Jens Ahrensfeld

jens pushed to master at jens/pyRBM

2026-06-05 18:22:44 +02:00

a41ddce6a8 scale up model, expand vocab, collect predicted text in forward pass

5ac4608ba8 tune hyperparams and fix forward_step shift, re-enable training

f7ed8563d7 refactor RnnModel: multi-unit delay training and unified vc interface

Compare 3 commits »

jens pushed to master at jens/pyRBM

2026-06-05 11:11:43 +02:00

b20ea4edb6 fixed wrong number of bits for binary encoding

jens pushed to master at jens/pyRBM

2026-06-05 10:07:25 +02:00

49a99dc750 tune label fitter: epochs, momentum, encoding, decode threshold

jens pushed to master at jens/pyRBM

2026-06-05 10:03:30 +02:00

248ae4f199 tune hyperparams, add README, simplify training loop

jens pushed to master at jens/pyRBM

2026-06-04 23:20:21 +02:00

b5a128fb61 fixed error for other WIN sizes

jens pushed to master at jens/pyRBM

2026-06-04 19:31:16 +02:00

06ab8d0166 refactored forward_step

7f6818670e - changed sample text

62dd450c9c increased difficulty

4bd516798a cleaned up

bb6680c931 getting Rnn to work

Compare 5 commits »

jens pushed to master at jens/pyRBM

2026-06-03 23:02:01 +02:00

851bad28d3 cleaned up

3a41014d50 improved context update for JayRnn

Compare 2 commits »

jens pushed to master at jens/pyRBM

2026-06-03 22:14:53 +02:00

2701daf1c4 added JayRnn

6a75e0b33b added more Rnn helber

5c3734a2a4 extend model member detection by list[Entity]

27460c503d fixed type alias

c62605de51 [update] README.rbm.md: add Q&A on image/video and code generation architectures

Compare 7 commits »

jens pushed to master at jens/pyRBM

2026-06-02 21:33:01 +02:00

3edc124a5c [refactor] move tests to tests/, add pytest functions and main() entrypoints

jens pushed to master at jens/pyRBM

2026-06-02 21:07:05 +02:00

829547d271 [notebook] update cifar notebook outputs and add N_HID constant

0b4794de4e [fix] use np.asnumpy in SubImageCombine to support CuPy arrays

Compare 2 commits »

jens pushed to master at jens/pyRBM

2026-06-02 08:55:52 +02:00

1954064a54 [refactor] move rbm.model → model.model

jens pushed to master at jens/pyRBM

2026-06-02 08:36:46 +02:00

b41274c7a6 [refactor] move rbm.torch → compat.torch

3d235b4214 [refactor] move rbm.label → label.label

Compare 2 commits »

jens pushed to master at jens/pyRBM

2026-06-02 08:24:01 +02:00

807de19fa7 [fix] restore SubImageCombine name in cifar notebook (clobbered by bulk rename)

a07698497d [refactor] SubImage → SubImageExtract, add SubImageCombine

2da2262368 [refactor] move rbm.stack* + rbm.rnn_helper → stack/

b0d370a56b [refactor] move rbm.image → image.sub_image

Compare 4 commits »

jens pushed to master at jens/pyRBM

2026-06-01 20:07:14 +02:00

599ba4f6d5 [cifar_sub_image] run with STRIDE=6 (25 patches/image, ~33% overlap)

jens pushed to master at jens/pyRBM

2026-06-01 19:59:45 +02:00

dddb937909 [cifar_sub_image] add full 32x32 reconstruction from sub-images with overlap blending

00993e29f1 [rnn_helper] add shift functions, ch2idx/idx2ch, and tests

cce3943b27 [RNN-RBM] rename to RNN-RBM, add unrolled depth-2 mode, update README

Compare 3 commits »

jens pushed to master at jens/pyRBM

2026-05-31 20:55:56 +02:00

73ff6c2bb4 [context33] - momentum 0.9, epochs 500

jens pushed to master at jens/pyRBM

2026-05-31 20:07:10 +02:00

4a915f165a [context33] - increase momentum 0.5→0.7; run outputs 99% recon, 88% next-step

586803856f [context33] - add batch tiling (N_REPEAT=500) for GPU utilisation

bbc03d37e1 [context33] - 200-char sentence, auto-uppercase, adaptive tick spacing

0beb7a5c1b [context33] - replace armadillo data with 20-char sentence, moby encoding

5506fef0fa [context33] - fix row order; sequence now reads 1 5 6 9 … R U X

Compare 11 commits »

jens pushed to master at jens/pyRBM

2026-05-30 22:42:03 +02:00

37c6e33948 [CheckpointStatus] - add checkpoint status handler; Model.train() accepts status param

jens pushed to master at jens/pyRBM

2026-05-30 22:26:10 +02:00

2f9fe1b66b [bugfix] - support large CPU-side datasets; fix numpy/CuPy interop

jens pushed to master at jens/pyRBM

2026-05-30 21:15:05 +02:00

8c10bda002 [README] - update faces_sub_image with training results and CLI args