d2913fd6f1
converting to CPU
2019-10-08 13:09:02 -04:00
dda7013f07
returning predictions from the model during eval at every timestep
2019-10-08 12:56:20 -04:00
46b552e075
updated eps value
2019-10-08 02:44:13 -04:00
02fb2430f0
removed logging of every horizon
2019-10-08 02:10:59 -04:00
765142de00
refactored
2019-10-07 20:55:26 -04:00
5d7694e293
logging to 4 decimals
2019-10-07 20:54:58 -04:00
f6e6713f74
fixed range bug
2019-10-07 20:40:54 -04:00
2560e1d954
Added per timestep loss
2019-10-07 20:03:00 -04:00
3d93008a3e
improved saving and restoring of model
2019-10-07 11:56:14 -04:00
a5a1063160
fixed docstring
2019-10-07 10:48:48 -04:00
5509e9aae5
Ensured all parameters are added to the optimizer
2019-10-07 09:47:38 -04:00
de42a67391
added logging statement
2019-10-07 07:59:41 -04:00
941675d6a7
Added kwargs
2019-10-06 18:57:13 -04:00
96d8dc4417
handling nans in loss tensor
2019-10-06 18:55:35 -04:00
5dd0f1dd3a
implemented masked mae loss, added tensorflow writer, changed % logic
2019-10-06 18:08:13 -04:00
a8814d5d93
Added docstrings
2019-10-06 17:12:06 -04:00
9fb999c3bb
squash! Added dcrnn_cell
2019-10-06 17:01:49 -04:00
d1964672c2
Added dcrnn_cell
...
Rough implementation complete - could forward pass it through the network
Ensured sparse mm for readability, logging sparsely as well
moving tensors to GPU
moving tensors to GPU [v2]
moving tensors to GPU [v3]
logging and refactor
logging and refactor
logging and refactor
logging and refactor
logging and refactor
logging and refactor
logging and refactor
ensured row major ordering
fixed log message
2019-10-06 17:00:23 -04:00
b65df994e4
Added dcrnn_cell
2019-10-06 11:55:02 -04:00
e80c47390d
Merge branch 'pytorch_implementation' into pytorch_scratch
2019-10-06 11:49:49 -04:00
5a790d5586
cuda no grad
2019-10-04 23:30:10 -04:00
593e3db1bf
Using model.cuda() if cuda is available
2019-10-04 22:45:08 -04:00
8d3b1d0d66
Implemented lr annealing schedule
2019-10-04 21:18:05 -04:00
ba880b8230
Implementing load and save models and early stopping
2019-10-04 17:25:03 -04:00
d9f41172dc
Implemented eval and function
2019-10-04 17:07:38 -04:00
20c6aa5862
Fixed bugs with refactoring
2019-10-04 16:05:52 -04:00
2b8d5e6b31
Refactored code and moved everything into a DCRNN forward pass
2019-10-04 13:02:50 -04:00
f41dc442b0
Implemented gradient clipping and returning output from training one batch
2019-10-03 19:35:54 -04:00
9834b12d5a
Cleaned up code, fixed bugs in implementation, seems like it starts training with GRU
2019-10-02 22:20:43 -04:00
f96a8c0d59
Dirty commit - setup model but [GRUCell] not working, tried ParameterList, did not work
2019-10-02 18:09:33 -04:00
c876cbfba3
Setup training loop and logging
2019-10-02 17:34:07 -04:00
86c4c5704d
Implemented single batch forward pass
2019-10-01 22:47:59 -04:00
69288460b1
Simplified encoder decoder model and moved curriculum learning outside
2019-10-01 10:07:34 -04:00
a1c9af2bad
Setup curriculum learning framework
2019-09-30 21:58:55 -04:00
bdce241a8f
Implemented abstract method and changed scheme to do all layers first for each timestep
2019-09-30 20:32:31 -04:00
0769a3b2e2
Implemented seq2seq without DCGRU and curriculum learning
2019-09-30 19:14:24 -04:00
7349f2ed67
Figured out decoder shapes
2019-09-30 16:04:03 -04:00
6386ac7eb4
Implemented encoder using GRUCell instead so that it's easier to swap that with DCGRUCell
2019-09-29 17:40:52 -04:00
66fb202d21
Implemented Encoder with GRU - should swap GRU with DCGRU
2019-09-29 12:51:49 -04:00
7a5e3c0216
model partially implemented - does not work yet
2019-09-29 11:13:08 -04:00
adbfa19146
Implemented DCGRUCell in pytorch (untested)
2019-09-08 19:28:20 -04:00
00c70b3a27
Implemented fc layer and changed docker image to use pytorch
2019-09-08 18:47:19 -04:00
7ba7fa320d
Using pytorch image
2019-09-07 17:53:46 -04:00
81b4626193
Remove unused code.
2019-01-18 19:00:34 -08:00
d59d44e4f0
Code refactor.
2018-10-01 17:45:46 -07:00
7e05414d99
Combine Val and Test model.
2018-09-30 22:14:45 -07:00
5dc36fed7c
Refactor DCRNN Model.
2018-09-30 21:53:40 -07:00
e8d6c08db0
Merge log_helper into utils and change logging mechanism.
2018-09-26 11:33:21 -07:00
9ec161543b
Code refactoring, including data loading, logging, configuration, removing redundant code.
2018-09-26 11:19:00 -07:00
e08002d72d
Update the seq2seq training logic.
2018-06-07 11:16:33 +08:00
Chintan Shah
Chintan Shah
Yaguang