LaDe-D / README.md
jinyan218's picture
Update README.md
e1f85a3
metadata
license: apache-2.0
tags:
  - Spatial-Temporal
  - Graph
  - Logistic
  - Last-mile Delivery
size_categories:
  - 10M<n<100M
dataset_info:
  features:
    - name: order_id
      dtype: int64
    - name: region_id
      dtype: int64
    - name: city
      dtype: string
    - name: courier_id
      dtype: int64
    - name: lng
      dtype: float64
    - name: lat
      dtype: float64
    - name: aoi_id
      dtype: int64
    - name: aoi_type
      dtype: int64
    - name: accept_time
      dtype: string
    - name: accept_gps_time
      dtype: string
    - name: accept_gps_lng
      dtype: float64
    - name: accept_gps_lat
      dtype: float64
    - name: delivery_time
      dtype: string
    - name: delivery_gps_time
      dtype: string
    - name: delivery_gps_lng
      dtype: float64
    - name: delivery_gps_lat
      dtype: float64
    - name: ds
      dtype: int64
  splits:
    - name: delivery_jl
      num_bytes: 5568309
      num_examples: 31415
    - name: delivery_cq
      num_bytes: 168574531
      num_examples: 931351
    - name: delivery_yt
      num_bytes: 36796326
      num_examples: 206431
    - name: delivery_sh
      num_bytes: 267095520
      num_examples: 1483864
    - name: delivery_hz
      num_bytes: 335088000
      num_examples: 1861600
  download_size: 290229555
  dataset_size: 813122686

1. About Dataset

LaDe is a publicly available last-mile delivery dataset with millions of packages from industry. It has three unique characteristics: (1) Large-scale. It involves 10,677k packages of 21k couriers over 6 months of real-world operation. (2) Comprehensive information, it offers original package information, such as its location and time requirements, as well as task-event information, which records when and where the courier is while events such as task-accept and task-finish events happen. (3) Diversity: the dataset includes data from various scenarios, such as package pick-up and delivery, and from multiple cities, each with its unique spatio-temporal patterns due to their distinct characteristics such as populations.

If you use this dataset for your research, please cite this paper: {xxx}

2. Download

LaDe is composed of two subdatasets: i) LaDe-D, which comes from the package delivery scenario. ii) LaDe-P, which comes from the package pickup scenario. To facilitate the utilization of the dataset, each sub-dataset is presented in CSV format.

LaDe-D is the first subdataset from LaDe.

LaDe can be used for research purposes. Before you download the dataset, please read these terms. And Code link. Then put the data into "./data/raw/".
The structure of "./data/raw/" should be like:

* ./data/raw/  
    * delivery    
        * delivery_sh.csv   
        * ...    

LaDe-D contains 5 files, with each representing the data from a specific city, the detail of each city can be find in the following table.

City Description
Shanghai One of the most prosperous cities in China, with a large number of orders per day.
Hangzhou A big city with well-developed online e-commerce and a large number of orders per day.
Chongqing A big city with complicated road conditions in China, with a large number of orders.
Jilin A middle-size city in China, with a small number of orders each day.
Yantai A small city in China, with a small number of orders every day.

3. Description

Below is the detailed field of each LaDe-D.

Data field Description Unit/format
Package information
package_id Unique identifier of each package Id
Stop information
lng/lat Coordinates of each stop Float
city City String
region_id Id of the region Id
aoi_id Id of the AOI Id
aoi_type Type of the AOI Categorical
Courier Information
courier_id Id of the courier Id
Task-event Information
accept_time The time when the courier accepts the task Time
accept_gps_time The time of the GPS point whose time is the closest to accept time Time
accept_gps_lng/accept_gps_lat Coordinates when the courier accepts the task Float
delivery_time The time when the courier finishes delivering the task Time
delivery_gps_time The time of the GPS point whose time is the closest to the delivery time Time
delivery_gps_lng/delivery_gps_lat Coordinates when the courier finishes the task Float
Context information
ds The date of the package delivery Date

4. Leaderboard

Blow shows the performance of different methods in Shanghai.

4.1 Route Prediction

Experimental results of route prediction. We use bold and underlined fonts to denote the best and runner-up model, respectively.

Method HR@3 KRC LSD ED
TimeGreedy 57.65 31.81 5.54 2.15
DistanceGreedy 60.77 39.81 5.54 2.15
OR-Tools 66.21 47.60 4.40 1.81
LightGBM 73.76 55.71 3.01 1.84
FDNET 73.27 ± 0.47 53.80 ± 0.58 3.30 ± 0.04 1.84 ± 0.01
DeepRoute 74.68 ± 0.07 56.60 ± 0.16 2.98 ± 0.01 1.79 ± 0.01
Graph2Route 74.84 ± 0.15 56.99 ± 0.52 2.86 ± 0.02 1.77 ± 0.01

4.2 Estimated Time of Arrival Prediction

Method MAE RMSE ACC@30
LightGBM 30.99 35.04 0.59
SPEED 23.75 27.86 0.73
KNN 36.00 31.89 0.58
MLP 21.54 ± 2.20 25.05 ± 2.46 0.79 ± 0.04
FDNET 18.47 ± 0.25 21.44 ± 0.28 0.84 ± 0.01

4.3 Spatio-temporal Graph Forecasting

Method MAE RMSE
HA 4.63 9.91
DCRNN 3.69 ± 0.09 7.08 ± 0.12
STGCN 3.04 ± 0.02 6.42 ± 0.05
GWNET 3.16 ± 0.06 6.56 ± 0.11
ASTGCN 3.12 ± 0.06 6.48 ± 0.14
MTGNN 3.13 ± 0.04 6.51 ± 0.13
AGCRN 3.93 ± 0.03 7.99 ± 0.08
STGNCDE 3.74 ± 0.15 7.27 ± 0.16

5. Citation

To cite this repository:

@software{pytorchgithub,
    author = {xx},
    title = {xx},
    url = {xx},
    version = {0.6.x},
    year = {2021},
}