CVS log for Babybot/Babybot.py
	Help

Request diff between arbitrary revisions

updated code to work under new version of pyrobot

updated all files to import from pyrobot

tools.py: made rnd always return floats of 2 decimals

Babybot.py: we turned on gov aspect, we lowered learning rates
0.7 as the highEpsilon was causing huge changes in weights (often
within one backprop step).

BabybotOffline.py: we tried to make it match Babybot experiment in the
reinforcement procedure and in the scale of the environment.

Working under Pyro4; changed GovernorSRN epsilon and delta to allow model
vectors, but I'm not sure if these will be quality MV (this is with the new
weight mask).

Doug and Lisa changed call to network's step function so that it
always included values for every input and output layer.  Even though
certain layers are inactive, we needed dummy values in order to use
the ravq.  This is just a quick work around.  We'll eventually want to
come up with a more general way of sharing weights between particular
layers rather than between entire networks.

Changed all calls to self.net.step to self.net.networkStep (which is
what the governor overrides).

Added some code in the setupNetwork method to set an appropriate ravq
mask for this task.

removed debugging pring statements

In Avoid.py: for some tests we had made the robot be stationary. I
removed this and let the function determineMove set translate and
rotate.

In Babybot.py: Replaced SRN with a Governed SRN.  Added the
"ordered-continuous" setting. Updated setEpsilon calls to take an
additional parameter.

In playpen.world: Commented out the additional stationary robot at the
bottom of the world.

In setup.py: The test function was obsolete.  Babybot.py includes a
way to test particular weight files.

srn now requires type of input to be set

Improved efficiency of Babybot.py code.  We were repeatedly
recalculating the SaliencyMap.  Saved it in self.currentSaliencyMap
and self.previousSaliencyMap instead.

Cleaned up Babybot code. Fixed various representations to match new
way of representing saliency information.

Have the basic next version working; I think it has some noise too

Doug: finished biasedError() and createSaliencyMap(); needs update pyro

Added blob; need to get biasedError bitmap from errorCam

See NOTES.txt; have plan will code

See NOTES.txt: working on adding error as input

See NOTES

Fixed output; fixed true false 1 0 problem; added new perf measures;

Added params to perform: xmin ymin expression

Added go(seconds); changed Babybot's reinforcement function to reward when in center always

Currently testing stationary decoy with no punishment when it doesn't
see anything. Added NOTES

Brought up to speed with changes made to Pyro 3.2.2

Added binify.py; new errorplot with bin = 25

Added error input; representation of error

Disabled red robot; auto load weights when testing = 1

Added record flag to save images from cameras to ppm files; added printColorMap for BW printing

adjusted reinforcement reward values

Made both robots blue for now; made reinforcement return 3 things: signal (+, -, 0); epsilon, and temperature. Seems to work.

did a bunch of random crap all night long, nothing worked, tomorrow we write!

added greaterThanPredictedErrorMap

Added param variables, and print out to log; current settings small linear

code for running experiment 7

Added predict flag

code for experiment 6 (see notes.txt)

Removed positive reinforcement for kepping decoy robot in same place; working on stochastic; added errorMap variable in Babybot

Added printout of raw output

Added param.stochastic to MotorRepBinary; added different variation of complement with stochastic; slowed down robot to between +-.5 rather than +-.9; added setup() to setup.py; add motorRotateHistoryProfile to Babybot

major improvements to InputCam/ErrorCam code (subsumed into Cam); changed
reinforcement signal to reward centroid stability

changed a lot of stuff

Added flag for err/vis centroid; add temperature global

uses 2D visual rep, binary motor rep, error cam, RL signal based on error

Refined output; changed output directory of weights and filename

modified discretize to be deterministic and cleaned up code a little

Bug in logging performance on save

Connected log to performance measure

Added measure of performance and restarting code

added logging to Babybot

changed offline.py code to work with different visual rep modules

added a few hacks to allow offline training with linear visual rep

modified Collect to give smoother robot movements, and tinkered with offline.py

cleaned up and reorganized everything into separate representational modules

removed a few unnecessary files

Initial version of Babybot

Initial revision