Diferentially Private Decision Making

This repo includes codes for our recent publication.

One can run simulate.py to start the simulation study. The results will appear in the folder "results".

Required packages: numpy, csv, gurobipy

Required solver: Gurobi

For each scenario, there will be four actions.

• Two txt files will be created with names showing the dynamic parameters (can be seen below). The files include the following
  • Optimal objective function values
  • Data private model iteration results (in an every 10 iterations)
  • Data private model objective function value
  • Data private model decision variable values
  • Differentially private model iteration results (in an every 10 iterations)
  • Differentially private model objective function value
  • Differentially private model decision variable values
• For each scenario, a row will be added to "results.csv". The row includes the following
  • the parameters of the scenario
  • objective function values of 3 models
  • primal solutions of 3 models
    • # (parties * max_product) entries for the initial model
    • # (parties * max_product) entries for the distributed model
    • # (parties * max_product) entries for the differentially private model
  • dual solutions of 3 models
    • # (m + parties * max_private_cap) entries for the initial model
    • # parties * (m + max_private_cap) entries for the distributed model
    • # parties * (m + max_private_cap) entries for the differentially private model
  3. For each scenario, a row will be added to both dataPrivate.csv and diffPrivate.csv. Rows include the followings
     • dataPrivate.csv
       • seed value
       • \bar{s}_k values for each party
       • best objective function value so far
     • diffPrivate.csv
       • seed value
       • \bar{s}_k values for each party
       • shared resource capacities
       • best s_k values
       • epsilon
       • delta
       • best objective function value so far

Current dynamic parameters for the simulation in test.py

• Number of parties (parties): 5
• Number of shared resources (m): 5
• Epsilon (epsilon): [0.05, 0.10, 0.15, 0.20, 0.25]
• Delta (delta): [0.05, 0.10, 0.15, 0.20]
• Upper Limit on shared resources (upper_limit_policy, upper_limit_coef): [0.15, 0.30, 0.50, 1.00]
• Seed (seed): 1 to 100 with increments 1

Current static parameters for the simulation in test.py

• maximum number of iterations for data-private model (max_iter_subgradient): 5000
• maximum number of iterations for differentially private model (max_iter): 100
• maximum number of product for each party (max_product): 15
• maximum number of private capacity for each party (max_private_cap): 10

The inputs of the original model (A_k, B_k, b_k, r_k, c) are mostly generated randomly. The parameters of these inputs can be seen in Section 3 of the paper.