Autonomous Negotiation Using Comparison-Based Gradient Estimation

This is the repository for the numerical and human interaction experiments performed in the paper Autonomous Negotiation Using Comparison-Based Gradient Estimation.

The numerical experiments presented in the paper were run in Python 3.9.7 on a Dell XPS 13 laptop with a 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz processor and 16.0 GB of RAM.

Required Packages

The versions of python packages used to generate our results are given in the requirements.txt file. The packages can be installed by running:

    pip install -r requirements.txt

Running Numerical Tests

Use the following command to run the numerical tests:

    python run_numerical_tests.py [Arguments]

Command-Line Arguments

The available command-line arguments are as follows:

• --debug <flag>

  • Description: Enable debug mode for detailed logging and troubleshooting.
  • Default: Disabled (i.e., debug mode is off).

• --integer_constraint <flag>

  • Description: Apply integer constraints in the trading scenarios.
  • Default: Disabled (i.e., no integer constraints applied).

• --random_seed <int>

  • Description: Set the random seed for initializing testing scenarios. This ensures reproducibility of results.
  • Default: 10

• --max_trade_value <int>

  • Description: Specify the maximum number of items from each category that can be exchanged in a single trade.

  • Default: 5

• --theta_closeness <float>

  • Description: Define the semi-vertical angle threshold for the ST-CR algorithm. This parameter controls the precision of the cone refinement.

  • Default: 0.00001

• --deviation_interval <float>

  • Description: Set the deviation interval for random trading with momentum. This parameter affects the randomness of trading scenarios.

  • Default: 0.05

• --max_deviation_magnitude <float>

  • Description: Determine the maximum deviation magnitude for random trading with momentum.

  • Default: 5

• --num_scenarios <int>

  • Description: Specify the number of randomized testing scenarios to be generated. Note that this argument should be an integer.

  • Default: 100

• --offer_budget <float>

  • Description: Set the offer budget for the negotiation algorithms. This budget dictates the maximum value of offers that can be made.

  • Default: 1000

Results

The testing script outputs the plots that show normalized cumulative benefit as the number of offers increases. The plots are stored in the results folder with the format: query_benefit_curves_alignment_{mixing_constant}_items_{num_items}. We store the numerical results from the paper in the folder numerical_results/paper_results.

Running GPT Tests

The human interaction test script are located in the gpt_integration folder.

    cd gpt_integration

To run the human interaction test, run the following command

    python gpt_integration_test.py

When the script starts, you should see a link to a locally hosted webpage of the form:

Starting Website on http://127.0.0.1:5000
 * Serving Flask app 'gpt_integration_test'
 * Debug mode: on  

Clicking on the link will take you to a webpage which will ask you for relevant information before trading begins. We note that you will need to provide your own OpenAI API key to run these tests. ###Results The testing script stores a log of the negotiation with humans. The transcript is stored in a folder with a timestamp (Ex: chat_folders/1723827209/log.txt). The example logs that we show in the appendix are stored in chat_folders\paper_examples