Download presentation
Presentation is loading. Please wait.
Published bySophie Murphy Modified over 9 years ago
1
TAC Classic: VegBot Team Members Venkata Yellapantula Evan Liu George Alexander
2
TAC Classic ● Flight tickets: continuously clearing, infinite supply ● Hotel tickets: 1 random closure every minute ● Event tickets: continuous double auctions ● MAS aspects: – Stochastic, dynamic environment – Self-interested agents – Non-local effects (other agents affect the environment of our agent)
3
Problem Dimensions ● Price prediction : what will the price of item X be at time T ● Item valuation : what is item X worth to me? ● When should I place my bids? ● Should I counterspeculate about other agents?
4
Agent Design Ideas ● Learning based on historical TAC data (e.g. ATTac) ● Alternate bidding strategies based on assessment of competition (e.g. Risk- averseness) ● Ignore other agents, bid based on personal valuation of goods ● Simple bidding heuristics
5
Evaluation – 2 Possibilities ● Alternative A – Connect to SICS TAC Classic server – Run lots of games – Evaluate our agent’s average performance ● Alternative B – Download TAC Classic server + some agents from agent repository – Run many games locally using the same set of competitors – Evaluate our agent’s performance
6
Project Timeline ● Research/design: 2/21 - 3/21 ● Implementation/debugging: 3/21 - 4/11 ● Experimentation: 4/11 - 4/25 ● Analysis: 4/25 - 5/2 ● Presentation of Results: 5/9
7
References ● TAC Classic website: – http://www.sics.se/tac/page.php?id=3 ● Amy Greenwald’s website: – http://www.cs.brown.edu/~amygreen/
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.