pub_soft/foldingdiff
3
0
Fork 0
mirror of https://github.com/microsoft/foldingdiff.git synced 2026-06-04 21:34:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 21ae033e95 Reimplement noising and modulo process Kevin Wu 2022-08-03 23:37:01 +00:00
  • eaa9498245 Change shifting strategy to be just add pi Kevin Wu 2022-08-03 22:49:09 +00:00
  • 827dd61ef9 Reimplment where shift to [0, 2pi] happens in cath dataset Kevin Wu 2022-08-03 22:46:27 +00:00
  • 9c00adcd5a More tests for loss Kevin Wu 2022-08-03 22:07:43 +00:00
  • 2dcbb3803d Code cleanup Kevin Wu 2022-08-03 21:49:03 +00:00
  • c33a5165f0 Code to compute modulo with negative min Kevin Wu 2022-08-03 21:47:25 +00:00
  • 72d308bc7b Simplify radian smooth L1 loss implementatoin Kevin Wu 2022-08-03 21:00:53 +00:00
  • c1b5276083 Log learning rate and some fixes Kevin Wu 2022-08-03 20:56:30 +00:00
  • d462025fe0 More bugfix Kevin Wu 2022-08-03 20:41:11 +00:00
  • 0837ee7460 Bugfix Kevin Wu 2022-08-03 20:38:57 +00:00
  • f8506a215d Implement single timestep training Kevin Wu 2022-08-03 20:19:29 +00:00
  • 7a3b04ef44 Bugfix Kevin Wu 2022-08-03 03:29:15 +00:00
  • 285f59c95c Update default to cosine variance schedule Kevin Wu 2022-08-03 01:30:47 +00:00
  • 42b8b6cb39 Bugfix and add docs Kevin Wu 2022-08-03 00:38:39 +00:00
  • 767f25628f Plotting of alpha bar t Kevin Wu 2022-08-03 00:07:20 +00:00
  • 0de8455b4b Fix noise to be zero centered even with modulo Kevin Wu 2022-08-02 23:21:20 +00:00
  • 951f413902 Update default params Kevin Wu 2022-08-02 22:50:15 +00:00
  • 3a0deb641c Plot losses at end of training Kevin Wu 2022-08-02 22:43:50 +00:00
  • 2b1ac150dd Make loss plotting code more general Kevin Wu 2022-08-02 22:43:30 +00:00
  • f04ef224c9 Correctly use singlular loss when doing single angle debug Kevin Wu 2022-08-02 22:06:09 +00:00
  • 93bd337c04 Implement running a single angle to debug Kevin Wu 2022-08-02 21:49:35 +00:00
  • 157d01c612 Better error logging Kevin Wu 2022-08-02 21:37:02 +00:00
  • cef30c8f19 Some checks Kevin Wu 2022-08-02 20:56:03 +00:00
  • 586db9625d Change how we shift angles to [0, 2pi] -- modulo instead of add Kevin Wu 2022-08-02 20:53:42 +00:00
  • 6caa6a4166 Support for alternative gaussian fourier time encoding Kevin Wu 2022-08-02 19:50:07 +00:00
  • 537ab1a066 Bugfix Kevin Wu 2022-08-02 01:01:33 +00:00
  • 3058e33b8e Configurable position embedding type Kevin Wu 2022-08-02 00:58:05 +00:00
  • fea4fb158d Add embeddings, simply code Kevin Wu 2022-08-02 00:57:36 +00:00
  • f42f2e1100 Properly initialize weights Kevin Wu 2022-08-02 00:27:00 +00:00
  • a6cff80e84 Simplify logic for running small datasets Kevin Wu 2022-08-01 23:57:12 +00:00
  • 6352929793 Log training loss Kevin Wu 2022-08-01 20:07:23 +00:00
  • ccea67a53e Improved logging Kevin Wu 2022-08-01 20:06:50 +00:00
  • fe222e6d6c Configure whether to use exhaustive t evaluation Kevin Wu 2022-08-01 19:32:01 +00:00
  • d1d312d287 Support for subsetting training set Kevin Wu 2022-08-01 19:23:39 +00:00
  • dcd412cf32 Initial handwavy implementation of score matching Kevin Wu 2022-08-01 18:52:31 +00:00
  • 9d9f457d48 Add some comments Kevin Wu 2022-07-31 22:04:34 +00:00
  • bff6e5ef97 Initial import Kevin Wu 2022-07-31 21:42:24 +00:00
  • 84e74ef406 Use exhaustive timesteps for validatoin/test sets Kevin Wu 2022-07-29 18:36:00 +00:00
  • 375bdeab97 Noising object with exhaustive timesteps Kevin Wu 2022-07-29 18:21:40 +00:00
  • bffa4cd2d5 Separately specify min and max epochs Kevin Wu 2022-07-29 17:16:15 +00:00
  • f154d773a8 Support for stochastic weight averaging Kevin Wu 2022-07-29 17:14:16 +00:00
  • c38fd2c0eb Refactor callbacks code Kevin Wu 2022-07-29 16:47:46 +00:00
  • 805e9d6252 Code to extract and plot losses Kevin Wu 2022-07-29 15:57:08 +00:00
  • c8cbdd7b02 Typo fix Kevin Wu 2022-07-29 04:14:30 +00:00
  • f3b88ea63e Make training script more convenient Kevin Wu 2022-07-29 04:06:10 +00:00
  • e5aa510148 Helper function to update args Kevin Wu 2022-07-29 04:03:30 +00:00
  • 69c18df129 Fix issue with modulo operation not actulaly happening in noised datasets Kevin Wu 2022-07-29 03:37:08 +00:00
  • 3d77b35ff3 Ability to disable cache Kevin Wu 2022-07-28 21:44:36 +00:00
  • f3068556a2 More robustly determine cached filename Kevin Wu 2022-07-28 21:35:22 +00:00
  • ef805888ed logging Kevin Wu 2022-07-28 21:33:22 +00:00
  • ba0954ce30 Allow configuring recomputing for AlphaFold dataset Kevin Wu 2022-07-28 21:32:12 +00:00
  • 5f1bec6938 Cache the results of reading pdb files Kevin Wu 2022-07-28 21:29:37 +00:00
  • 8e8b7dfecc Report lengths Kevin Wu 2022-07-28 21:02:05 +00:00
  • 590cad3811 Initial implementation of AlphaFold dataset Kevin Wu 2022-07-28 20:57:25 +00:00
  • 9923a39e2d Ignore alphafold dataset pdb gz files Kevin Wu 2022-07-28 18:22:02 +00:00
  • b986be7a78 Hyperparam sweep for smaller model Kevin Wu 2022-07-28 05:21:29 +00:00
  • 349117da43 Use abspath of exec script Kevin Wu 2022-07-28 05:21:18 +00:00
  • 58164500ee Reduce default beta value for smooth L1 loss Kevin Wu 2022-07-27 22:26:41 +00:00
  • 3b572ca6d2 Additional test Kevin Wu 2022-07-27 22:09:45 +00:00
  • 333d852af8 Tweak CLI Kevin Wu 2022-07-27 22:04:15 +00:00
  • e19d1335f5 Script to get alphafold data Kevin Wu 2022-07-27 21:31:52 +00:00
  • b8ad46a6f1 Report Wasserstein distance for bond/angle comparison Kevin Wu 2022-07-27 20:39:50 +00:00
  • 2a722f5c87 Dynamic loading of bert config for sampling Kevin Wu 2022-07-27 20:39:40 +00:00
  • 3157e0185d Support architectural params Kevin Wu 2022-07-27 19:44:29 +00:00
  • 0b735de088 Update some default params and configs Kevin Wu 2022-07-27 19:37:43 +00:00
  • 8d470562dd Use AdamW optimizer by default Kevin Wu 2022-07-27 19:36:21 +00:00
  • e8e0dfefe3 Add docs Kevin Wu 2022-07-27 19:18:19 +00:00
  • 59a339fda4 Simplify modulo in noise generation Kevin Wu 2022-07-27 19:01:42 +00:00
  • 9bcbf4159c Use the simplified broadcast mod when sampling Kevin Wu 2022-07-27 18:53:43 +00:00
  • fdeac034dd Handle 0 values Kevin Wu 2022-07-27 18:27:21 +00:00
  • 04ed86692c Efficient metohd for broadcasted mod Kevin Wu 2022-07-27 18:21:37 +00:00
  • e306e5e351 Fix sampling so that iterations don't fall off manifold Kevin Wu 2022-07-27 00:17:51 +00:00
  • 445523ca51 Configurable noise prior Kevin Wu 2022-07-26 22:26:30 +00:00
  • 8b56a239bf Initial implementation of uniform noise generator Kevin Wu 2022-07-26 21:21:33 +00:00
  • 356f3caa7a Fix off by one error Kevin Wu 2022-07-26 19:06:27 +00:00
  • b583bcd223 Simplify some plotting Kevin Wu 2022-07-26 19:01:10 +00:00
  • 949848c611 Configurable shifting to [0, 2pi] on training config Kevin Wu 2022-07-26 19:01:01 +00:00
  • dce31a6205 Create plots of training distributions Kevin Wu 2022-07-26 18:34:52 +00:00
  • 0a5e80e352 Save fig Kevin Wu 2022-07-26 18:10:01 +00:00
  • ceaf258516 More typo fix Kevin Wu 2022-07-26 00:07:37 +00:00
  • fff5763f2a Typo fix Kevin Wu 2022-07-26 00:07:23 +00:00
  • a9f407d140 Simplify some duck typing Kevin Wu 2022-07-25 23:59:56 +00:00
  • d7fd6cc656 Misc. changes, update default loss to radian smooth L1 Kevin Wu 2022-07-25 23:57:24 +00:00
  • 67bca5529a Implement sampling and generating noise with modulo in mind Kevin Wu 2022-07-25 23:53:24 +00:00
  • 6c4392658b Add noise modulo for initial sampling starting point Kevin Wu 2022-07-25 23:33:48 +00:00
  • 8ea8a9fae8 Refactoring ot allow for adaptive noise sampling Kevin Wu 2022-07-25 22:49:34 +00:00
  • b6a31422f2 Centralize code for plotting vals Kevin Wu 2022-07-25 22:37:13 +00:00
  • 0240f565f4 Typo fix Kevin Wu 2022-07-25 21:36:27 +00:00
  • c27733c63e Smooth L1 radian loss Kevin Wu 2022-07-25 21:32:23 +00:00
  • 21ca2b10a1 Save the BertConfig Kevin Wu 2022-07-25 21:05:35 +00:00
  • 31cc02c2ce Support for running without a config json Kevin Wu 2022-07-25 19:10:38 +00:00
  • b435ae587c Update some default values Kevin Wu 2022-07-25 19:03:21 +00:00
  • 0de5981178 Bond angle analysis with dynamic length sampling Kevin Wu 2022-07-25 18:37:06 +00:00
  • 6503ab2fc0 Properly trim by sequence length Kevin Wu 2022-07-25 18:19:47 +00:00
  • d06e5a4f99 Allow sampling of lengths from cath dataset Kevin Wu 2022-07-25 18:02:04 +00:00
  • 0a424115b2 Add type annotations Kevin Wu 2022-07-21 23:11:26 +00:00
  • a9c86f10f3 Return position IDs Kevin Wu 2022-07-21 23:09:44 +00:00
  • daf13186b3 Attention masks in sampling Kevin Wu 2022-07-21 23:07:53 +00:00
  • 987bc45968 More checkpointing Kevin Wu 2022-07-20 03:25:47 +00:00
  • 86e78bec3e Log the four separate losses Kevin Wu 2022-07-19 01:09:48 +00:00