# -*- coding: utf-8 -*-
""" Deep Multiclass Graph Description (DMGD)"""
# Author: Kay Liu <zliu234@uic.edu>
# License: BSD 2 clause
import torch
import warnings
from torch_geometric.nn import MLP, GCN
from . import DeepDetector
from ..nn import DMGDBase
[docs]
class DMGD(DeepDetector):
"""
Deep Multiclass Graph Description
DMGD is a support vector based multiclass outlier detector. Its
backbone is an autoencoder that reconstructs the adjacency matrix
of the graph with MSE loss and homophily loss. It applies k-means
to cluster the nodes embedding and then uses support vector to
detect outliers.
See :cite:`bandyopadhyay2020integrating` for details.
Parameters
----------
hid_dim : int, optional
Hidden dimension of model. Default: ``64``.
num_layers : int, optional
Total number of layers in model. Default: ``2``.
dropout : float, optional
Dropout rate. Default: ``0.``.
weight_decay : float, optional
Weight decay (L2 penalty). Default: ``0.``.
act : callable activation function or None, optional
Activation function if not None.
Default: ``torch.nn.functional.relu``.
backbone : torch.nn.Module
The backbone of the deep detector implemented in PyG.
Default: ``torch_geometric.nn.GCN``.
contamination : float, optional
The amount of contamination of the dataset in (0., 0.5], i.e.,
the proportion of outliers in the dataset. Used when fitting to
define the threshold on the decision function. Default: ``0.1``.
lr : float, optional
Learning rate. Default: ``0.004``.
epoch : int, optional
Maximum number of training epoch. Default: ``100``.
gpu : int
GPU Index, -1 for using CPU. Default: ``-1``.
batch_size : int, optional
Minibatch size, 0 for full batch training. Default: ``0``.
num_neigh : int, optional
Number of neighbors in sampling, -1 for all neighbors.
Default: ``-1``.
alpha : float, optional
Weight of the radius loss. Default: ``1``.
beta : float, optional
Weight of the reconstruction loss. Default: ``1``.
gamma : float, optional
Weight of the homophily loss. Default: ``1``.
warmup : int, optional
The number of epochs for warm-up training. Default: ``2``.
k : int, optional
The number of clusters. Default: ``2``.
verbose : int, optional
Verbosity mode. Range in [0, 3]. Larger value for printing out
more log information. Default: ``0``.
save_emb : bool, optional
Whether to save the embedding. Default: ``False``.
compile_model : bool, optional
Whether to compile the model with ``torch_geometric.compile``.
Default: ``False``.
**kwargs
Other parameters for the backbone model.
Attributes
----------
decision_score_ : torch.Tensor
The outlier scores of the training data. Outliers tend to have
higher scores. This value is available once the detector is
fitted.
threshold_ : float
The threshold is based on ``contamination``. It is the
:math:`N \\times` ``contamination`` most abnormal samples in
``decision_score_``. The threshold is calculated for generating
binary outlier labels.
label_ : torch.Tensor
The binary labels of the training data. 0 stands for inliers
and 1 for outliers. It is generated by applying
``threshold_`` on ``decision_score_``.
emb : torch.Tensor or tuple of torch.Tensor or None
The learned node hidden embeddings of shape
:math:`N \\times` ``hid_dim``. Only available when ``save_emb``
is ``True``. When the detector has not been fitted, ``emb`` is
``None``. When the detector has multiple embeddings,
``emb`` is a tuple of torch.Tensor.
"""
def __init__(self,
hid_dim=64,
num_layers=2,
dropout=0.,
weight_decay=0.,
act=torch.nn.functional.relu,
backbone=GCN,
contamination=0.1,
lr=4e-3,
epoch=100,
gpu=-1,
batch_size=0,
num_neigh=-1,
alpha=1,
beta=1,
gamma=1,
warmup=2,
k=2,
verbose=0,
save_emb=False,
compile_model=False,
**kwargs):
if num_neigh != 0 and backbone == MLP:
warnings.warn('MLP does not use neighbor information.')
num_neigh = 0
self.alpha = alpha
self.beta = beta
self.gamma = gamma
self.warmup = warmup
self.k = k
super(DMGD, self).__init__(hid_dim=hid_dim,
num_layers=num_layers,
dropout=dropout,
weight_decay=weight_decay,
act=act,
backbone=backbone,
contamination=contamination,
lr=lr,
epoch=epoch,
gpu=gpu,
batch_size=batch_size,
num_neigh=num_neigh,
verbose=verbose,
save_emb=save_emb,
compile_model=compile_model,
**kwargs)
def process_graph(self, data):
DMGDBase.process_graph(data)
def init_model(self, **kwargs):
if self.save_emb:
self.emb = torch.zeros(self.num_nodes,
self.hid_dim)
return DMGDBase(in_dim=self.in_dim,
hid_dim=self.hid_dim,
num_layers=self.num_layers,
dropout=self.dropout,
act=self.act,
backbone=self.backbone,
alpha=self.alpha,
beta=self.beta,
gamma=self.gamma,
warmup=self.warmup,
k=self.k,
**kwargs).to(self.device)
def forward_model(self, data):
batch_size = data.batch_size
x = data.x.to(self.device)
edge_index = data.edge_index.to(self.device)
x_, nd, emb = self.model(x, edge_index)
loss, score = self.model.loss_func(x[:batch_size],
x_[:batch_size],
nd[:batch_size],
emb[:batch_size])
return loss, score.detach().cpu()
def decision_function(self, data, label=None):
if data is not None:
warnings.warn("This detector is transductive only. "
"Training from scratch with the input data.")
self.fit(data, label)
return self.decision_score_