jens/Rbm-legacy
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Commit Graph

  • 431f1c47a6 - clean working copy for leaving SVN master jens 2022-06-28 18:51:10 +02:00
  • 3355aaa8f7 - cleaned up jens 2019-11-07 17:32:01 +00:00
  • 0b85dd7148 - no weight decay per default jens 2019-10-29 07:20:20 +00:00
  • 756c4d8670 - Noise_Init() depoends only on seed - more detrmistic calls of shuffle() - Shuffle() init weights with uniform noise jens 2019-10-29 07:02:45 +00:00
  • 2ac8d27e75 - better error report jens 2019-10-28 23:09:33 +00:00
  • 1d02f08242 - use normalize per training batch jens 2019-10-18 06:23:11 +00:00
  • ea6a1ee6ed - use normalize per training sample jens 2019-10-18 06:16:27 +00:00
  • 2c2039ec2f - added early stop of training - fixed non-visible training data after load jens 2019-10-17 19:48:14 +00:00
  • 0f2fe2b315 - fully refactored traning - removed non working stuff jens 2019-10-17 18:58:12 +00:00
  • b46c65c492 - improved DrawData() jens 2019-10-16 17:42:38 +00:00
  • 262ae30576 - refactored - print L1 and L2 values jens 2019-10-16 16:34:38 +00:00
  • 149385d292 - RBM use L1 weight decay jens 2019-10-14 19:02:27 +00:00
  • 9d8bdf45b3 - check result of fscanf jens 2019-10-14 19:01:12 +00:00
  • 77a3bca92d - check result of fscanf jens 2019-10-14 19:00:51 +00:00
  • 82e7286848 - added tooltip window jens 2019-10-01 17:51:15 +00:00
  • 4dc1a01c34 - changed path for prj files to /home/jens/.rbm jens 2018-10-05 20:13:25 +00:00
  • 80a67ba314 [RBM] - committed last changes jens 2018-06-12 17:01:49 +00:00
  • 9f8d84008c - load training: no size between samples, progress jens 2016-07-12 16:54:21 +00:00
  • 2eabaa42ec [RBM] - miniBatchSize is parameter of train() jens 2016-07-07 23:34:03 +00:00
  • 0259727a2c [RBM] - improved multi layer weight reconstruction using weight convolution jens 2016-07-07 22:54:54 +00:00
  • 57cdfd7065 [RBM] - GUI: added gaussian hidden, added mini batch size - Rbm: added mini batch training revised sample functions, reverted to old weight decay jens 2016-07-07 19:27:33 +00:00
  • 4979ab41b4 [RBM] - introduced mini batch training jens 2016-06-30 07:52:49 +00:00
  • 6e93c07863 [RBM] - render lower layer weights as linear combination from top weights using up pass of reconstructions jens 2016-06-29 20:29:17 +00:00
  • 54e4065a00 [RBM] - splitted bm into hpp and cpp jens 2016-06-27 07:36:28 +00:00
  • b1fcf14e99 [RBM] - sample h before negative phase jens 2016-06-23 06:45:57 +00:00
  • 605ee4da46 update jens 2016-06-22 19:29:13 +00:00
  • 866a0349db [RBM] - DrawComponent: use fixed value scaling - Rbm: fixed weight decay - Rbm: fixed sparsity jens 2016-06-22 19:11:09 +00:00
  • 9d00654932 [RBM] - removed classes HiddenLayer.hpp and VisibleLayer.hpp - fixed warnings - RbmComponent inherits Rbm - improved Rbm::train - use gaussion weight initialization jens 2016-06-19 19:52:41 +00:00
  • 1705287e37 [RBM] - DBN fixes - batch sample inside training loop - implemented RbmComponent stacking jens 2016-06-18 17:57:32 +00:00
  • 46bffda38f [RBM] - added DBN stack - concentrated RBM params into structure jens 2016-06-17 22:14:55 +00:00
  • b10d06d96c [RBM] - refactored RbmComponent jens 2016-06-16 20:09:16 +00:00
  • ff88a62e73 [RBM] - pulled LayerArray out of RbmComponent jens 2016-06-16 19:16:33 +00:00
  • 4dccddd50f [RBM] - added getHiddenBatch() and getVisibleBatch() - cleaned up jens 2016-06-15 21:30:28 +00:00
  • 3a200cb46e [RBM] - fix: skip variable variance calculation if no training data is present jens 2016-06-15 20:53:54 +00:00
  • 2de8e9ac48 [RBM] - refactored and cleaned up jens 2016-06-15 20:32:28 +00:00
  • 3d56e4a163 [RBM] - MainComponent:fixed Jucer contention - RBM : clear before file load jens 2016-06-15 16:57:17 +00:00
  • 93b6144864 [RBM] - introduce RbmComponent jens 2016-06-14 18:52:08 +00:00
  • 80135d0498 [RBM] - moved sigma and mean from WEIGHTS to RBM - use Gibbs slider also for reconstruction draw jens 2016-06-13 18:05:17 +00:00
  • 82d93ca7b5 more stable jens 2016-06-05 22:04:13 +00:00
  • 40037f526d added jens 2016-06-02 18:33:20 +00:00
  • cf4fb2a9ef added project jens 2016-06-02 18:27:12 +00:00
  • f25f5b0dab fixed RBM jens 2016-05-30 23:53:18 +00:00
  • 7fada2a227 [RBM] - committed local changes jens 2015-05-28 17:41:06 +00:00
  • 549089a440 - added data normalization jens 2014-11-12 13:08:20 +00:00
  • 6e7b8bc151 - train(): added reconstruction error metric - train2(): - added reconstruction error metric. - added Gaussian units - added sparsity jens 2014-11-01 18:01:04 +00:00
  • 93fa8c64ce - bugfix: weightDecay needs to be multiplied by mu_weights - Biases are initialized with zero - developing full matrix calculation in train2() jens 2014-10-28 21:12:03 +00:00
  • 6520c80eca - LayerArray: added copy constructor - minor changes jens 2014-10-26 08:46:13 +00:00
  • 376f2dfa35 -autoscale of data during draw. data remains unchanged jens 2014-10-25 21:08:57 +00:00
  • 4d1f2f2e0c - changed gaussProb() jens 2014-10-25 15:56:59 +00:00
  • 6b2fe10d05 - correct energy calc. for BB-RBMs jens 2014-10-25 15:34:12 +00:00
  • 025a45eb83 - weight decay is 0.0 per default jens 2014-10-25 10:53:08 +00:00
  • 21701d103e - GUI: increased number of possible gibbs iterations jens 2014-10-24 22:01:04 +00:00
  • 6a0ef8d8f7 - no need to free RBM instance when weight topology changes jens 2014-10-24 06:10:54 +00:00
  • ba04927592 - added weight init std dev on GUI jens 2014-10-20 06:47:51 +00:00
  • 0f38b78093 - changed noise API - noise initialized with help of clock() jens 2014-10-19 20:28:36 +00:00
  • d0243820d6 - init weight with dev = 0.001 - layer load() plausibilty check jens 2014-10-19 14:58:24 +00:00
  • 56b0c4bee6 - minor GUI fix jens 2014-10-18 17:25:22 +00:00
  • 57761fe743 - RBM: changed calcualation of pre-weight update data in RBM - always use expectations - removed "Use Expectations Button" - removed Robbins-Monro - added sparsity learning rate - added momentum - added weight decay - added Slider as progress bar jens 2014-10-18 15:38:11 +00:00
  • 3fb8ac0629 - rearranged dot product jens 2014-10-16 18:35:32 +00:00
  • 288775cf8c - update working weights after sparse pass jens 2014-10-15 21:12:22 +00:00
  • afc1853664 - probsUpdateLogistic() use var for multiplication - automatic pattern reconstruction on pattern slider. Test button is obsolete - revised sparse mode jens 2014-10-15 20:35:49 +00:00
  • 005342e829 - RBM-Training has own thread - realtime update of Weights and reconstruction during training - realtime update of reconstruction on relevant parameters - added progress bar jens 2014-10-14 23:06:34 +00:00
  • 7f4438d698 - added param sparsity - added gaussian visible unit - added param sigma decay - RBM modi and params are set using members - use sigma instead of variance jens 2014-10-14 21:23:01 +00:00
  • 0166b986cb - vectorized logSigmoid() and gaussProb() - added switch RBM_SPARSE - added some experimental expect functions jens 2014-10-12 18:18:09 +00:00
  • 19c69ac02a - use Matrix, linear algebra library Eigen 3.2.2 jens 2014-10-12 14:30:40 +00:00
  • 7b1a713adc - changed project settings: release is optimized for speed jens 2014-10-08 18:48:02 +00:00
  • fc1f53fdb6 - added useProbsForHiddenReconstruction jens 2014-10-08 18:45:15 +00:00
  • da30d95e31 - added Robbins-Monro (but doesn't work well) - added DrawListener - added realtime reconstruct - additional LayerArray constructor jens 2014-10-07 20:42:57 +00:00
  • 39bf50701b - added button and functionality for Rao-Blackwellized weight update - added button for Robins-Monro weight update jens 2014-10-07 06:50:30 +00:00
  • fc758b853a - use expectations - improved gibbs sampling - LayerArray is template class jens 2014-10-06 17:08:50 +00:00
  • f62f1283f8 further development jens 2014-10-04 19:09:35 +00:00
  • b403b9776a - added jens 2014-09-26 05:33:17 +00:00
  • b1a0c90ea0 - added grayscale display - reconstruction and hidden probs can be displayed jens 2014-09-26 05:30:16 +00:00
  • 470437eecb - initial version jens 2014-09-24 17:21:29 +00:00