Commit Graph

262 Commits

Author SHA1 Message Date
Yin Li
7be3153206 Revert removal of saving/loading optimizer & scheduler states
Why did I do that?
2020-09-12 18:30:07 -04:00
Yin Li
85efb9e3a3 Add autograd detect anomaly flag, and test cudnn.deterministic
cudnn.deterministic is only 10% slower
2020-09-12 18:26:44 -04:00
Yin Li
d8c6be797d Move figures from .data to .utils 2020-09-12 18:25:58 -04:00
Yin Li
154376d95a Add tgt_pad, rename pad to in_pad
tgt_pad can be useful for scale_factor > 1
2020-09-12 18:24:36 -04:00
Yin Li
39ad59436e Replace narrow_like by narrow_by in UNet/VNet
This makes it traceable / scriptable.
Note that the narrow_like in ResBlock used by VNet is not changed yet
2020-09-11 00:26:14 -04:00
Yin Li
c4ab7e065b Add plt_slices(size=None) 2020-09-09 14:56:07 -04:00
Yin Li
6c62cb09db Fix error pickling normalization function in multiprocessing dataloading
By avoiding importing functions as class attributes but importing before
calling, and using cache in the importing function

Thanks to help by Yu Feng
2020-09-09 14:56:07 -04:00
Yin Li
0e0ad8f071 Fix import bug that prevents pickling of norm functions
https://docs.python.org/3/library/importlib.html#importing-a-source-file-directly
https://stackoverflow.com/questions/43819653/pickling-objects-imported-with-importlib-util/53696167#53696167
2020-09-09 14:56:06 -04:00
Yin Li
f3216f4cbd Add multiple pkgs to import_attr 2020-09-09 14:55:36 -04:00
Yin Li
79b28561d5 Fix two power spectrum bugs about freq shift and flatten 2020-08-23 15:50:24 -04:00
Yin Li
6c67eaa788 Fix cropping bug when scale_factor>1 2020-08-23 15:50:24 -04:00
Yin Li
670364e54c Change Lag2Eul to lag2eul as a function 2020-08-22 23:25:08 -04:00
Yin Li
3eb1b0bccc Add power spectrum tracking 2020-08-22 23:24:25 -04:00
Yin Li
afaf4675fe Add power spectrum module 2020-08-22 23:18:48 -04:00
Yin Li
5d22594ede Add shape report after narrow_cast in training 2020-08-22 12:28:08 -07:00
Yin Li
01cc1b6964 Change VNet based on experiment on displacement 2020-08-17 19:28:10 -07:00
Yin Li
ebd962e333 Fix module before __init__ bug 2020-08-12 09:10:58 -07:00
Yin Li
61ca400942 Add center slicing in plt_slices
Also fix bug of hard coded `size`
2020-08-11 13:05:38 -07:00
Yin Li
9c8b331bf5 Add loglxe loss as log of product of Lagrangian and Eulerian losses 2020-08-09 12:59:24 -07:00
Yin Li
13edf3b96d Add optional trailing activation to residual block 2020-08-09 12:18:17 -07:00
Yin Li
a1e5399311 Fix args helps 2020-08-04 21:15:22 -07:00
Yin Li
6b65b7888d Change color normalization in plt_slices 2020-08-04 21:13:13 -07:00
Yin Li
d6b7a528df Remove pyramid model 2020-08-04 21:13:13 -07:00
Yin Li
728e0597f0 Add documentation 2020-08-04 21:13:13 -07:00
Yin Li
265587922d Add srsgan model and scale_factor to model arguments 2020-08-04 21:13:10 -07:00
Yin Li
6dfde5ee7f Fix --optimizer-args None default bug 2020-07-31 17:31:55 -04:00
Yin Li
5ac9016987 Replace specific slurm scripts with general ones 2020-07-29 12:12:46 -07:00
Yin Li
b54fc4ba3a Add a line about socket hopefully will avoid port conflicts 2020-07-28 07:38:07 -07:00
Yin Li
0d41bdae26 Add plt.close to avoid matplotlib warning 2020-07-28 07:10:46 -07:00
Yin Li
a886e53c54 Always use training loss for lr scheduler 2020-07-27 17:47:58 -07:00
Yin Li
2336d83f2d Add optimizer and scheduler kwargs via json.loads 2020-07-27 12:40:19 -07:00
Yin Li
8ce13e67f6 Fix training hang due to constrained layout of matplotlib 2020-07-27 12:39:41 -07:00
Yin Li
bf9b0ba426 Merge remote-tracking branch 'refs/remotes/origin/lag2eul' into lag2eul 2020-07-22 18:48:27 -07:00
Yin Li
ca39b2028f Replace cuda.device with env var to assign GPU to processes 2020-07-22 18:48:12 -07:00
Yin Li
f98b0fbbd0 Merge remote-tracking branch 'refs/remotes/origin/lag2eul' into lag2eul 2020-07-22 02:38:45 -04:00
Yin Li
98cdd4795c Add DistFieldSampler to benefit from page cache 2020-07-21 23:37:59 -07:00
Yin Li
632c73db16 Merge branch 'lag2eul' of github.com:eelregit/map2map into lag2eul 2020-07-22 00:20:28 -04:00
Yin Li
9f54e02c3a Add DistFieldSampler to benefit from page cache 2020-07-21 21:13:52 -07:00
Yin Li
ec46f41ba5 Add lag x eul loss tracking to tensorboard 2020-07-21 21:10:34 -07:00
Yin Li
509856cb4f Revert "Deprecate GroupedRandomSampler for use of memmap in place of cache"
This reverts commit bbf77c9f91.
2020-07-18 18:07:38 -04:00
Yin Li
56f8fa932b Revert "Deprecate GroupedRandomSampler for use of memmap in place of cache"
This reverts commit bbf77c9f91.
2020-07-16 15:22:52 -07:00
Yin Li
5bc099251f Merge remote-tracking branch 'refs/remotes/origin/lag2eul' into lag2eul 2020-07-15 19:04:34 -07:00
Yin Li
2d5234812b Add lagrangian and eulerian alternate training 2020-07-15 22:03:30 -04:00
Yin Li
337d65de68 Remove adversary 2020-07-15 22:03:30 -04:00
Yin Li
607bcf3f4c Add Lag2Eul to training 2020-07-15 22:03:30 -04:00
Yin Li
40220e9248 Add Lagrangian-to-Eulerian transformation 2020-07-15 22:03:30 -04:00
Yin Li
ff4c2e12fd Change default number of loader workers 2020-07-15 22:03:14 -04:00
Yin Li
7d3a598080 Remove shape dependence from narrow_by
Good for torchscript
2020-07-15 21:36:37 -04:00
Yin Li
4160ed25f7 Add lagrangian and eulerian alternate training 2020-07-15 15:57:10 -04:00
Yin Li
e8744d6c7b Remove adversary 2020-07-14 21:07:05 -04:00