Skip to content

Commit

Permalink
add testing of IPFP and SM
Browse files Browse the repository at this point in the history
  • Loading branch information
@rogerwwww
rogerwwww committed Mar 16, 2022
1 parent 869ccd7 commit 3f56516
Show file tree
Hide file tree
Showing 2 changed files with 74 additions and 26 deletions.
1 change: 1 addition & 0 deletions tests/requirements.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,3 +4,4 @@ Pillow>=7.2.0
numpy>=1.18.5
easydict>=1.7
torch
tqdm
99 changes: 73 additions & 26 deletions tests/test_classic_solvers.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,8 +6,9 @@
import torch
import functools
import itertools
from tqdm import tqdm

# Some test utils function
# Some test utils functions
def data_from_numpy(*data):
return_list = []
for d in data:
Expand All @@ -23,14 +24,15 @@ def data_to_numpy(*data):


# The testing function
def test_classic_solvers():
graph_num_nodes = list(range(10, 30, 2))
node_feat_dim = 10
batch_size = len(graph_num_nodes)
def _test_classic_solver_on_isomorphic_graphs(graph_num_nodes, node_feat_dim, solver_func, matrix_params, backends):
assert 'edge_aff_fn' in matrix_params
assert 'node_aff_fn' in matrix_params
if backends[0] != 'pytorch':
backends.insert(0, 'pytorch') # force pytorch as the reference backend

backends = ['pytorch', 'numpy']
batch_size = len(graph_num_nodes)

# Test matching of isomorphic graphs
# Generate isomorphic graphs
pygm.BACKEND = 'pytorch'
X_gt = torch.zeros(batch_size, max(graph_num_nodes), max(graph_num_nodes))
A1 = torch.zeros(batch_size, max(graph_num_nodes), max(graph_num_nodes))
Expand All @@ -40,23 +42,27 @@ def test_classic_solvers():
for b, num_node in enumerate(graph_num_nodes):
X_gt[b, torch.arange(0, num_node, dtype=torch.int64), torch.randperm(num_node)] = 1
A1[b, :num_node, :num_node] = torch.rand(num_node, num_node)
A1_diagonal = torch.diagonal(A1[b])
A1_diagonal[:] = 0
torch.diagonal(A1[b])[:] = 0
A2[b] = torch.mm(torch.mm(X_gt[b].t(), A1[b]), X_gt[b])
F2[b] = torch.mm(X_gt[b], F1[b])
F2[b] = torch.mm(X_gt[b].t(), F1[b])
n1 = torch.tensor(graph_num_nodes, dtype=torch.int)
n2 = torch.tensor(graph_num_nodes, dtype=torch.int)

A1, A2, F1, F2, n1, n2 = data_to_numpy(A1, A2, F1, F2, n1, n2)
A1, A2, F1, F2, n1, n2, X_gt = data_to_numpy(A1, A2, F1, F2, n1, n2, X_gt)

# call the solver
total = 1
for val in matrix_params.values():
total *= len(val)
for values in tqdm(itertools.product(*matrix_params.values()), total=total):
aff_param_dict = {}
solver_param_dict = {}
for k, v in zip(matrix_params.keys(), values):
if k in ['node_aff_fn', 'edge_aff_fn']:
aff_param_dict[k] = v
else:
solver_param_dict[k] = v

# RRWM
for alpha, beta, edge_aff_func, node_aff_func in \
itertools.product(
[0.1, 0.5, 0.9], # alpha
[0.1, 1, 10], # beta
[functools.partial(pygm.utils.gaussian_aff_fn, sigma=1.), pygm.utils.inner_prod_aff_fn], # edge aff func
[functools.partial(pygm.utils.gaussian_aff_fn, sigma=.1), pygm.utils.inner_prod_aff_fn], # node aff func
):
last_K = None
last_X = None
for working_backend in backends:
Expand All @@ -65,15 +71,56 @@ def test_classic_solvers():
_conn1, _edge1, _ne1 = pygm.utils.dense_to_sparse(_A1)
_conn2, _edge2, _ne2 = pygm.utils.dense_to_sparse(_A2)

_K = pygm.utils.build_aff_mat(_F1, _edge1, _conn1, _F2, _edge2, _conn2, _n1, _ne1, _n2, _ne2, node_aff_func, edge_aff_func)
_K = pygm.utils.build_aff_mat(_F1, _edge1, _conn1, _F2, _edge2, _conn2, _n1, _ne1, _n2, _ne2,
**aff_param_dict)
if last_K is not None:
print(np.abs(pygm.utils.to_numpy(_K) - last_K).sum())
assert np.abs(pygm.utils.to_numpy(_K) - last_K).sum() < 0.1
assert np.abs(pygm.utils.to_numpy(_K) - last_K).sum() < 0.1, \
f"Incorrect affinity matrix for {working_backend}, " \
f"params: {';'.join([k + '=' + str(v) for k, v in aff_param_dict.items()])};" \
f"{';'.join([k + '=' + str(v) for k, v in solver_param_dict.items()])}"
last_K = pygm.utils.to_numpy(_K)
_X = pygm.rrwm(_K, _n1, _n2, alpha=alpha, beta=beta)
_X = solver_func(_K, _n1, _n2, **solver_param_dict)
if last_X is not None:
print(np.abs(pygm.utils.to_numpy(_X) - last_X).sum())
assert np.abs(pygm.utils.to_numpy(_X) - last_X).sum() < 1e-4
assert np.abs(pygm.utils.to_numpy(_X) - last_X).sum() < 1e-4, \
f"Incorrect GM solution for {working_backend}, " \
f"params: {';'.join([k + '=' + str(v) for k, v in aff_param_dict.items()])};" \
f"{';'.join([k + '=' + str(v) for k, v in solver_param_dict.items()])}"
last_X = pygm.utils.to_numpy(_X)

pygm.hungarian(_X, _n1, _n2)
accuracy = (pygm.utils.to_numpy(pygm.hungarian(_X, _n1, _n2)) * X_gt).sum() / X_gt.sum()
assert accuracy == 1, f"GM is inaccurate for {working_backend}, " \
f"params: {';'.join([k + '=' + str(v) for k, v in aff_param_dict.items()])};" \
f"{';'.join([k + '=' + str(v) for k, v in solver_param_dict.items()])}"


def test_rrwm():
_test_classic_solver_on_isomorphic_graphs(list(range(10, 30, 2)), 10, pygm.rrwm, {
'alpha': [0.1, 0.5, 0.9],
'beta': [0.1, 1, 10],
'sk_iter': [10, 20],
'max_iter': [20, 50],
'edge_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=1.), pygm.utils.inner_prod_aff_fn],
'node_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=.1), pygm.utils.inner_prod_aff_fn]
}, ['pytorch', 'numpy'])


def test_sm():
_test_classic_solver_on_isomorphic_graphs(list(range(10, 30, 2)), 10, pygm.sm, {
'max_iter': [10, 50, 100],
'edge_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=1.), pygm.utils.inner_prod_aff_fn],
'node_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=.1), pygm.utils.inner_prod_aff_fn]
}, ['pytorch', 'numpy'])


def test_ipfp():
_test_classic_solver_on_isomorphic_graphs(list(range(10, 30, 2)), 10, pygm.ipfp, {
'max_iter': [10, 50, 100],
'edge_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=1.), pygm.utils.inner_prod_aff_fn],
'node_aff_fn': [functools.partial(pygm.utils.gaussian_aff_fn, sigma=.1), pygm.utils.inner_prod_aff_fn]
}, ['pytorch', 'numpy'])


if __name__ == '__main__':
test_rrwm()
test_sm()
test_ipfp()

0 comments on commit 3f56516

Please sign in to comment.