Welcome to the official repo of the TransTTE model -- transformer-based travel time estimation algorithm. Here we present the source code for accepted PKDD'22 paper "Logistics, Graphs, and Transformers: Towards improving Travel Time Estimation".

Natalia Semenova, Artyom Sosedka, Vladislav Tishin, Vladislav Zamkovoy, Vadim Porvatov

You can access inference of our model at transtte.online

arXiv PDF: https://arxiv.org/abs/2207.05835

Prerequisites

It is possible to run Visual Tool and Graphormer locally, but we strongly recomend to use provided Dockerfiles

Backend:

fastapi==0.67.0
pydantic==1.8.2
uvicorn==0.14.0
pandas==1.3.4
sklearn==0.0
python-igraph==0.9.6
loguru==0.5.3
torch==1.9.1+cu111

Model:

lmdb==1.3.0
torch-scatter==2.0.9
torch-sparse==0.6.12
torch-geometric==1.7.2
tensorboardX==2.4.1
ogb==1.3.2
rdkit-pypi==2021.9.3
dgl==0.7.2
igraph==0.9.10
setuptools==0.1.96
numpy==1.20.3

Additionally, you need to install fairseq to fit graphormer.

Local test

Prepare repository, data and weights:

• Clone repository: git clone https://github.com/Vloods/TransTTE_demo
• Download backend data and put it in backend/app/data
• Download graphormer models and put it in graphormer/app/models
• Download geo-datasets to fit graphormer (check the section below)

How to run Visual Tool:

• Install and run Docker
• Build Docker image with backend/Dockerfile via run command docker build . -t visual" in terminal
• Run Docker container via run docker run --rm -it -p 80:80 visual in terminal
• Go to http://127.0.0.1:80/

How to run Graphormer:

• Install and run Docker
• Build Docker image with graphormer/Dockerfile via run docker build . -t graphormer in terminal
• Run Docker container via run docker run --rm -it -p 80:80 graphormer in terminal
• Run python script to get times for each edge. Visual tool use this times in order to find the shortest way between two points.

Python script:

  r = requests.post('http://0.0.0.0:80/get_weights', headers = {'Content-Type': 'application/json'})
  weights_dict = r.json()

Datasets

We provide two datasets corresponding to the cities of Abakan and Omsk. For each of these datasets, there are two types of target values -- real travel time (considered in this study) and real length of trip.

Road network	Trips
Abakan Omsk Nodes 65524 231688 Edges 340012 1149492 Clustering 0.5278 0.53 Usage median 12 8	Abakan Omsk Trips number 119986 120000 Coverage 0.535 0.392 Average time 433.61 622.67 Average length 3656.34 4268.72

Provided data could be used for research purposes only. If you want to incorporate it in your study, please send request to semenova.bnl@gmail.com.

License

Established code released as open-source software under the MIT license.

Contact us

If you have some questions about the code, you are welcome to open an issue, I will respond to that as soon as possible.

Citation

@InProceedings{10.1007/978-3-031-26422-1_36,
author="Semenova, Natalia
and Porvatov, Vadim
and Tishin, Vladislav
and Sosedka, Artyom
and Zamkovoy, Vladislav",
title="Logistics, Graphs, and Transformers: Towards Improving Travel Time Estimation",
booktitle="Machine Learning and Knowledge Discovery in Databases",
year="2023",
publisher="Springer Nature Switzerland",
pages="589--593"
}