/GCN_predict-Pytorch

Traffic flow predict. Implementation of graph convolutional network with PyTorch

Primary LanguagePython

GCN_predict-Pytorch

Traffic flow predict. Implementation of graph convolutional network(GCN,GAT,Chebnet) with PyTorch

Requirements:

​ - Pytorch

​ - Numpy

​ - Pandas

​ - Matplotlib

Example Dataset:

​ The datasets are collected by the Caltrans Performance Measurement System (PEMS-04)

​ Numbers:307 detectors

​ Date:Jan to Feb in 2018 (2018.1.1——2018.2.28)

​ Features:flow, occupy, speed.

Exploring data analysis:

​ 1.there is three features:flow,occupy and speed.First, we conduct a visual analysis of data distribution

​ 2.run code: python data_view.py

​ 3.Every node(detector) has three fetures,but two features data distribution are basically stationary, so we only take the first dimension features.

Read dataset:

​ In the traffic_dataset.py file,the get_adjacent_matrix and get_flow_data functions are to read adjacent matrix and flow data.

Model training:

​ In the traffic_preditcion.py,there are three graph convolution neural network models:GCN,ChenNET and GAT.Correspondingly, you only need to modify the 45th line of code in this file, and then observe the different results of model training.

​ python traffic_preditcion.py