Store agent state

[Javier Alfonso]

How can I store the agent state among program executions? I want to preserve my agent training among program executions, i can't find a suitable way of doing this
PD: i'm very new to BURLAP

[James MacGlashan]

Hi,

Which agent are you using? Part of the changes in BURLAP 3 were based around making serialization much more trivial, but not all of the algorithms have default methods implemented just yet to take advantage of that and would require a bit of more manual serialization (but probably something that wouldn't be too hard to implement).

QLearning and SARSA, however, do currently have default serialization methods for their Q-functions, which is the primary statistic you would want to load between sessions. See the writeQTable and loadQTable methods:

http://burlap.cs.brown.edu/doc/burlap/behavior/singleagent/learning/tdmethods/QLearning.html#writeQTable-java.lang.String-

http://burlap.cs.brown.edu/doc/burlap/behavior/singleagent/learning/tdmethods/QLearning.html#loadQTable-java.lang.String-

James

[Javier Alfonso]

i'm using Q-Learning. To preserve what my agent has learned i should use the writeQTable and loadQTable methods

[James MacGlashan]

Yep!

James

[fifi.alf...@gmail.com]

Dear James,

I am very new to Burlap. I am sorry if my question sound so basic. Can you give me an example on how to use the writeQTable and loadQTable methods? Thank you.