/ST-ResNet-Pytorch

Deep Spatio-Temporal Residual Networks for Mobile Location Data Prediction (instead of original Crowd Flow Prediction) - Modified ST-ResNet

Primary LanguagePython

shadowbourne's Fork:

This is a modified fork of the original repo that instead allows forecasting of per square population density rather than 2 way flows within the grids. Used to test network's capabilities of forecasting mobile location data.

ST-ResNet in PyTorch

Demo code snippets (with dataset BikeNYC) of Deep Spatio-Temporal Residual Networks (ST-ResNet) from the paper "Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction".

Prerequisites

  • Python 3.6
  • Pytorch 1.0

Usage

This is only demo code snippets for reference.

Reference

Zhang, Junbo, Yu Zheng, and Dekang Qi. "Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction." AAAI. 2017. https://arxiv.org/pdf/1610.00081.pdf

Refered implementations

Keras: https://github.com/lucktroy/DeepST

Tensorflow: https://github.com/snehasinghania/STResNet