Information Technology Reference
In-Depth Information
•
Multi-Attempt Massive Strategy with Prediction of Gaps (MAMS-PG): As MAMS,
but this strategy addresses problem (5) by predicting gaps in the production given the
current inventory and the active orders.
•
Multi-Attempt Massive Strategy with Small Orders (MAMS-SO): This strategy op-
erates as MAMS, but orders small quantities of components for a period of time de-
termined at the beginning of the game in an attempt to avoid gaps in the production
(although these are not predicted as in the previous variation).
5.1
TAC SCM 2004
The TAC SMC 2004 competition involved five rounds: (a) qualifying, (b) seeding, (c)
quarter finals, (d) semi-finals, and (e) finals. In the qualifying round 31 teams partic-
ipated and games were played 24 hours each weekday for two weeks. All agents that
were active over 50% of the games were allowed to advance to the seeding round (30
teams). The seeding rounds took place over a two-week period similarly to the qualify-
ing round. The success of an agent was measured by a weighted average of its scores
in all games for which it was scheduled. Games played during the second week were
worth twice as much as those played in the first week. The top 24 agents proceeded to
the quarter finals for which they were divided in four groups according to their final
position in the seeding round:
group A: positions 1, 2, 3, 22, 23, 24; group B: positions 4, 5, 6, 19, 20, 21.
group C: positions 7, 8, 9, 16, 17, 18; group D: positions 10, 11, 12, 13, 14, 15.
Every group played 8 games. The top three teams from each group progressed to
the semi-finals in which the agents were divided into two groups: group 1: agents from
groups A and D, and group 2: agents from groups B and C. The top three teams in
each semifinal progressed to the finals, in which they played 16 games in total. The top
scoring agent in the finals was declared the winner.
In the first week of the qualifying rounds the MSS strategy was used and Socrates
ended up in position 20 - this was due to the problems discussed in section 4.2. In the
second week of the qualifying rounds the EMAMS-LO strategy was used. EMAMS-LO
was quite effective in keeping the factory utilisation up towards the end of the game,
if Socrates had already received the other components on time. One disadvantage of
this strategy is that in some games it is not worth ordering components towards the
end as the difference between the cost of the components purchased and the price of
the PCs sold gives the agent no profit. Socrates was placed 16th at the end of the 2nd
week.
In the seeding round Socrates (76 games) the strategy used was MAMS-PG which
attempts to predict gaps in the production after accepting all the initial offers from the
suppliers. Factory utilisation improved from 75% in the qualifying rounds to 81% in
the seeding rounds, and the average score improved as well. Socrates was placed in the
13th position at the end of the seeding rounds.
Socrates qualified for the quarter finals and was placed in the most competitive group,
D, in which all agents seemed to have similar average scores and used similar strategies.
Socrates strategy in the quarter finals was EMAMS. The performance of EMAMS was
