runcues.txt

CMAC神经网络机械臂控制的设计matlab源码

Compilation and Running Cues. 
The ziped archive with the source code for RMDP - SARSA - CMAC package also contains several additional files: 

cmac3-25.unx 
Contains structural parameters of the CMAC architectures used to represent action value functions. The name of this file can be passed as the command-line argument str=cmac3-25.unx (see MainParameters for the format of the command line arguments). You are free to change the architecture the way you like (preserving the format). 

myrmdp directory 
This directory contains a previously generated random MDP (RMDP), which you can load and learn. The name of this RMDP can be passed as the command-line argument rmdp=myrmdp (see RMDPparameters for the format of the RMDP related command line arguments). This RMDP has continuous state space [0,1]2, 2 actions and branching factor equal to 2. 

rmdpteststates.dat 
Contains 50 test states, which are used as start states for the policy evaluations on test trials. The name of this file can be passed as the command-line argument tsf=rmdpteststates.dat (see MainParameters for the format of the command line arguments). The states are sampled from a uniform distribution in the [0,1]2 state-space. Each state is specified on a different line with a blank space separating values of the state variables. 


To compile and create an exacutable, first unzip the archive and then use a C++ compiler. For example under unix, using g++:

        g++ -O2 *.cpp -o learn 


Then run the created exacutable with appropriate command-line arguments, e.g.:

         learn rmdp=myrmdp lambda=0 epsilon=0.01 schedule=visitation alpha=0.25 v=1.25 Trials=10000 Steps=500 t s=10 tf=100 tsn=50 tsf=rmdpteststates.dat str=cmac3-25.unx run=1 dir=. 


See documentation for the run() function for the explanation of the content of the produced files. 

--------------------------------------------------------------------------------