Milestones

·         Group formation 3/11/2010

·         A baseline agent implementation has to be completed until 15/12/2010

·         A design specification document has to be submitted by 13/1/2011, to be presented in class on 19/1/2011

·         The deadline for the final submission is 27/4/2011 (both software and project report)

       Group formation

3 to 5 members per group. At group formation the team will also select a unique agent name and password (used to configure the server for running simulations/competitions).

             A baseline Implementation

The baseline agent should implement a simple agent (with trivial modeler and estimator – see Appendix A below) that only bids to one query - the 'specialty' query (the query with the specialty manufacturer and component). The optimizer should decide on bid and limit for the next day based on past days results. The submitted baseline agent will include the following:

1.      Specific agent and architecture configuration files.

2.      Agent components packages replacing the provided samples.

3.      The two server-generated logs of simulations in which the agent participated (see Appendix B below).

             A design specification document

The design specification document should describe the planned implementation of each of the agent components. Specifically, the following should be detailed for each of the components:

1.      Learning method:  algorithm and related mechanisms/data structures.

2.      Training/Bootstrapping: data to be collected, and management/usage method.

3.      High level architecture: main classes, interfaces and relations. 

The design specification will be presented in class (see milestones above). It is expected that the above will evolve throughout the implementation. A refined updated design document is required as part of the final submission.  

             Final Submission

The complete project should contain the following:

1.      software - similar to the content of the baseline implementation, but with the final agent components packages.  The code itself must be well documented.

2.      design spec – updated to reflect the final implementation.

3.      project report - 

·         A brief introduction and an overview of the project and relevant background

·         A detailed explanation regarding the learning methods used (including reasoning, explaining and motivating your decisions). Also, a detailed explanation regarding of methods that were unsuccessful (if any), and a description of the training process.

·         Performance/simulation results and analysis.

             The Grade Structure

The grade will be composed from all your activities during the workshop, and will include, but not limited to, the following:

·         The quality of the program: should be at least reasonable! extra points for high performance!

·         Milestones - quality and timely submissions.

·         The ideas and the testing that were tried.

·         An overall impression of the project and the project report.

Appendices

             A. Software Infrastructure

The Ad-Auctions agent should be based on the provided software infrastructure:

1.      reference libraries – implements the generic agent's mechanisms and related data structures (communication with the server, logging, etc').

2.      src: set of java packages that (combined with the referenced libraries) comprise a full agent implementation (see the architecture diagram in the workshop site):

·         agent – fully implements the control module and server messaging. This package should not be modified.

·         arch – interfaces and abstract classes of agent components. This package should not be modified.

·         sampleEstimator, sampleModeler, sampleOptimizer – sample implementations of the agent model, estimation, and optimization components (illustrating the usage of arch). Those packages should be replaced by your own implementation. Note that inter-component interaction is only allowed through the interfaces defined by the arch package, this will allow to subsequently mix and match components to achieve high performing agents.

3.      config – an agent configuration file (with the agent name and password, server contact info, etc'), and an architecture configuration file (names of classes implementing the agent's components). Those files should be replaced by your own

             B. Simulations

Two simulation servers will be available for agents to connect and run simulations (specific connection details will be provided during the workshop). One of the servers will be always up and agents for teams' agents to compete against past years' agents. Also, a second server will run a weekly competition between agents of interested teams (details to follow, participation is optional).