load it back (just for demo purposes).
tree.save('crater_tree.lst')
tree2 = dcl.load_tree('crater_tree.lst')
print tree2
## Get leaf clusters and leaf node indices.
leaves = tree.get_leaf_nodes()
labels = tree.get_clusters(method='leaf')
## Plot the tree, coloring the nodes used to cluster.
fig = tree.plot(form='mass', color_nodes=leaves, colormap=plt.cm.Spectral)[0]
fig.show()
## Plot the clusters in feature space.
X_clusters = X[labels[:, 0], :]
