Loading...
Close
XCurveXCurve
Sidebar

Search tutorials

Get Started

Documentation

Example Projects

XCurve.Losses.AdversarialAUROC

The AdversarialAUROC consists of two parts: AdvAUROCLoss, and RegAdvAUROCLoss.

AdvAUROCLoss

The following class supports the computation of AdvAUROCLoss: 
  CLASS AdvAUROCLoss (imratio=None, a=None, b=None, alpha=None)     [SOURCE]
Parameters:
imratio (float)

The imbalance ration in your dataset. It must be precisely estimated based on your training set, i.e., $\frac{\#Pos.Num}{\#Tot.Num}$.

a, b, alpha (float)

The initial values for a, b, $\alpha$ in instance-wise AUROC loss (Default=0.0).

The calculation of AdvAUROC Loss should use the following optimizer, where the parameters are the same as above: 

  CLASS AdvAUCOptimizer (params, a, b, alpha, lr=required, momentum=0.9, dampening=0, weight_decay=0, nesterov=False)

Example: 
  import torch
  from XCurve.AUROC.losses import AdvAUROCLoss
  
  feats = torch.rand(16).cuda()
  targets = torch.tensor([0,1,1,0,1,0,1,1,0,0,1,0,1,1,1,0]).cuda()
  criterion_AdvAUROC = AdvAUROCLoss()
  loss_AdvAUROC = criterion_AdvAUROC(feats, targets)
  print(loss_AdvAUROC.item())

RegAdvAUROCLoss

  CLASS RegAdvAUROCLoss (imratio=None, a=None, b=None, alpha=None, lambda1=None, lambda2=None)     [SOURCE]
Parameters:
imratio (float)

The imbalance ration in your dataset. It must be precisely estimated based on your training set, i.e., $\frac{\#Pos.Num}{\#Tot.Num}$.

a, b, alpha (float)

The initial values for a, b, $\alpha$ in instance-wise AUROC loss (Default=0.0).

lambda1, lambda2 (float)

The initial coefficient values for regularization (Default=1.0).

The calculation of RegAdvAUROCLoss Loss should use the following optimizer, where the parameters are the same as above: 

  CLASS RegAdvAUCOptimizer (params, a, b, alpha, lambda1, lambda2, c1=1, c2=1, gamma=1, lamda=1, lr=required)

Example: 
  import torch
  from XCurve.AUROC.losses import RegAdvAUROCLoss
  
  feats = torch.rand(16).cuda()
  targets = torch.tensor([0,1,1,0,1,0,1,1,0,0,1,0,1,1,1,0]).cuda()
  criterion_RegAdvAUROC = RegAdvAUROCLoss()
  loss_RegAdvAUROC = criterion_RegAdvAUROC(feats, targets)
  print(loss_RegAdvAUROC.item())