def test_gen_knn_hg():
X = np.random.rand(50, 100)
hg = gen_knn_hg(X, n_neighbors=5)
assert hg.num_nodes() == 50
assert hg.num_edges() == 50
assert hg.incident_matrix().nnz == 300
def test_fuse_multi_hg():
X = np.random.rand(200, 10)
knn_hg = gen_knn_hg(X, n_neighbors=5)
clu_hg = gen_clustering_hg(X, n_clusters=20)
hg = concat_multi_hg([knn_hg, clu_hg])
assert hg.num_nodes() == knn_hg.num_nodes()
assert hg.num_nodes() == clu_hg.num_nodes()
assert hg.num_edges() == knn_hg.num_edges() + clu_hg.num_edges()
def main():
print_log("loading data")
X_train, X_test, y_train, y_test = load_myocardium([3])
X = np.concatenate([X_train, X_test], axis=0)
y = np.concatenate(
[y_train.reshape(-1), -1 * np.ones_like(y_test).reshape(-1)])
print_log("generating hypergraph")
# grid hg
grid_hg_s = [gen_grid_neigh_hg(X[i].shape) for i in range(X.shape[0])]
grid_hg = fuse_mutli_sub_hg(grid_hg_s)
# knn hg
X_patch_ft = np.stack([
gather_patch_ft(X[i][:, :, np.newaxis], (5, 5))
for i in range(X.shape[0])
])
knn_hg = gen_knn_hg(X_patch_ft.reshape(-1, X_patch_ft.shape[-1]),
n_neighbors=7)
# concatfeat hg
concat_hg = concat_multi_hg([grid_hg, knn_hg])
print_log("learning on hypergraph")
y_predict = trans_infer(concat_hg, y, 100)
print_log("iou: {}".format(iou_socre(y_predict, y_test.reshape(-1))))
print_log("postprocessing")
y_postpro = postprocess(y_predict.reshape(y_test.shape))
print_log("iou(postprocess): {}".format(
iou_socre(y_postpro.reshape(-1), y_test.reshape(-1))))
# visualize
print_log("visualizing")
fig = plt.figure()
ax1 = fig.add_subplot(2, 2, 1)
ax1.imshow(X_test.squeeze())
ax1.title.set_text('X_test')
ax1.axis('off')
ax2 = fig.add_subplot(2, 2, 2)
ax2.imshow(y_test.squeeze())
ax2.title.set_text('y_test')
ax2.axis('off')
ax3 = fig.add_subplot(2, 2, 3)
ax3.imshow(y_predict.reshape(y_test.squeeze().shape))
ax3.title.set_text('y_predict')
ax3.axis('off')
ax4 = fig.add_subplot(2, 2, 4)
ax4.imshow(y_postpro.reshape(y_test.squeeze().shape))
ax4.title.set_text('y_postpro')
ax4.axis('off')
plt.show()
def main():
print_log("loading data")
X_train, X_test, y_train, y_test = load_MSRGesture3D()
X = np.concatenate([X_train, X_test])
y = np.concatenate([y_train, -1 * np.ones_like(y_test)])
print_log("generating hypergraph")
hg = gen_knn_hg(X, n_neighbors=4)
print_log("learning on hypergraph")
y_predict = trans_infer(hg, y, lbd=100)
print_log("accuracy: {}".format(accuracy_score(y_test, y_predict)))
def main():
print_log("loading data")
X_train, X_test, y_train, y_test = load_modelnet(selected_mod=(0, 1))
X = [np.vstack((X_train[imod], X_test[imod])) for imod in range(len(X_train))]
y = np.concatenate((y_train, -1 * np.ones_like(y_test)))
print_log("generating hypergraph")
hg_list = [
gen_knn_hg(X[imod], n_neighbors=10)
for imod in range(len(X_train))
]
print_log("learning on hypergraph")
y_predict = multi_hg_trans_infer(hg_list, y, lbd=100)
print_log("accuracy: {}".format(accuracy_score(y_test, y_predict)))