[IEEE T-ITS Submission for Regular Paper] Multi-faceted Route Representation Learning for Travel Time Estimation

This is a PyTorch implementation of Multi-faceted Route Representation Learning for Travel Time Estimation (MulT-TTE).

Requirements

pip install -r requirements.txt

Data

We conduct our experiments on two real-world taxi trajectory datasets and their corresponding road networks, including Chengdu and Porto. And the Map Matching Algorithm with corresponding road network information are referenced here.

The sampled data and corresponding network information are stored in the following directory:

• MulT-TTE/data/chengdu/...
• MulT-TTE/data/chengdu/network-chengdu/...

Run

The command of processing is listed bellow:

(base) PS \MulT-TTE> python .\main.py --help
usage: main.py [-h] [-m {MulT_TTE}] [-M {train,resume,test}]
               [-d {chengdu_MulT_TTE,porto_MulT_TTE}] [-i IDENTIFY]
               [-D DEVICE] [-o {Adam}] [-c {rmse,mse,mape,mae,smoothL1}]
               [-cl LOSS_VAL] [-e EPOCHS] [-b BETA] [-l LR] [-w WEIGHT_DECAY]
               [-p PATIENCE] [-r MASK_RATE]

optional arguments:
  -h, --help            show this help message and exit
  -m {MulT_TTE}, --model {MulT_TTE}
                        input the model name
  -M {train,resume,test}, --mode {train,resume,test}
                        input the process mode
  -d {chengdu_MulT_TTE,porto_MulT_TTE}, --dataset {chengdu_MulT_TTE,porto_MulT_TTE}
                        input the dataset name
  -i IDENTIFY, --identify IDENTIFY
                        input the specific identification information
  -D DEVICE, --device DEVICE
                        input the chosen device
  -o {Adam}, --optim {Adam}
                        input the chosen optimization function
  -c {rmse,mse,mape,mae,smoothL1}, --loss {rmse,mse,mape,mae,smoothL1}
                        input the chosen loss function
  -cl LOSS_VAL, --loss_val LOSS_VAL
                        intput the specific parameter for smoothL1
  -e EPOCHS, --epochs EPOCHS
                        input the max epochs
  -b BETA, --beta BETA  intput the learning preference between MSG and TTE
                        (the bigger the value, the more preference for TTE.)
  -l LR, --lr LR        intput the initial learning rate
  -w WEIGHT_DECAY, --weight_decay WEIGHT_DECAY
                        intput the weight decay of optimization
  -p PATIENCE, --patience PATIENCE
                        intput the max iteration times of early stop
  -r MASK_RATE, --mask_rate MASK_RATE
                        intput the mask rate of segments in a trajectory

For example, you can train the model through the following commands:

python .\main.py --model MulT_TTE --dataset chengdu_MulT_TTE --identify example_0 --device cuda:0   

python .\main.py --model MulT_TTE --dataset porto_MulT_TTE --identify example_1 --device cuda:1   

Cite (coming soon)

If you find the paper useful, please cite as following:

@inproceedings{,
  title={},
  author={},
  booktitle={},
  year={},
  organization={}
}