def __init__(self,
cluster_number: int,
embedding_dimension: int,
hidden_dimension: int,
ae: torch.nn.Module,
alpha: float = 1.0):
"""
Module which holds all the moving parts of the DEC algorithm, as described in
Xie/Girshick/Farhadi; this includes the AutoEncoder stage and the ClusterAssignment stage.
:param cluster_number: number of clusters
:param embedding_dimension: embedding dimension, input to the encoder
:param hidden_dimension: hidden dimension, output of the encoder
:param ae: autoencoder to use, must have .encoder attribute
:param alpha: parameter representing the degrees of freedom in the t-distribution, default 1.0
"""
super(DEC, self).__init__()
self.ae = ae # AutoEncoder stage
if not hasattr(ae, 'encoder'):
raise ValueError('Autoencoder must have a .encoder attribute.')
self.embedding_dimension = embedding_dimension
self.hidden_dimension = hidden_dimension
self.cluster_number = cluster_number
self.alpha = alpha
self.assignment = ClusterAssignment(cluster_number,
self.hidden_dimension, alpha)
def setUpClass(cls):
cls.ca = ClusterAssignment(
cluster_number=2,
embedding_dimension=2,
cluster_centers=torch.Tensor([
[-1, -1],
[1, 1]
]).float()
)
def __init__(self,
cluster_number: int,
hidden_dimension: int,
encoder: torch.nn.Module,
alpha: float = 1.0):
"""
Module which holds all the moving parts of the DEC algorithm, as described in
Xie/Girshick/Farhadi; this includes the AutoEncoder stage and the ClusterAssignment stage.
:param cluster_number: number of clusters
:param hidden_dimension: hidden dimension, output of the encoder
:param encoder: encoder to use
:param alpha: parameter representing the degrees of freedom in the t-distribution, default 1.0
"""
super(DEC, self).__init__()
self.encoder = encoder
self.hidden_dimension = hidden_dimension
self.cluster_number = cluster_number
self.alpha = alpha
self.assignment = ClusterAssignment(cluster_number, self.hidden_dimension, alpha)
def setUpClass(cls):
cls.ca = ClusterAssignment(2,
2,
cluster_centers=torch.Tensor([[-1, -1],
[1,
1]]).float())