julius/prototorch
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ML_omega_distance: allow for 2 or more omegas and masks

Browse Source
This commit is contained in:
julius
2023-11-07 16:45:05 +01:00
parent 3a5a2bb473
commit f35a08a070
1 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
prototorch/core/distances.py
View File

@@ -73,10 +73,14 @@ def omega_distance(x, y, omega):
return distances return distances
def ML_omega_distance(x, y, omega_0, omega_1, mask_0, mask_1): def ML_omega_distance(x, y, omegas, masks):
"""Multi-Layer Omega distance.""" """Multi-Layer Omega distance."""
x, y = (arr.view(arr.size(0), -1) for arr in (x, y)) x, y = (arr.view(arr.size(0), -1) for arr in (x, y))
omega = (omega_0 * mask_0) @ (omega_1 * mask_1) # omega = (omega_0 * mask_0) @ (omega_1 * mask_1)
omegas = [torch.mul(_omega, _mask) for _omega, _mask in zip(omegas, masks)]
omega = omegas[0] @ omegas[1]
for _omega in omegas[2:]:
omega = omega @ _omega
projected_x = x @ omega projected_x = x @ omega
projected_y = y @ omega projected_y = y @ omega
distances = squared_euclidean_distance(projected_x, projected_y) distances = squared_euclidean_distance(projected_x, projected_y)