Code

Project: Box2D_friction_mod

test_TopDownCar.py      WheelConstraint.h      test_TopDownFrictionJoint.py      box2d_friction_joint.patch      b2FrictionJoint.h      TopDownCar.h      b2FrictionJoint.cpp      TestEntries.cpp      python_friction_joint.patch     

Project: jvfeatures

jvtypes.h      jvtypes.cpp      jvfeatures.h      chessSeg.cpp      jvtest.cpp      jvfeatures.cpp     

Project: Other

migrateMailbox.scpt.txt     

Project: Infinite HMM Tutorial

HMMProblem.m      HMM.m      README.txt      HDP.m      HDP_HMM.m      run.m      iHMM_tutorial.zip      ConditionalProbabilityTable.m     

Project: Arduino_Code

arduino-serial.c      Arduino_Code.zip      convert_range2D.py      plot_ardunio_data.sh      oscilloscope.pde      oscilloscope.sh      ranger_test.pde      accelerometer_test.pde      pwm_manual.pde      servo_test.pde      motordriver.pde      ranger_plane_sweep.pde      helicopter_controller.pde      clodbuster_controller.pde     

Project: RRT

BidirectionalRRT.cpp      RRT.tgz      AbstractRRT.cpp      RRT.h      RRT.cpp      rrt_test.cpp      plot_output.py     

Project: ArduCom

arducom.py      setup.py     

Project: Dirichlet Process Mixture Tutorial

EM_GM.m      DP_Demo.m      DP_Tutorial.zip      DirichletProcess.m      DPMM.m      gaussian_EM.m     

Project: support

Protector.php      geshi.php     

Project: Cogent

CodePane.php      Cogent.php      NotesPane.php      PubsTable.php      PicsPane.php     
Click here to download "resources/code/Infinite HMM Tutorial/run.m"

resources/code/Infinite HMM Tutorial/run.m

% Example run configurations for the HMM models in this demo.  This code
% was left fairly alpha (see README.txt), but the examples below work
% consistently.
% Note that the 3rd one, which starts underpowered, seems to converge to a
% locally optimal representation with fewer than 10 states.  I played with
% this a lot here and in Jurgen's code, and it just seems like a property
% of the problem itself.**  
%
% Jonathan Scholz
% jkscholz@gatech.edu
% 11/2/2011

% Build a problem
p=HMMProblem;

% Run fixed-size HMM on a 10 state representation
p.runFixed(10, 100, true, false);

% Run iHMM on 30 state representation
%p.runHDP(30, 100, true, false);

% Run iHMM on 5 state representation (interestingly, doesn't work as well)
%p.runHDP(5, 100, true, false);

% **
% Even Jurgen's beam sampler got stuck in this mode, which I guess means
% these modes are very pronounced and tough to walk out of.  It makes
% sense if you watch the transition and emission matrices: suboptimal
% modes end up assigning all 30 of the tokens for several of the true
% states to one state, which means that the bin counts tend to occur in
% multiples of 30 in the modes.  This aliasing of the true states would
% require resampling all 30 incorrect states at once to a new correct one,
% which is clearly unlikely.  Watching it online, you can see the counts
% drift down from 60 (or 30*x), and just when you get excited the
% super-state will gobble them back up again.  

About me

Pic of me