|
|
197de7997c
|
Fixed bug with SDE
|
2022-08-22 13:36:17 +02:00 |
|
|
|
a9e3f295b2
|
Fixed numerical issues with Wasserstein
|
2022-08-17 23:25:24 +02:00 |
|
|
|
9fffe048af
|
Fixed Spherical_Chol not accepting batches
|
2022-08-17 22:55:42 +02:00 |
|
|
|
86e6bfb65b
|
Higher epsilon to deal with numerical instabilities
|
2022-08-17 19:31:54 +02:00 |
|
|
|
64a7d5ec59
|
Guarante minimum epsilon when ensuring non-zero (CholNet)
|
2022-08-16 20:02:33 +02:00 |
|
|
|
d35c3d8520
|
Fixed all the bugs in TRPL
|
2022-08-15 16:55:17 +02:00 |
|
|
|
28d0c609bc
|
Fixed SDE: sampling had dimension mismatches
|
2022-08-14 20:09:10 +02:00 |
|
|
|
e1c59cffd0
|
Removed debug-print
|
2022-08-14 18:45:17 +02:00 |
|
|
|
bb1f9ecf2b
|
Fixed UniversalGaussianDistribution lost SDE when cloning
|
2022-08-14 18:42:19 +02:00 |
|
|
|
0ee65e789b
|
Fixing sde's bugs
|
2022-08-14 16:10:22 +02:00 |
|
|
|
0e4eedae5e
|
Fixed gradient throught spherical-chol
|
2022-08-10 11:55:08 +02:00 |
|
|
|
520dc98eb5
|
Implemented SDE
|
2022-08-10 11:54:52 +02:00 |
|
|
|
12e422aec7
|
Why does KL double free?
|
2022-08-07 18:04:40 +02:00 |
|
|
|
75d73049b4
|
Fixing bugs with w2 and sqrt_induced_gaussian
|
2022-08-06 21:25:49 +02:00 |
|
|
|
802094a50f
|
Enabled w2 (can now get sqrt from dist)
|
2022-08-06 14:54:59 +02:00 |
|
|
|
508ebf51f0
|
Implemented sqrt-induced-gaussian for W2-Projection
|
2022-08-06 14:46:42 +02:00 |
|
|
|
54113fd40c
|
Removed unused dependency
|
2022-08-06 14:37:06 +02:00 |
|
|
|
2c1689fbbc
|
Fixed smol bugs when instanciating based on string-names of enums
|
2022-08-06 14:36:35 +02:00 |
|
|
|
8b82347056
|
Automatic casting to enums
|
2022-08-05 21:06:31 +02:00 |
|
|
|
683644f77d
|
Removed weird line...
|
2022-08-05 21:06:05 +02:00 |
|
|
|
a78e81b9e1
|
Allowing prob squashing
|
2022-07-21 09:42:25 +02:00 |
|
|
|
199ce0c8cb
|
ProbSquashing implemented (tanh)
|
2022-07-20 10:32:19 +02:00 |
|
|
|
0162a36824
|
Renamed TRL_SAC => SAC
|
2022-07-19 10:08:14 +02:00 |
|
|
|
7b667e9650
|
SAC is back; with SDC; without Projections
|
2022-07-19 10:07:50 +02:00 |
|
|
|
5f32435751
|
Smashing bugs: dont confuse chol with chol_net
|
2022-07-19 10:07:20 +02:00 |
|
|
|
b7de99b1fc
|
EnforcePositiveType makes no sense for Strength.NONE
|
2022-07-19 10:06:40 +02:00 |
|
|
|
49f9acff3e
|
Fixed: Wrong simplification for Hybrid[SCALAR=>FULL]
|
2022-07-17 00:47:47 +02:00 |
|
|
|
046fa78206
|
Fixed: _chol_from_sphe_chol was unable to handle batches
|
2022-07-16 17:34:25 +02:00 |
|
|
|
c141599662
|
Fixed Bug: _chol_from_sphe_chol dependet on action_dim, not able to use
for Hybrid methods
|
2022-07-16 15:47:09 +02:00 |
|
|
|
3fa6de7e66
|
Broader sampling of stds for logging with batched full covs
|
2022-07-16 15:28:16 +02:00 |
|
|
|
bc0e188a0d
|
Removed debugging-code
|
2022-07-16 15:19:56 +02:00 |
|
|
|
d2d84d3287
|
Fixed bug for logging std-estimates when using batched data
|
2022-07-16 15:18:24 +02:00 |
|
|
|
4a24381f46
|
Fixed bug when using batches with SPHERICAL_CHOL
|
2022-07-16 15:17:48 +02:00 |
|
|
|
04529e8261
|
Removed debug-point
|
2022-07-16 14:58:29 +02:00 |
|
|
|
4854346f2d
|
Fixed bug with logging of std for full-cov
|
2022-07-16 14:58:00 +02:00 |
|
|
|
cb9ee4f302
|
Fixed bugs for Hybrid[Diag=>Full]
|
2022-07-16 14:57:34 +02:00 |
|
|
|
ad584d70fd
|
Removed debugging prints
|
2022-07-16 13:07:08 +02:00 |
|
|
|
72754525cd
|
Allow using newly implemented hybrid method
|
2022-07-16 13:06:07 +02:00 |
|
|
|
fa167b3e5f
|
Hybrid Diag -> Full Implemented; Made spherical_chol more efficient
|
2022-07-16 13:05:35 +02:00 |
|
|
|
f184b88f19
|
Allow std logging for full and diagonal cov policies
|
2022-07-15 18:46:42 +02:00 |
|
|
|
74697e8773
|
Smol bug fixes
|
2022-07-15 18:46:17 +02:00 |
|
|
|
2e0f46b0f3
|
Fixing ser/deser bug (cloudpickle cant handle some enums)
|
2022-07-15 18:45:38 +02:00 |
|
|
|
a86d19053d
|
Smashing bugs (dimension mismatch between Normal and
Independent/MultivariateNormal)
|
2022-07-15 15:46:31 +02:00 |
|
|
|
ab557a8856
|
Making MultivariateNormal Policies work (and porting Normal to
Independent)
|
2022-07-15 15:03:51 +02:00 |
|
|
|
b1ed9fc2b8
|
Renamed TRL_PG to PPO
|
2022-07-13 19:51:33 +02:00 |
|
|
|
3304fd49f6
|
Working on UniversalGaussianDistribution
|
2022-07-13 19:38:57 +02:00 |
|
|
|
fae19509bc
|
Implemented Policies with Contextual Covariance
|
2022-07-13 19:38:20 +02:00 |
|
|
|
41e4170b2f
|
Fixes + spherical_chol
|
2022-07-11 17:28:08 +02:00 |
|
|
|
e4440428f8
|
Working on SDC
|
2022-07-11 11:55:23 +02:00 |
|
|
|
4c4b12ee0e
|
Allow cloning UniversalGaussianDistribution (new_dist_like)
|
2022-07-09 14:46:11 +02:00 |
|
|
|
c08ea1cb91
|
Making UniversalGaussianDistribution ready for tanh-squashing-support
|
2022-07-09 14:33:07 +02:00 |
|
|
|
249754ee89
|
Wrote a little helper-function to generate all allowed combinations of
cov-parameterizations
|
2022-07-09 14:03:56 +02:00 |
|
|
|
e09950b30c
|
Work on Contextual Covariances
|
2022-07-09 12:26:39 +02:00 |
|
|
|
0911a04e98
|
Factored out Gaussian Collection for RolloutBuffer
|
2022-07-04 21:21:16 +02:00 |
|
|
|
1f3eac5398
|
Cleanup
|
2022-07-02 16:42:14 +02:00 |
|
|
|
91f64c10d7
|
Fixed bug with initialization of buffer
|
2022-07-01 20:02:09 +02:00 |
|
|
|
14100cccc8
|
Working on SDC-impl.
|
2022-07-01 15:14:41 +02:00 |
|
|
|
84d1cda96c
|
Trying to get kl to work
|
2022-07-01 13:45:58 +02:00 |
|
|
|
2626ec82a6
|
Added code-src-note
|
2022-07-01 11:52:50 +02:00 |
|
|
|
ab1b269af9
|
Allow checking whether a dist is contextual
|
2022-07-01 11:52:14 +02:00 |
|
|
|
a8b9c63965
|
Making dez covariances contextual
|
2022-07-01 11:29:12 +02:00 |
|
|
|
2e378d0a7d
|
Rebranding to Metastable Baselines
|
2022-06-30 20:40:30 +02:00 |
|
