Building a Recommendation Engine: The XELOPES Library - Realtime Data Mining - page 270

// Put it all together:
agent.setAgentSettings(settings);
agent.setAgentEnvironmentInformation(env);
agent.verify();

// Apply and learn:
int indItem ¼ metaData.getAttributeIndex("itemID");
int indUnits ¼ metaData.getAttributeIndex("units");
while ( inputData.next() ) {
MiningVector mv ¼ inputData.read();
Category itemID ¼ mv.getValueCategory(indItem);

// Get recommended price:
double price ¼ agent.apply(mv);

// Recommend price for item ID, get units ordered (0 if no
order):
int unitsOrdered ¼ getResponse(itemID, price); // not
shown here
mv.setValue(indUnits, unitsOrdered);

// Learn from response (stateful):
agent.addData(mv);
agent.learn();
// Alternative - Learn from response (stateless):
// agent.learn(mv);

}
// Write to PMML:
FileWriter

¼

writer new FileWriter("data/pmml/

LinElastAgent.xml");
agent.writePmml(writer);

■

Finally, we notice that the Agent package provides a useful wrapper class
MiningModelAgent , which extends Agent , in order to use data mining models/
algorithms as described in Sect. 12.1.3 inside the agent framework. This is of
particular interest for updateable mining models such as na¨ve Bayes or adaptive
decision trees. At this, to MiningModelAgent ,a MiningModel and a suitable
MiningAlgorithm can be assigned and can then be used as an agent.

12.2.2 The Reinforcement Learning Package

In XELOPES, the reinforcement learning package is implemented as subpackage
RL of the Agent package (previous section). The RL package in turn contains the
following subpackages: