def test_estimate_transform_spd(self):
point = spd.SPDMatrices(3).random_uniform()
points = gs.stack([point, point])
transformer = ToTangentSpace(geometry=spd.SPDMetricAffine(3))
transformer.fit(X=points)
result = transformer.transform(points)
expected = gs.zeros((2, 6))
self.assertAllClose(expected, result, atol=1e-5)
def __init__(self, dim, metric_type="bw"):
super(SPDManifold, self).__init__(dim)
if metric_type == "bw":
self.metric = spd.SPDMetricBuresWasserstein(dim)
elif metric_type == "euclid":
self.metric = spd.SPDMetricEuclidean(dim)
elif metric_type == "logeuclid":
self.metric = spd.SPDMetricLogEuclidean(dim)
else:
self.metric = spd.SPDMetricAffine(dim)
def test_inverse_transform_spd(self):
point = spd.SPDMatrices(3).random_uniform(10)
transformer = ToTangentSpace(geometry=spd.SPDMetricLogEuclidean(3))
X = transformer.fit_transform(X=point)
result = transformer.inverse_transform(X)
expected = point
self.assertAllClose(expected, result, atol=1e-4)
transformer = ToTangentSpace(geometry=spd.SPDMetricAffine(3))
X = transformer.fit_transform(X=point)
result = transformer.inverse_transform(X)
expected = point
self.assertAllClose(expected, result, atol=1e-4)
def test_spd_kmeans_fit(self):
gs.random.seed(0)
dim = 3
n_points = 2
space = spd_matrices.SPDMatrices(dim)
data = space.random_point(n_samples=n_points)
metric = spd_matrices.SPDMetricAffine(dim)
kmeans = RiemannianKMeans(metric, n_clusters=1, lr=1.0)
kmeans.fit(data)
result = kmeans.centroids
mean = FrechetMean(metric=metric, point_type="matrix", max_iter=100)
mean.fit(data)
expected = mean.estimate_
self.assertAllClose(result, expected)
