A Dynamic Embedding Method for Passenger Flow Estimation

Introduction

This repository contains the code for replicating results from

• A Dynamic Embedding Method for Passenger Flow Estimation
• Wei-Yi Chung; Yen-Nan Ho; Yu-Hsuan Wu; Jheng-Long Wu
• In 2021 10th International Congress on Advanced Applied Informatics (IIAI-AAI)
• Slide | Competitive Paper Award

Flow Chart

Getting Started

• Clone the repo and get in to project cd ./Project
• Build a new virtual environment
• Install python3 requirements: pip install -r requirements.txt
• Run cd ./model to the model folder
• Use your own dataset (optional)
  • Construct the longtitude and latitude information of station in to mrt_vd.csv
  • Adjustment the format of passanger flow data to the demo input format
  • Run python distance_matrix to create the distance matrix
• Build training data, run python Data_preparing_threeloss to generate the training data
• Train your own models of pretrained stage
• repace the station feature from Node2Vec to BERT output in GMAN

Training Insturctions

• Experiment configurations are found in ./model/BERT_three_loss/run_train.sh
• Training: sh run_train.sh
• Results model and logs are stored in the output directory under BERT_three_loss.
• Evaluation: python ./BERT_three_loss/predict_embedding.py