RDP

Codes for IJCAI2020 paper "Unsupervised Representation Learning by Predicting Random Distances” https://arxiv.org/abs/1912.12186

Installation

The repo is tested on Ubuntu 16.04, Python 3.5.2, PyTorch 1.1.0 and Sklearn 0.21.1.

Anomaly Detection

Data Preparation

Some of example datasets are put in ./data folder due to the large file size limitation. You may downloaded them from the urls listed in the paper appendix.

Train

If you are under Dev mode (tweak it in train.py), just run

python train.py

If you are under Server mode, the following scripts can be used to help you run experiments in batch

python train.py [data/csv_file] [save_path] > [output_log] 2>&1 &

e.g.

python train.py data/apascal.csv save_model/apascal/ > logs/apascal.log 2>&1 &
...

Test

If you are under Dev mode, just run

python test.py

If you are under Server mode, the following scripts can be used to help you run experiments in batch

python test.py [data/csv_file] [load_path] [tree_depth] > [output_log] 2>&1 &

e.g.

python test.py data/apascal.csv save_model/apascal/ 8 1 > logs/apascal_l8_test.log 2>&1 &
...

Clustering

Train

If you are under Dev mode (tweak it in train.py), just run

python train.py

If you are under Server mode, the following scripts can be used to help you run experiments in batch

python train.py [data] [batch_size] [out_c] > [output_log] 2>&1 &

e.g.

python train.py r8 192 512 > logs/r8_b192_[512].log 2>&1 &
...

Test

If you are under Dev mode, just run

python test.py

If you are under Server mode, the following scripts can be used to help you run experiments in batch

python test.py [data] [batch_size] [out_c] > [output_log] 2>&1 &

e.g.

python test.py r8 192 512 > logs/r8_b192_[512]_test.log 2>&1 &
...