Loading...
Close
XCurveXCurve
Sidebar

Search tutorials

Get Started

Documentation

Example Projects

XCurve.Losses.AUPRC

The following functions build the class to compute the List-stable AUPRC loss: 
  CLASS XCurve.AUPRC.RetrievalDataset(data_dir, list_dir, subset, input_size, batchsize, num_sample_per_id, normal_mean=[0.485, 0.456, 0.406], normal_std=[0.229, 0.224, 0.225], split='train')     [SOURCE]
Parameters:
tau1 (float)

Control the surrogate loss of pos-neg pairs. See \tau_1 in Eq.(7).

tau2 (float)

Control the surrogate loss of pos-pos pairs. See \tau_2 in Eq.(7).

beta (float)

Control the exponential moving average. See \beta in Eq.(10).

prior_mul (float)

Imbalance ratio of the id with most positive examples.

num_sample_per_id (int)

Number of examples for each id.

var_reg_weight_pos (float)

Weight of the variance regular term w.r.t. positive examples.

var_reg_weight_neg (float)

Weight of the variance regular term w.r.t. negative examples.

Example: 
  import torch.nn.functional as F
  from XCurve.AUPRC import (ListStableAUPRC, DefaultLossCfg, \
      RetrievalDataset, DefaultInatDatasetCfg)
  
  dataset = RetrievalDataset(**DefaultInatDatasetCfg, split='train')
  criterion = ListStableAUPRC(**DefaultLossCfg)
  criterion.update_cnt_per_id(dataset.get_cnt_per_id())
  feats = F.normalize(torch.randn((16, 128)).cuda(), dim=1, p=2)
  targets = torch.tensor([5, 5, 5, 5, 3, 3, 3, 3, 1, 1, 1, 1, 9, 9, 9, 9]).cuda()
  loss = criterion(feats, targets)
  print(loss.item())