huiminren / mtrajrec

Map-Constrained Trajectory Recovery (KDD'21)

MTrajRec

About

Source code of the KDD'21: MTrajRec: Map-Constrained Trajectory Recovery via Seq2Seq Multi-task Learning

Requirements

• Python==3.6
• pytorch==1.7.1
• rtree==0.9.4
• GDAL==2.3.3
• networkx==2.3

Usage

Installation

Clone this repo:

git clone [email protected]:huiminren/MTrajRec.git
cd MTrajRec

Running

python multi_main.py

Dataset

We provide sample data under data/.

Please note that the sample data is generated with the structure as the original data. For the data preprocessing, please refer to tptk.

Acknowledge

Thanks to Sijie to support data preprocessing.

Citation

If you find this repo useful and would like to cite it, citing our paper as the following will be really appropriate:

@inproceedings{ren2021mtrajrec,
  title={MTrajRec: Map-Constrained Trajectory Recovery via Seq2Seq Multi-task Learning},
  author={Ren, Huimin and Ruan, Sijie and Li, Yanhua and Bao, Jie and Meng, Chuishi and Li, Ruiyuan and Zheng, Yu},
  booktitle={Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery \& Data Mining},
  pages={1410--1419},
  year={2021}
}

mtrajrec's Issues

no sample data

how to use the results

i use the code python multi_main.py,then get 3 files in the data/. ,and get 1 file in the results/. ,but now i don't know how to use those files, can you tell me the way to use them? 3q

About the sample data

In the sample data provided, the longitude and latitude of the trajectory data used for training is not within the latitude and longitude range of the road network data. Could you provide the matched data?

关于sample data

Hour feature has some problem

In datasets.py file, def get_pro_features function

    def get_pro_features(self, ds_pt_list, hours, weather_dict):
        holiday = is_holiday(ds_pt_list[0].time)*1
        day = ds_pt_list[0].time.day
        # TODO: check hour data have some questions,
        hour = {'hour': np.bincount(hours).max()}  # find most frequent hours as hour of the trajectory
        weather = {'weather': weather_dict[(day, hour['hour'])]}
        features = self.one_hot(hour) + self.one_hot(weather) + [holiday]
        return features

I thought you mean get the frequent hours as hour of the trajectory, just like if I have hours = [12,12,12,12,13], the hour=12 but the code

hour = {'hour': np.bincount(hours).max()}  # find most frequent hours as hour of the trajectory

seems to get the number of the frequent hours, i.e. hour=4, not the hour.

So, I wanna know is my understanding correct, or is it other meaning?

get_rid_grid function seems incomplete

Please note that the code of "get_rid_grid" in models_utils.py is incomplete, and the part that calculates the grid cell indices for the intermediate coordinates is missing. Additionally, the calculated mid_xid and mid_yid values are not used anywhere in the code, so this part of the code would need to be completed and integrated into the rest of the logic to correctly handle the intermediate grid cells between two consecutive points.

a little confusion about train and evaluate

In the train process and evaluate process,epoch_train_id_loss is calculated.But in train id_loss calculate is epoch_train_id_loss = loss_train_ids.item(),while in evaluate is epoch_train_id_loss += loss_train_ids.item().How is this difference produced?