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

refactor: use LinearTransform instead of torch.nn.Linear

Browse Source
This commit is contained in:
Jensun Ravichandran 2021-06-21 22:52:22 +02:00 committed by Alexander Engelsberger
parent 71602bf38a
commit 72af03b991
1 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
prototorch/models/glvq.py
View File

@ -7,6 +7,7 @@ from ..core.competitions import wtac
from ..core.distances import lomega_distance, omega_distance, squared_euclidean_distance from ..core.distances import lomega_distance, omega_distance, squared_euclidean_distance
from ..core.initializers import EyeTransformInitializer from ..core.initializers import EyeTransformInitializer
from ..core.losses import glvq_loss, lvq1_loss, lvq21_loss from ..core.losses import glvq_loss, lvq1_loss, lvq21_loss
from ..core.transforms import LinearTransform
from ..nn.activations import get_activation from ..nn.activations import get_activation
from ..nn.wrappers import LambdaLayer, LossLayer from ..nn.wrappers import LambdaLayer, LossLayer
from .abstract import ImagePrototypesMixin, SupervisedPrototypeModel from .abstract import ImagePrototypesMixin, SupervisedPrototypeModel
@ -208,18 +209,22 @@ class SiameseGMLVQ(SiameseGLVQ):
super().__init__(hparams, **kwargs) super().__init__(hparams, **kwargs)
# Override the backbone # Override the backbone
self.backbone = torch.nn.Linear(self.hparams.input_dim, omega_initializer = kwargs.get("omega_initializer",
self.hparams.latent_dim, EyeTransformInitializer())
bias=False) self.backbone = LinearTransform(
self.hparams.input_dim,
self.hparams.output_dim,
initializer=omega_initializer,
)
@property @property
def omega_matrix(self): def omega_matrix(self):
return self.backbone.weight.detach().cpu() return self.backbone.weights
@property @property
def lambda_matrix(self): def lambda_matrix(self):
omega = self.backbone.weight # (latent_dim, input_dim) omega = self.backbone.weight # (input_dim, latent_dim)
lam = omega.T @ omega lam = omega @ omega.T
return lam.detach().cpu() return lam.detach().cpu()