def test_increment():
# arguments values
mode_values = ["concatenation", "subtraction"]
n_features_per_vertex_values = [2, 3]
sparse_values = [True, False]
n_components_values = [None, 30]
# create graph
n_vertices = 6
edges = np.array([[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 0]])
graphs = [
DirectedGraph.init_from_edges(edges, n_vertices),
UndirectedGraph(np.zeros((n_vertices, n_vertices))),
]
for n_features_per_vertex in n_features_per_vertex_values:
# create samples
n_samples = 100
samples = []
for i in range(n_samples):
samples.append(np.random.rand(n_vertices * n_features_per_vertex))
for graph in graphs:
for mode in mode_values:
for sparse in sparse_values:
for n_components in n_components_values:
# Incremental GMRF
gmrf1 = GMRFVectorModel(
samples[:50],
graph,
mode=mode,
sparse=sparse,
n_components=n_components,
dtype=np.float64,
incremental=True,
)
gmrf1.increment(samples[50::])
# Non incremental GMRF
gmrf2 = GMRFVectorModel(
samples,
graph,
mode=mode,
sparse=sparse,
n_components=n_components,
dtype=np.float64,
)
# Compare
if sparse:
assert_array_almost_equal(
gmrf1.precision.todense(),
gmrf2.precision.todense())
assert_array_almost_equal(gmrf1.mean_vector,
gmrf2.mean_vector)
def test_mahalanobis_distance():
# arguments values
mode_values = ['concatenation', 'subtraction']
n_features_per_vertex_values = [2, 3]
sparse_values = [True, False]
subtract_mean_values = [True, False]
n_components_values = [None, 30]
# create graph
n_vertices = 6
edges = np.array([[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 0]])
graphs = [
DirectedGraph.init_from_edges(edges, n_vertices),
UndirectedGraph(np.zeros((n_vertices, n_vertices)))
]
for n_features_per_vertex in n_features_per_vertex_values:
# create samples
n_samples = 50
samples = []
for i in range(n_samples):
samples.append(
PointCloud(np.random.rand(n_vertices, n_features_per_vertex)))
test_sample = PointCloud(
np.random.rand(n_vertices, n_features_per_vertex))
for graph in graphs:
for mode in mode_values:
for sparse in sparse_values:
for n_components in n_components_values:
# train GMRF
gmrf = GMRFModel(samples,
graph,
mode=mode,
sparse=sparse,
n_components=n_components,
dtype=np.float64)
for subtract_mean in subtract_mean_values:
# compute costs
if graph.n_edges == 0:
cost1 = _compute_sum_cost_block_diagonal(
samples, test_sample, graph,
n_features_per_vertex, subtract_mean)
else:
cost1 = _compute_sum_cost_block_sparse(
samples, test_sample, graph,
n_features_per_vertex, subtract_mean, mode)
cost2 = gmrf.mahalanobis_distance(
test_sample, subtract_mean=subtract_mean)
assert_almost_equal(cost1, cost2)
def test_mahalanobis_distance():
# arguments values
mode_values = ['concatenation', 'subtraction']
n_features_per_vertex_values = [2, 3]
sparse_values = [True, False]
subtract_mean_values = [True, False]
n_components_values = [None, 30]
# create graph
n_vertices = 6
edges = np.array([[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 0]])
graphs = [DirectedGraph.init_from_edges(edges, n_vertices),
UndirectedGraph(np.zeros((n_vertices, n_vertices)))]
for n_features_per_vertex in n_features_per_vertex_values:
# create samples
n_samples = 50
samples = []
for i in range(n_samples):
samples.append(PointCloud(np.random.rand(n_vertices,
n_features_per_vertex)))
test_sample = PointCloud(np.random.rand(n_vertices,
n_features_per_vertex))
for graph in graphs:
for mode in mode_values:
for sparse in sparse_values:
for n_components in n_components_values:
# train GMRF
gmrf = GMRFModel(
samples, graph, mode=mode, sparse=sparse,
n_components=n_components, dtype=np.float64)
for subtract_mean in subtract_mean_values:
# compute costs
if graph.n_edges == 0:
cost1 = _compute_sum_cost_block_diagonal(
samples, test_sample, graph,
n_features_per_vertex, subtract_mean)
else:
cost1 = _compute_sum_cost_block_sparse(
samples, test_sample, graph,
n_features_per_vertex, subtract_mean, mode)
cost2 = gmrf.mahalanobis_distance(
test_sample, subtract_mean=subtract_mean)
assert_almost_equal(cost1, cost2)
def test_increment():
# arguments values
mode_values = ['concatenation', 'subtraction']
n_features_per_vertex_values = [2, 3]
sparse_values = [True, False]
n_components_values = [None, 30]
# create graph
n_vertices = 6
edges = np.array([[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 0]])
graphs = [DirectedGraph.init_from_edges(edges, n_vertices),
UndirectedGraph(np.zeros((n_vertices, n_vertices)))]
for n_features_per_vertex in n_features_per_vertex_values:
# create samples
n_samples = 100
samples = []
for i in range(n_samples):
samples.append(np.random.rand(n_vertices * n_features_per_vertex))
for graph in graphs:
for mode in mode_values:
for sparse in sparse_values:
for n_components in n_components_values:
# Incremental GMRF
gmrf1 = GMRFVectorModel(
samples[:50], graph, mode=mode, sparse=sparse,
n_components=n_components, dtype=np.float64,
incremental=True)
gmrf1.increment(samples[50::])
# Non incremental GMRF
gmrf2 = GMRFVectorModel(
samples, graph, mode=mode, sparse=sparse,
n_components=n_components, dtype=np.float64)
# Compare
if sparse:
assert_array_almost_equal(gmrf1.precision.todense(),
gmrf2.precision.todense())
assert_array_almost_equal(gmrf1.mean_vector,
gmrf2.mean_vector)
