The Positronic Economist: A Computational System for Analyzing Economic Mechanisms
Authors: David Thompson, Neil Newman, Kevin Leyton-Brown
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We now present experimental evidence that our two structure inference algorithms produce compact games in a practical amount of time and that these games can be used to compute sample Nash equilibria efficiently. To evaluate our algorithms, we turned to games for which previous work has manually identified compact encodings; they happen to be perfectinformation games (thus, AGGs rather than BAGGs). Specifically, we recreated GFP and w GSP position auctions games from Thompson and Leyton-Brown (2009) and two-approval voting games from Thompson et al. (2013). We generated straightforward specifications of these settings in Pos Ec and then ran WBSI and BBSI. For every setting, for every number of agents, we generated 10 different games. The results are summarized in Figure 3. |
| Researcher Affiliation | Academia | David Thompson, Neil Newman, Kevin Leyton-Brown Department of Computer Science University of British Columbia, Canada {daveth, newmanne, kevinlb}@cs.ubc.ca |
| Pseudocode | Yes | Algorithm 1: White-Box Structure Inference, Algorithm 2: Black-box structure inference |
| Open Source Code | Yes | The package is open source and pointers to the software, compatible equilibriumfinding algorithms, and further documentation are available at https://www.cs.ubc.ca/research/posec/. |
| Open Datasets | Yes | Specifically, we recreated GFP and w GSP position auctions games from Thompson and Leyton-Brown (2009) and two-approval voting games from Thompson et al. (2013). |
| Dataset Splits | No | The paper does not provide specific training/validation/test dataset splits. It describes generating different games for evaluation, but not how data within those games is partitioned for training or validation purposes typically seen in machine learning contexts. |
| Hardware Specification | Yes | All of our experiments were run on Intel Xeon E5-2640 v2 processors on nodes with 96 GB of RAM. |
| Software Dependencies | No | The paper mentions a 'Python-based declarative language' and 'compatible equilibriumfinding algorithms' but does not specify version numbers for Python, libraries, or any other software dependencies. |
| Experiment Setup | Yes | For our position auction games we used 4 positions, 20 bid increments, and the Varian (2007) preference model. We varied the per-agent click-through rates, valuations, and quality scores across games. We restricted to conservative strategies (Caragiannis et al. 2011; Roughgarden and Tardos 2012) in which bidders do not play the weakly dominated strategies of bidding above their valu-ations. For our two-approval games, we considered settings with 5 candidates and a variable number of voters. For each game, we randomly assigned each bidder some permutation of 0, 1, 2, 3 and 4 utility points for each of the different candidates. ... Each algorithm s runtime depends on its starting point, which we initialized randomly; we considered the median time to find an equilibrium over 10 such randomly sampled points, reporting the distribution over games for each setting. |