def test_multi_level(self, dtype, device):
N = 16
G = rand_udg(N, 0.2, dtype, device)
bptG, beta, _, _, _ = harary(G)
basis = QmfOperator(bptG, beta, order=4, device=device)
x = torch.ones(N, 1, dtype=dtype, device=device)
y = basis(x)
z = basis.inverse_transform(y)
print("\nsnr: ",
snr(z.permute(-1, -2), x.permute(-1, -2)).item(), "dB.")
print("dis: ", (z - x).abs().sum())
def test_multi_level(self, dtype, device):
N = 120
G = rand_udg(N, 0.4, dtype, device)
bptG, beta, _, _, _ = harary(G)
basis = BiorthOperator(bptG, beta, k=2, device=device)
x = torch.ones(N, 1, dtype=dtype, device=device)
y = basis.transform(x)
z = basis.inverse_transform(y)
print("\nsnr: ",
snr(z.permute(-1, -2), x.permute(-1, -2)).item(), "dB.")
print("dis: ", (z - x).abs().sum())
self.display_density(basis.operator, basis.inv_operator)
def test_transform(self, device, dtype):
M, N = 2, 32
K = 30
Bs = []
beta = []
bpg, bt = rand_bipartite(N // 2,
N - N // 2,
p=0.2,
dtype=dtype,
device=device,
return_partition=True)
Bs.append(bpg)
beta.append(bt)
mask = Bs[0].to_dense() != 0
for i in range(M - 1):
bpg, bt = rand_bipartite(
N // 2,
N - N // 2,
p=0.2,
dtype=dtype,
device=device,
return_partition=True,
)
dense = bpg.to_dense()
temp_mask = dense != 0
dense[mask] = 0
mask = temp_mask + mask
Bs.append(SparseTensor.from_dense(dense))
if i % 2 == 0:
beta.append(bt)
else:
beta.append(~bt)
beta = torch.stack(beta).T
qmf = QmfCore(bipartite_graphs=Bs, beta=beta, order=K)
x = torch.rand(N, dtype=dtype, device=device)
y = qmf.analyze(x)
assert y.shape == (2**M, N, 1)
z = qmf.synthesize(y)
assert z.shape == (2**M, N, 1)
assert (z.sum(0).squeeze() - x).abs().sum() != 0
z.squeeze_()
f_hat = z.sum(0)
dis = (f_hat - x).abs()
pf = snr(f_hat, x)
ppprint(dis, pf)
def test_transform(self, dtype, device, strategy, M, N):
ray.init(num_cpus=2, log_to_driver=False, ignore_reinit_error=True)
graph = rand_udg(N, device=device, dtype=dtype)
qmf = NumQmf(graph, strategy=strategy, level=M)
f = torch.rand(N, device=device, dtype=dtype)
y = qmf.analyze(f)
z = qmf.synthesize(y)
z.squeeze_()
f_hat = z.sum(0)
dis = (f_hat - f).abs()
pf = snr(f_hat, f).item()
print(f"\n|----- Strategy: {strategy:8s}, M={M}, "
f"Device: {str(device):5s}, Dtype: {str(dtype):6s} -----")
ppprint(dis, pf)
assert pf > 20
def test_transform(self, dtype, device, strategy, Ci):
N = 60
graph = rand_udg(N, device=device, dtype=dtype)
qmf = ColorQmf(graph, strategy=strategy, in_channels=1)
M = qmf.num_bgraph
f = torch.rand(N, Ci, device=device, dtype=dtype)
y = qmf.analyze(f)
z = qmf.synthesize(y)
z.squeeze_()
f_hat = z.sum(0)
dis = (f_hat - f).abs()
pf = snr(f_hat, f.squeeze_()).item()
print(f"\n|-- Strategy: {strategy:8s}, M={M}, "
f"Device: {str(device):5s}, Dtype: {str(dtype):6s} ---")
ppprint(dis, pf)
assert pf > 20
def test_transform(self, dtype, device, strategy, zero_dc):
N = 100
graph = rand_udg(N, device=device, dtype=dtype)
bio = ColorBiorth(graph, strategy=strategy)
M = bio.num_bgraph
f = torch.rand(N, device=bio.device, dtype=bio.dtype)
y = bio.analyze(f)
z = bio.synthesize(y)
z.squeeze_()
f_hat = z.sum(0)
dis = (f_hat - f).abs()
pf = snr(f_hat, f).item()
print(f"\n|----- Strategy: {strategy:8s}, M:{M:4d} "
f"Device: {str(device):5s}, Dtype: {str(dtype):6s}")
ppprint(dis, pf)
assert pf > 20
def test_one_level(self, dtype, device):
N1 = 6
N2 = 4
bptG, beta = rand_bipartite(N1,
N2,
0.5,
dtype=dtype,
return_partition=True)
basis = QmfOperator([bptG.to_scipy("csr")],
beta.view(-1, 1),
order=20,
device=device)
x = torch.ones(N1 + N2, 1, dtype=dtype, device=device)
y = basis.transform(x)
z = basis.inverse_transform(y)
print("\nsnr: ",
snr(z.permute(-1, -2), x.permute(-1, -2)).item(), "dB.")
print("dis: ", (z - x).abs().sum())
def test_one_level(self, dtype, device):
N1 = 60
N2 = 40
bptG, beta = rand_bipartite(N1,
N2,
0.2,
dtype=dtype,
return_partition=True)
basis = BiorthOperator([bptG.to_scipy("csr")],
beta.view(-1, 1),
k=2,
device=device)
x = torch.ones(N1 + N2, 1, dtype=dtype, device=device)
y = basis.transform(x)
z = basis.inverse_transform(y)
print("\nsnr: ",
snr(z.permute(-1, -2), x.permute(-1, -2)).item(), "dB.")
print("dis: ", (z - x).abs().sum())
self.display_density(basis.operator, basis.inv_operator)
def test_transform(self, dtype, device, strategy, M, part):
ray.init(num_cpus=2, log_to_driver=False, ignore_reinit_error=True)
N = 32 * 3
print(f"\n|----- Strategy: {strategy:8s}, M={M}, "
f"Device: {str(device):5s}, Dtype: {str(dtype):6s} ----")
kwargs = dict()
if strategy == "admm":
print("|admm-lbga part strategy: {}".format(part))
kwargs["part"] = part
graph = rand_udg(N, device=device, dtype=dtype)
bio = NumBiorth(graph, strategy=strategy, level=M, **kwargs)
f = torch.randn(N, device=device, dtype=dtype)
y = bio.analyze(f)
z = bio.synthesize(y)
z.squeeze_()
f_hat = z.sum(0)
dis = (f_hat - f).abs()
pf = snr(f_hat, f).item()
ppprint(dis, pf)
assert pf > 20