Create the half-step methods

[GoogleCodeExporter]

Want to be able (from an experiment) to do:

s = RL_start_env
a = RL_start_agent(s)

(s,r,t) = RL_step_env

if(t)
    RL_agent_end(r)
else
    RL_agent_step(s,r)

This will allow us great flexibility in the experiment program.

Still need to clear it with Adam 100%.  Maybe the way to think about it is
that these methods are exposed to the experiment, but RL_episode and
RL_step are convenience methods that make life easy.

Original issue reported on code.google.com by brian.ta...@gmail.com on 12 Nov 2008 at 8:59

[GoogleCodeExporter]

This is implemented, but will remain an undocumented API for now.  I really only
needed it so that I Could do neat stuff with RL-Viz, so I've added it to the 
java
codec and the rl_glue server only for now.  It will get added to other codecs if
anyone wants it or if it becomes useful more generally.

The methods we ended up adding are:
/**
    New Experimental Methods, not part of the public API
**/
const action_t* RL_agent_start(const observation_t* observation);
const action_t* RL_agent_step(double reward, const observation_t* observation);
void RL_agent_end(double reward);  
const observation_t* RL_env_start();
const reward_observation_terminal_t* RL_env_step(const action_t* action);

Original comment by brian.ta...@gmail.com on 3 Dec 2008 at 4:23

• Changed state: Fixed