prototorch_models/prototorch/models/neural_gas.py
Alexander Engelsberger 1c3613019b Update Examples to new initializer architecture.
Visualization still borken for some examples.
2021-05-06 14:10:09 +02:00

70 lines
2.2 KiB
Python

import torch
from prototorch.components import Components
from prototorch.components import initializers as cinit
from prototorch.functions.distances import euclidean_distance
from prototorch.modules.losses import NeuralGasEnergy
from .abstract import AbstractPrototypeModel
class EuclideanDistance(torch.nn.Module):
def forward(self, x, y):
return euclidean_distance(x, y)
class ConnectionTopology(torch.nn.Module):
def __init__(self, agelimit, num_prototypes):
super().__init__()
self.agelimit = agelimit
self.num_prototypes = num_prototypes
self.cmat = torch.zeros((self.num_prototypes, self.num_prototypes))
self.age = torch.zeros_like(self.cmat)
def forward(self, d):
order = torch.argsort(d, dim=1)
for element in order:
i0, i1 = element[0], element[1]
self.cmat[i0][i1] = 1
self.age[i0][i1] = 0
self.age[i0][self.cmat[i0] == 1] += 1
self.cmat[i0][self.age[i0] > self.agelimit] = 0
def extra_repr(self):
return f"agelimit: {self.agelimit}"
class NeuralGas(AbstractPrototypeModel):
def __init__(self, hparams, **kwargs):
super().__init__()
self.save_hyperparameters(hparams)
# Default Values
self.hparams.setdefault("input_dim", 2)
self.hparams.setdefault("agelimit", 10)
self.hparams.setdefault("lm", 1)
self.hparams.setdefault("prototype_initializer",
cinit.ZerosInitializer(self.hparams.input_dim))
self.proto_layer = Components(
self.hparams.num_prototypes,
initializer=self.hparams.prototype_initializer)
self.distance_layer = EuclideanDistance()
self.energy_layer = NeuralGasEnergy(lm=self.hparams.lm)
self.topology_layer = ConnectionTopology(
agelimit=self.hparams.agelimit,
num_prototypes=self.hparams.num_prototypes,
)
def training_step(self, train_batch, batch_idx):
x = train_batch[0]
protos = self.proto_layer()
d = self.distance_layer(x, protos)
cost, order = self.energy_layer(d)
self.topology_layer(d)
return cost