This is an implementation of the ‘Order Answer-rate Prediction’ application in Section 4.1 and the 'Policy Evaluation' application in Section 4.3 of paper 'Graph-Based Equilibrium Metrics for Dynamic Supply-Demand Systems with Applications to Ride-sourcing Platforms'. All the required codes to reproduce the main results of the two applications in the paper are provided. In this repository, we do not share the raw data while some processed data can be obtained by submitting an application to DiDi Gaia Open Data Initiative.
-
Python Packages: cvxpy 1.1.5, numpy, scipy, pandas, sklearn
-
Optimization Package: GLPK
The required data can be obtained from the following website after sigining in:
https://outreach.didichuxing.com/appEn-vue/DatasetProjectDetail?id=1026
The demand and supply data, named as 'area5_xxxx_order'.npz and 'area5_xxxx_driver.npz' can be found in the folder 'area_driver' and 'area_order', where the number 'xxxx' represents the date.
Put all the supply-demand data from 20180421 to 20180521 in the path
Application1/data/processed/area5/
Put all the supply-demand data from 20181112 to 20181216 in the path
Application3/data/processed/area5/
The file 'grids_to_id.pickle' in the folder 'grids' needs to be copied to the above two paths.
File 'data_cost_lambda_eq1.csv' needs to be placed in the path
Application3/data/ab_test/area5/
All files in the folder 'AA_AB_test' should be placed in the path
Application3/
Compute supply-demand ratio and GEM from 20180421 to 20180521 for Application 1, from 20181112 to 20181216 for Application 3.
cd Application1/
# from 20180421 to 20180430
sh run0.sh
# from 20180501 to 20180521
sh run1.sh
cd Application3/
# from 20181112 to 20181125
sh run0.sh
# from 20181203 to 20181216
sh run1.sh
When the above jobs are done, you can find the outputs,
- 'area5_xxxx_metric.npz' : calculated by GEM, include
ratioandglobal_metric. - 'area5_xxxx_metric_1.npz' : calculated by Wasserstein Distance, include
ratioandglobal_metric_w.
where the number 'xxxx' represents the date, under the path
Application1/data/output/area5/
and
Application3/data/output/area5/
cd Application1/
python functional_prepare.py
Output data.npy in the path
Application1/data/prediction/area5/
To get the prediction results for GEM, Hellinger Distance, L2 Distance and Wasserstein Distance
nohup R CMD BATCH model_t+1.R &
nohup R CMD BATCH model_t+6.R &
Integrate data in'area5_xxxx_order.npz','area5_xxxx_driver.npz', area5_xxxx_metric.npz' of different dates, where the number 'xxxx' represents the date.
cd Application3/
python cal_out.py
python AB_test.py