Commit Graph

21 Commits

Author SHA1 Message Date
23fddbe5b9 updated comment for PairwiseComparator 2025-12-25 14:53:35 +00:00
acbccebb2c renamed ADAMW_WiDECAY 2025-12-25 14:48:07 +00:00
edf8d46123 wut 2025-12-25 02:13:23 +00:00
4f500e8b4c write all data on training to training log 2025-12-25 02:07:49 +00:00
921e24b451 embeded # of hidden neurons in model save data
added more valves for options.
will update the console every 5 seconds when training now
2025-12-25 01:52:19 +00:00
d46712ff53 make the defaults more sane for the task
its still overkill
2025-12-25 01:23:54 +00:00
cd72cd7052 changes to figure sizes and add labels to embedding chart 2025-12-23 13:44:38 -05:00
755161c152 unused var 2025-12-23 13:29:52 -05:00
1d70935b64 general clean up & added help text
removed symbolic link calling path
2025-12-23 13:06:07 -05:00
0e2098ceec remove linked files 2025-12-23 12:30:18 -05:00
cfef24921d moved output_graphs into main script and restructured
training log is seperate from normal output & is compressed
slightly adjusted lr
made final test stage test 4xTRAIN_BATCHSZ number of samples
2025-12-23 10:18:02 -05:00
9ea8ef3458 default k=0.5 2025-12-23 10:14:08 -05:00
6e31865a84 added episilon for equality check
major layout changes in the network
2025-12-22 21:47:16 -05:00
997303028e output plots and csv data for e(x) over the trained range. 2025-12-22 12:23:59 -05:00
c3fbc44a34 add cpu pytorch instructions 2025-12-20 01:59:50 +00:00
5e5ad1bc20 fix log entry for model state saving
adjust default batch size and passes to be a bit more CPU pytorch friendly
2025-12-20 01:54:42 +00:00
0d6a92823a all files (logs/models/images) go into ./files
using AdamW optim again, AdamW is the go to for small toys and transformers.
refactored NN classes to thier own module under pairwise_comp_nn.py
2025-12-19 09:15:41 -05:00
cfcec07b9c updated .gitignore 2025-12-19 09:13:07 -05:00
bf15e7b6c8 save the plots for viewing 2025-12-18 22:24:23 -05:00
0385b0acc8 testing adadelta optimizer 2025-12-18 22:24:01 -05:00
5c89149388 first commit 2025-12-18 20:05:57 -05:00