Traffic4cast-TLab

This repository is our solution to NeurIPS 2019 traffic4cast competition.

We ranked 4^th in the public leaderboard.

For more information, please read traffic4cast-report.pdf in this repository.

How to run the codes.

1. run process.py

   e.g. python process.py --city Moscow -i ./data/

   process.py: generate all useful features. In order to increase IO performance, each feature is a npy file.

2. run train_v17.py

   e.g. python train_v17.py --city Moscow -ch 2 --crop 0 -o ./

   train_v17.py: train models.

3. run test_v17.py

   e.g. python test_v17.py --city Moscow -ch 2 --crop 0 -o ./ -m ./

   test_v17.py: output prediction.

4. run gen_v17.py

   e.g. python gen_v17.py --city Moscow

   gen_v17.py: generate submission.