|
2007 TAC-SCM Prediction Challenge |
Welcome to the home of the 2007 TAC-SCM Prediction Challenge.
In this challenge, agents will make daily predictions over the course of a number of TAC SCM games about four different types of prices:
current and future computer prices, and current and future component prices. Instead of making predictions about live TAC SCM games in which they are participating,
agents will make predictions for games that are "re-played" from game logs on behalf of another agent named
SCMPredictionAgent.
Schedule:
Prediction Challenge Software
SCMPredictionAgent
Sample game logs
Qualifying Round
Final Round
Monday, July 2: Qualifying round
Tuesday, July 24: Final round
The following is the beta release of the challenge software. Feel free to email us with any questions or comments.
Download the software for parsing game logs and processing predictions, as well as sample prediction code:
standalone version (useful for agent development)
client/server version (to be used in the competition)
View the JavaDocs.
Download the source and binary of the SCMPredictionAgent (the TAC SCM agent for which predictions will be made) here.
This agent is based on the
TacTex Starter Agent.
Download 5 sample game logs here (28MB).
Download 16 more sample game logs here (92MB). (Agents: SCMPredictionAgent, PhantAgent05, PhantAgent06, TacTex06, Maxon06,
Mertacor05)
The qualifying round consisted of 3 groups of 16 games. In addition to the SCMPredictionAgent, each group contained the following agents
(from the TAC SCM repository):
A: DeepMaize06, RationalSCM05, Foreseer05, PhantAgent05 x 2
B: MinneTAC06, TacTex05, TacTex06, Maxon06 x 2
C: GoBlueOval05, Foreseer05, GeminiJK05, Mertacor05, MinneTAC05
Download the game logs (253MB).
Results:
Average rank:
1. DeepMaize: 1.75
2. TacTex: 2
3. Botticelli: 2.5
4. Kshitij: 3.75
Current computer prices RMS error:
1. DeepMaize: 0.04632
2. TacTex: 0.04812
3. Kshitij: 0.04882
4. Botticelli: 0.09449
Future computer prices RMS error:
1. TacTex: 0.09318
2. DeepMaize: 0.09473
3. Botticelli: 0.13984
4. Kshitij: 0.15428
Current component prices RMS error:
1. Botticelli: 0.04163
2. TacTex: 0.04239
3. DeepMaize: 0.04810
4. Kshitij: 0.10654
Future component prices RMS error:
1. DeepMaize: 0.09925
2. Botticelli: 0.10067
3. TacTex: 0.10714
4. Kshitij: 0.16324
The final round consisted of 3 groups of 16 games. In addition to the SCMPredictionAgent, each group contained the following agents
(from the TAC SCM repository):
A: PhantAgent06, Maxon06, RationalSCM05, Tiancalli06, PhantAgent05
B: GoBlueOval05, GeminiJK05, RationalSCM05, PhantAgent05, TacTex06
C: Maxon06, MinneTAC05, DeepMaize05, Foreseer05, PhantAgent06
Download the game logs (247MB).
Results:
Average Rank:
1. DeepMaize: 1.5
2. TacTex: 2
3. Botticelli: 2.5
4. Kshitij: 4
Current computer prices RMS error:
1. TacTex: 0.04554
2. DeepMaize: 0.04682
3. Botticelli: 0.04714
4. Kshitij: 0.04873
Future computer prices RMS error:
1. TacTex: 0.09156
2. DeepMaize: 0.09586
3. Botticelli: 0.10238
4. Kshitij: 0.11094
Current component prices RMS error:
1. DeepMaize: 0.03919
2. Botticelli: 0.04168
3. TacTex: 0.04284
4. Kshitij: 0.13333
Future component prices RMS error:
1. DeepMaize: 0.09427
2. Botticelli: 0.09700
3. TacTex: 0.10338
4. Kshitij: 0.13886