Home > marl > batch > processexp.m

processexp

PURPOSE ^

processes a series of experiments ran using runexp()

SYNOPSIS ^

function [stats, worlds] = processexp(datafile, mode, plotcount, plotfields, cfg)

DESCRIPTION ^

 processes a series of experiments ran using runexp()
 STATS = PROCESSEXP(DATAFILE[, MODE, PLOTCOUNT, PLOTFIELDS, CFG])
  Processes a series of experiments ran using the MA-RL toolbox
  via the function runexp().

  Parameters:
   DATAFILE    - the name of the datafile where the experiment results
       were saved.
   MODE        - the processing mode. One of four available modes, 'plot',
       'replay', 'replayeach', 'manual'. See below for a detailed
       description of each. The default is 'plot'.
   PLOTCOUNT   - in 'plot' mode, how many plots should be displayed on a
       single figure. Useful when the experiments sequence is composed of
       constant-length subsequences each characterizing one category of 
       settings (e.g. one such subsequence could be characterized by a
       given configuration of the world, and within it various learning
       parameter configurations might be tried). The default is 0, which
       means all plots are collected on a single figure. Use 1 to give
       each plot its separate figure.
   PLOTFIELDS  - which fields of the saved statistics are interesting for
       plotting. A cell array of field names. Default is {'iter'}. Field
       evolutions will be plotted in separate figures, one per field.
   CFG         - additional configuration fields (see below for commented
       defaults)

  Returns:
   STATS       - replay statistics, in replay/replayeach modes.
       Either an nexp-cell array of stat structures, in replay mode;
       or an nexp-cell array of nrun-cell arrays of stat structures, in
       replayeach mode.
   WORLDS      - world objects, in replay/replayeach modes.
       An nexp-cell array of world structures

  The function outputs the configuration of each experiment at the console
  before processing it.

  Processing modes
 ==================
  In 'plot' mode, statistics of the learning process are being plotted.
  The values are averaged over the number of saved independent runs of
  each experiment.
  In 'replay' mode, if objects have been stored during running the
  experiments, the learned behaviour of the agents is replayed. A random
  or specified experiment run is chosen from the saved independent runs.
  In 'replayeach' mode, if objects have been stored, the learned behaviour
  of the agents in each run of each experiment is replayed.
  In 'manual' mode, the user is given full control of the processing
  sequence. A command prompt is displayed, where for each experiment the
  following options are available:
   p   - [Plot] the experiment statistics in the current figures, open 
       new figures if none are open.
   np  - [New figure(s) & Plot] open new figure(s) for the interesting
       fields, then plot the experiment statistics.
   r   - [Replay] behaviour from a random run.
   re  - [Replay Each] replay behaviour from each run of the experiment.
   pa  - [Plot All] terminates manual mode and processes all the remaining
       experiments in 'plot' mode.
   ra  - [Replay All] terminates manual mode and processes all the
       remaining experiments in 'replay' mode.
   q   - [Quit] terminates processing discarding any remaining experiments.

  For the 'replay' and 'replayeach' modes, the world must support a
  view.


 See also runexp

CROSS-REFERENCE INFORMATION ^

This function calls: This function is called by:
Generated on Wed 04-Aug-2010 16:55:08 by m2html © 2005