0.2.0
New features
- [Feature] First class dim compatibility by @vmoens in #525
- [Feature] CompositeDistribution by @vmoens in #517
- [Feature] torch.where by @vmoens in #507
- [Feature]
dense_stack_tds
by @matteobettini in #506 - [Feature] TensorDictParams by @vmoens in #500
- [Feature] Hooks and Buffers for TensorDictParams by @vmoens in #502
- [Feature] PyTree compatibility by @vmoens in #501
- [Feature] from_module by @vmoens in #423
- [Feature] Prevent lock/unlock conflicts for stacked and nested tensordicts by @vmoens in #436
- [Feature] Lock check decorator by @vmoens in #440
- [Feature] Cache for locked tensordicts by @vmoens in #438
- [Feature] tensordict.transpose by @vmoens in #467
What's Changed
-
[Benchmark] Benchmark flatten / unflatten by @vmoens in #439
-
[Benchmark] Benchmark items, values and keys by @vmoens in #428
-
[Benchmark] Benchmark vmap and stacked tds by @vmoens in #430
-
[Benchmark] More benchmarks for locked tds by @vmoens in #437
-
[Benchmark] Total time per benchmark for efficiency by @vmoens in #446
-
[BugFix] Fix EnsembleModule parameters exposure by @vmoens in #508
-
[BugFix] Fix error for incongruent devices in load_state_dict by @vmoens in #529
-
[BugFix] Fix lazy stack / stack_onto + masking lazy stacks by @vmoens in #497
-
[BugFix] Fix lazy-stack to call to to_tensordict() by @vmoens in #519
-
[BugFix] Fix logical operations in MemmapTensor by @vmoens in #484
-
[BugFix] Fix regex in tensorclass error checks by @vmoens in #398
-
[BugFix] Fix writing on stacked tds in vmap by @vmoens in #456
-
[BugFix] Heterogeneous lazy stack apply(_) by @vmoens in #400
-
[BugFix] Keyword arguments with dispatch + select keys by @vmoens in #409
-
[BugFix] Nested keys to probabilistic modules by @matteobettini in #479
-
[BugFix] Retrieve all keys for
LazyStackedTensorDict
by @matteobettini in #512 -
[Bugfix] Indexing memmaps with torch.Tensor by @tcbegley in #383
-
[CI] Fix Doc CI bis by @matteobettini in #490
-
[CI] Fix doc CI by @matteobettini in #489
-
[CI] Removing myself from notified users in benchmarks by @tcbegley in #483
-
[Feature] Add reset_parameters support to TensorDictModuleBase by @smorad in #476
-
[Feature] Better error message in tdmodule by @vmoens in #390
-
[Feature] Cat LazyStackedTensorDicts by @matteobettini in #499
-
[Feature] Create memmaps folders with makesdir by @vmoens in #431
-
[Feature] Module to add state independent normal scale by @albertbou92 in #515
-
[Feature] Print lazy fields by @matteobettini in #493
-
[Feature] Registering a tensordict as a module buffer by @vmoens in #395
-
[Feature] Support classmethods in tensorclass by @vmoens in #448
-
[Feature] as_decorator and enter exit methods by @vmoens in #455
-
[Minor] Add dist/ to gitignore. by @skandermoalla in #492
-
[Minor] Better errors for kwargs and key selection by @vmoens in #415
-
[Minor] Key filtering error fixes during deletion by @vmoens in #481
-
[Performance] Better names handling in LazyStackTD by @vmoens in #482
-
[Performance] use dict for modules in swap states by @vmoens in #419
-
[Refactor] Better keys for cache to reduce mem footprint by @vmoens in #454
-
[Refactor] Faster indexing of memmap tensors by @vmoens in #433
-
[Refactor] Minor refactorings to tensorclass by @vmoens in #435
-
[Refactor] Refactor to create empty tds where necessary by @vmoens in #522
-
New Contributors
-
@matteobettini made their first contribution in #479
-
@skandermoalla made their first contribution in #492
-
@albertbou92 made their first contribution in #515
Full Changelog: v0.1.2...v0.2.0