[ECML PKDD 2023]This is a Pytorch and DGL implementation of the following paper : "FDTI: Fine-grained Deep Traffic Inference with Roadnet-enriched Graph".
Please download the Manhattan data from https://drive.google.com/file/d/1TxVluhAEU3oFhlzoXq6FxmP7TTJjOuZG/view?usp=sharing and unzip it in ./data
folder
The Root is described as below
ROOT
+-- data
| +-- manhattan_train.dgl
| +-- manhattan_val.dgl
| +-- ...
+-- outputs
| +-- ...
+-- model.py
+-- test.py
+-- train.py
+-- utils.py
data
the dataset folder. Here it contains Manhattan dataset.outputs
contains the training log and model file. Each time runtrain.py
to launch a new training process, it will automatically create a folder to store the training log and model file.train.py
python script of training FDTI .test.py
python script of testing the model.utils.py
some useful functionmodel.py
contains the GNN module.
python train.py --setting manhattan
After finish training, a process for evaluation will be automatically created.
python test.py --setting manhattan --test_setting manhattan --test_folder $TRAIN_FOLDER
$TRAIN_FOLDER is the result folder name in ./outputs/manhattan/ which is created based on the time that training process begins.