Skip to content
PyTorch implementation for the paper Classification from Positive, Unlabeled and Biased Negative Data.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
cifar10
mnist
newsgroups
params
.gitignore
README.md
pu_biased_n.py
settings.py
training.py
utils.py

README.md

PUbiasedN

PyTorch implementation for experiments in the paper Classification from Positive, Unlabeled and Biased Negative Data.

Requirements

  1. Python >= 3.6
  2. PyTorch >= 0.4.0, scikit-learn, NumPy
  3. yaml to load parameters
  4. nltk, allennlp, h5py to prepare the 20newsgroups ELMO embedding

Usage

The file pu_biased_n.py allows to reproduce most of the experimental results described in the paper:

python(3) pu_biased_n.py --dataset [dataset] --params-path [parameter-path] --random-seed [random-seed]

where dataset is either mnist, cifar10 or newsgroups and parameter-path is a yml file containing the hyperparameters of the experiment. The hyperparameter files used for the results shown in Table 1 can be found under the params/ directory.

You can’t perform that action at this time.