This repository is the official implementation of Replication study of "Data-Driven Methods for Balancing Fairness and Efficiency in Ride-Pooling"

To install requirements for simulating the taxi environment, training the location embeddings, or training the neural value function:

conda env create -f environment.yml

To install requirements for preprocessing raw New York taxi data, obtaining a graph for (part of) a city using OpenStreetMap, or obtaining corresponding travel time and shortest path files:

conda env create -f preprocessing/environment.yml

The labels that assign each node in the graph to a neighborhood and are required to train/evaluate models and to generate the results are created as follows:

python preprocessing/kmeans.py

To create the graph (i.e. streetnetwork) that is needed to replicate the original experiments, run respectively for Manhattan and Brooklyn:

python preprocessing/get_graph_Manhattan.py
python preprocessing/get_graph_Brooklyn.py

The txt files containing the rider request (start, dest, counter) can be obtained by running these commands:

python preprocessing/flow_files_manhattan.py
python preprocessing/flow_files_brooklyn.py

To generate the csv file with the shortest paths and the travel time of these paths, run this command:

python preprocessing/generate_paths_and_traveltimes.py

To obtain the initial locations of the taxies, run:

python preprocessing/initializations_taxies.py

The pretrained location embeddings are obtained by running these commands:

python src/utils/generate_embeddings.py

To train and evaluate the model(s) in the paper, run these commands:

# driver side fairness objective
python src/main.py training_days 3 testing_days 1 num_agents 200 value_num 10 write_file True print_verbose False lambda 0.67 data_dir "'../data/ny/'"
# income objective
python src/main.py training_days 3 testing_days 1 num_agents 200 value_num 15 write_file True print_verbose False data_dir "'../data/ny/'"
# rider side fairness objective
python src/main.py training_days 2 testing_days 1 num_agents 200 value_num 14 write_file True print_verbose False lambda 1000000000 data_dir "'../data/ny/'"
# requests objective
python src/main.py training_days 3 testing_days 1 num_agents 200 value_num 1 write_file True print_verbose False data_dir "'../data/ny/'"

The figures required to support the claims, which were generated here: