ml-for-graph-analysis / traffic-prediction

Project - Traffic forecasting by spatio-temporal graph model

Traffic-Prediction

This is the pytorch implementation for class project of AIGS703I(인공지능특론:그래프분석을 위한 기계학습)

This problem is challenging due to the complicated and dynamic spatio-temporal dependencies between different regions in the road network.

Models are mainly based on GraphWaveNet and refer to other state-of-date models.

Dataset: Metr-LA, PeMS-Bay

EDA

• Role analysis of traffic graph data
• Community analysis of traffic graph data(Louvain.ipynb)

Train Commands

Model 1.2

python train.py --data data/METR-LA --savehorizon True --sheetname Model2_METR-LA_60min --adjdata data/sensor_graph/adj_mx.pkl --batch_size 32 --adjtype doubletransition

python train.py --data data/PEMS-BAY --savehorizon True --sheetname Model2_PEMS-BAY_60min --num_nodes 325 --adjdata data/sensor_graph/adj_mx_bay.pkl batch_size 32 --nhid 16 --adjtype doubletransition

Model 2

python train.py --data data/METR-LA --savehorizon True --sheetname Model1_METR-LA_60min --adjdata data/sensor_graph/adj_mx.pkl --nhid 16 --batch_size 32 --adjtype transition

python train.py --data data/PEMS-BAY --savehorizon True --sheetname Model1_PEMS-BAY_60min --num_nodes 325 --adjdata data/sensor_graph/adj_mx_bay.pkl --nhid 16 --batch_size 32 --adjtype transition

Model 3

--batch_size 32

Model 4

--batch_size 32