Dual Sequential Network for Temporal Sets Prediction

Introduction

DSNTSP (Dual Sequential Network for Temporal Sets Prediction) is a novel model used for temporal sets prediction prediction problem.

Please refer to our SIGIR 2020 paper "Dual Sequential Network for Temporal Sets Prediction" for more details.

Project Architecture

The descriptions of principal files in this project are explained as follows:

• config/: containing JSON format configuration files.
• data/: containing JSON format datasets and DataLoader implementations for our models.
• learner/: codes for the learner and metric definitions for our models.
• model/: codes for our temporal sets prediction models.
• paper/: containing our published paper.
• registry/: codes for registering models, only the registered model classes can be recognized by the argument parser.
• run.py: script used for training and testing models.
• requirements.txt: containing a list of dependencies for conda.

How to use:

Train the model:

python run.py --mode=train --config=./config/taobao_buy/dsntsp.json --cuda=0

Test the model:

python run.py --mode=test --config=./config/taobao_buy/dsntsp.json --cuda=0

You can modify the value of the attribute best_epoch in the JSON format configuration file in the config/ to choose which trained model to test.