DTS-ERA

Intro

This repository contains the source code implementation for our submission paper Deep Temporal Sets with Evidential Reinforced Attentions for Unique Behavioral Pattern Discovery The python file RAM_ASD_TD_mean_train_test.py are the core framework codes to run this project under different settings, by integrating training code and evaluation code all in one file.

Installation

Generate conda environment using ADS.yml file, then run the RAM_ASD_TD.py to try different RAM model variants (one for mean, one for concatenation). For recovering main results in the paper, just run RAM_ASD_TD_mean_train_test.py About how to build conda environment using .yml files. Please refer to this link.

Dependencies

• See ADS.yml file.

Dataset

• See data/maze for processed version of Maze dataset. For the original dataset of Maze, Wording Scanning and Coloring datasets, please ask authors for permission.

Pre-trained Models

• See results/checkpoint

Baselines Result

For baseline result comparison, users can easily reproduce them by running the RAM_ASD_TD_mean_train_test.py and replacing the main model RAM_ASD_TD_mean with any baselines. (e.g. GRU-LSTM, XCM, etc)