def test(self):
from topologylayer.nn import LevelSetLayer2D
# sublevel set
for alg in ['hom', 'hom2', 'cohom']:
layer = LevelSetLayer2D(size=(3, 3),
maxdim=1,
sublevel=False,
alg=alg)
x = torch.tensor([[2, 1, 1], [1, 0.5, 1], [1, 1, 1]],
dtype=torch.float).requires_grad_(True)
dgms, issub = layer(x)
self.assertEqual(issub, False,
"Expected superlevel set layer. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[0]),
torch.tensor([[2., -np.inf]]))), True,
"unexpected 0-dim barcode. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[1]),
torch.tensor([[1., 0.5]]))), True,
"unexpected 1-dim barcode. alg=" + alg)
d1 = remove_zero_bars(dgms[1])
p = d1[0, 0] - d1[0, 1]
p.backward()
self.assertEqual(x.grad[1, 1].item(), -1.0,
"unexpected gradient. alg=" + alg)
def test(self):
from topologylayer.nn import RipsLayer
for alg in ['hom', 'hom2', 'cohom']:
# superlevel set
layer = RipsLayer(4, maxdim=1, alg=alg)
x = torch.tensor([[1, 1], [1, -1], [-1, -1], [-1, 1]],
dtype=torch.float).requires_grad_(True)
dgms, issub = layer(x)
self.assertEqual(issub, True,
"Expected sublevel set layer. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(
remove_infinite_bars(remove_zero_bars(dgms[0]), issub),
torch.tensor([[0., 2.], [0., 2.], [0., 2.]]))), True,
"unexpected 0-dim barcode. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[1]),
torch.tensor([[2., 2.8284270763397217]]))), True,
"unexpected 1-dim barcode. alg=" + alg)
d0 = remove_infinite_bars(remove_zero_bars(dgms[0]), issub)
p = torch.sum(d0[:, 1] - d0[:, 0])
p.backward()
self.assertEqual(
torch.all(
torch.eq(
x.grad,
torch.tensor([[1, 1], [1, -1], [-1, 0], [-1, 0]],
dtype=torch.float))), True,
"unexpected gradient. alg=" + alg)
def test(self):
from topologylayer.nn import AlphaLayer
# superlevel set
layer = AlphaLayer(maxdim=1)
x = torch.tensor([[1, 1], [1, -1], [-1, -1], [-1, 1]],
dtype=torch.float).requires_grad_(True)
dgms, issub = layer(x)
self.assertEqual(issub, True, "Expected sublevel set layer")
self.assertEqual(
torch.all(
torch.eq(
remove_zero_bars(dgms[0]),
torch.tensor([[0., 2.], [0., 2.], [0., 2.],
[0., np.inf]]))), True,
"unexpected 0-dim barcode")
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[1]),
torch.tensor([[2., 2.8284270763397217]]))), True,
"unexpected 1-dim barcode")
d0 = remove_zero_bars(dgms[0])
p = d0[0, 1] - d0[0, 0]
p.backward()
self.assertEqual(
torch.all(
torch.eq(
x.grad,
torch.tensor([[0, 1], [0, -1], [0, 0], [0, 0]],
dtype=torch.float))), True,
"unexpected gradient")
def test(self):
# first, we build our complex
s = SimplicialComplex()
# a cycle graph on vertices 1,2,3,4
# cone with vertex 0
s.append([0])
s.append([1])
s.append([2])
s.append([3])
s.append([4])
s.append([0, 1])
s.append([0, 2])
s.append([0, 3])
s.append([0, 4])
s.append([1, 2])
s.append([1, 3])
s.append([4, 2])
s.append([4, 3])
s.append([0, 1, 2])
s.append([0, 1, 3])
s.append([0, 2, 4])
s.append([0, 3, 4])
# initialize internal data structures
s.initialize()
# function on vertices
# we are doing sub-level set persistence
# expect single H0 [0,inf]
# expect single H1 [0,2]
f = torch.Tensor([2., 0., 0., 0., 0.])
# extend filtration to simplical complex
s.extendFloat(f)
# compute persistence with MAXDIM=1
ret = persistenceForwardCohom(s, 1)
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(ret[0]),
torch.tensor([[0., np.inf]]))), True,
"Unexpected 0-Dim persistence")
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(ret[1]), torch.tensor([[0., 2.]]))),
True, "Unexpected 1-Dim persistence")
def forward(self, dgminfo):
dgms, issublevel = dgminfo
dgm = dgms[self.dim]
if self.remove_zero:
dgm = remove_zero_bars(dgm)
lengths, means = get_barcode_lengths_means(dgm, issublevel)
return torch.sum(
torch.mul(torch.pow(lengths, self.p), torch.pow(means, self.q)))
def test(self):
from topologylayer.nn import LevelSetLayer1D
# superlevel set
for alg in ['hom', 'hom2', 'cohom']:
layer = LevelSetLayer1D(size=3, sublevel=False, alg=alg)
y = torch.tensor([1, 0, 1], dtype=torch.float).requires_grad_(True)
dgms, issub = layer(y)
self.assertEqual(issub, False,
"Expected superlevel set layer. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(
remove_infinite_bars(remove_zero_bars(dgms[0]), issub),
torch.tensor([[1., 0.]]))), True,
"unexpected barcode. alg=" + alg)
p = torch.sum(
remove_infinite_bars(remove_zero_bars(dgms[0]), issub)[0])
p.backward()
self.assertEqual(y.grad[1].item(), 1.0,
"unexpected gradient. alg=" + alg)
def test(self):
from topologylayer.nn import LevelSetLayer1D
# sublevel set
for alg in ['hom', 'hom2', 'cohom']:
layer = LevelSetLayer1D(size=3, sublevel=True, alg=alg)
y = torch.tensor([1, 0, 1], dtype=torch.float).requires_grad_(True)
dgms, issub = layer(y)
self.assertEqual(issub, True,
"Expected sublevel set layer. alg=" + alg)
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[0]),
torch.tensor([[0., np.inf]]))), True,
"unexpected barcode. alg=" + alg)
def test(self):
from topologylayer.nn import LevelSetLayer1D
# superlevel set
layer = LevelSetLayer1D(size=3, sublevel=False)
y = torch.tensor([1, 0, 1], dtype=torch.float).requires_grad_(True)
dgms, issub = layer(y)
self.assertEqual(issub, False, "Expected superlevel set layer")
self.assertEqual(
torch.all(
torch.eq(remove_zero_bars(dgms[0]),
torch.tensor([[1., 0.], [1., -np.inf]]))), True,
"unexpected barcode")
p = torch.sum(dgms[0][1])
p.backward()
self.assertEqual(y.grad[1].item(), 1.0, "unexpected gradient")
s.append([0,4])
s.append([1,2])
s.append([1,3])
s.append([4,2])
s.append([4,3])
s.append([0,1,2])
s.append([0,1,3])
s.append([0,2,4])
s.append([0,3,4])
# initialize internal data structures
s.initialize()
# function on vertices
# we are doing sub-level set persistence
# expect single H0 [0,inf]
# expect single H1 [0,2]
f = torch.Tensor([2., 0., 0., 0., 0.])
# extend filtration to simplical complex
s.extendFloat(f)
# compute persistence with MAXDIM=1
ret = persistenceForward(s, 1)
for k in range(2):
print("dimension %d bars" % k)
print(remove_zero_bars(ret[k]))