Supply chain management sales using XCSR

Abstract

The Trading Agent Competition in its category Supply Chain Management (TAC SCM) is an international forum where teams construct agents that control a computer assembly company in a simulated environment. TAC SCM involves the following problems: to determine when to send offers, to determine final sales prices of offered goods and to plan factory and delivery schedules. The main goal of this work was to develop an agent called TicTACtoe, using Wilson's XCSR classifier system to decide the final sales prices. We develop an adaptation to the classifier system, that we called blocking classifiers technique, which allows the use of XCSR in an environment with parallel learning. Our results show that XCSR learning allows generating a set of rules that solves the TAC SCM sales problem in a satisfactory way. Moreover, we found that the blocking mechanism improves the performance of the XCSR learning in an environment with parallel learning.