* Updated Project and DevProject to 2020.3 LTS.
* Updated supported LTS version to 2020.3.
* Updated changelog.
* Bumped up CI Unity version from 2020.2 to 2020.3.
* Update CHANGELOG.md
* Removing VS Studio dependency from manifest.json.
* Update packages-lock.json
* Updated versions for main. Also updated validate versions script to remove exp.1 from package version.
* Reverted validate versions.
* Updated versions.
* Bumped package version for patch.
* Update DQN GridWorld colab example to be capable of outputting an ONNX model that works for inference.
-Allow GridWorld to work without a GoalSensor. Defaults to the GreenPlus target in that case.
-Add GridWorldColab.onnx file, trained and generated with the colab.
* Upgrade DQN GridWorld Colab to produce model that can work with (modified) GridWorld Scene.
-permute observation images from NHWC to NCHW right after env_step. Everything following is NCHW.
-update input_shape indexes and remove equivalent permute from the VisualQNetwork
-fix reference to global qnet inside generate_trajectories (use the passed in version)
-create wrapper network to add required constants and dummy input masks required by the
runtime inference (Barracuda)
-export the wrapped network to ONNX format compatible with Barracuda
* Add additional GridWorld scene called GridWorldColab that has the goal sensors removed (target is always green plus) and masks disabled. The purpose of this is to have a scene compatible with the DQN GridWorld Colab example.
* Minor edit in Colab docs.
* Delete unnecessary .lighting file.
* Update changelog.
* Update colab/Colab_UnityEnvironment_2_Train.ipynb
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Fix error in colab metadata name.
* Add suggested comments and fix incorrect parameter ordering.
* Update docs for Elo curriculum completion measure.
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Progress on propagating the setting to the action model.
* Added the _sample_action logic and tests.
* Add information to the changelog.
* Prioritize the CLI over the configuration file.
* Update documentation for config file.
* CR refactor.
* Update docs/Training-Configuration-File.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
Update ml-agents/mlagents/trainers/settings.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
Update ml-agents/mlagents/trainers/cli_utils.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Fix CR requests
* Add tests for discrete.
* Update ml-agents/mlagents/trainers/torch/distributions.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Added more stable test.
* Return deterministic actions for training (#5615)
* Added more stable test.
* Fix the tests.
* Fix pre-commit
* Fix help line to pass precommit.
* support for deterministic inference in onnx (#5593)
* Init: actor.forward outputs separate deterministic actions
* changelog
* Renaming
* Add more tests
* Package changes to support deterministic inference (#5599)
* Init: actor.forward outputs separate deterministic actions
* fix tensor shape for discrete actions
* Add test and editor flag
- Add tests for deterministic sampling
- update editor and tooltips
* Reverting to "Deterministic Inference"
* dissect tests
* Update docs
* Update CHANGELOG.md
* Fix the deterministic showing up all the tiime (#5621)
Co-authored-by: Chingiz Mardanov <chingiz.mardanov@unity3d.com>
Co-authored-by: cmard <87716492+cmard@users.noreply.github.com>
As part of the package signing initiative, we'd like bundled packages to be prepared for upcoming constraints regarding the need to have packages with a valid signature during APV testing.
Because of this, we are going through all bundled packages' repositories creating a PR that would enable package signing on your packages.
With signing enabled, upm-ci will perform an extra step where it will send each package to a signing service and then generate the final signed artifacts.
Internally, in an attempt to ensure that everything works, we are running yamato tests where all bundled packages (including yours) are signed and validated, proving that the packages in this repository should yield no errors. The validation step is also performed in all platforms.
In the case that any error does arise, this can be simply reverted back by removing the environment variable, UPMCI_ENABLE_PACKAGE_SIGNING in the publishing jobs.
For more information, please refer to: https://confluence.unity3d.com/display/PAK/Package+Signature
You can contact us through:
#team-pkd-176-internal-package-signing
@cassandra & @felipemunoz
* Progress on propagating the setting to the action model.
* Added the _sample_action logic and tests.
* Add information to the changelog.
* Prioritize the CLI over the configuration file.
* Update documentation for config file.
* CR refactor.
* Update docs/Training-Configuration-File.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Miguel Alonso Jr. <76960110+miguelalonsojr@users.noreply.github.com>
Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
Update ml-agents/mlagents/trainers/settings.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
Update ml-agents/mlagents/trainers/cli_utils.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Fix CR requests
* Add tests for discrete.
* Update ml-agents/mlagents/trainers/torch/distributions.py
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Added more stable test.
* Return deterministic actions for training (#5615)
* Added more stable test.
* Fix the tests.
* Fix pre-commit
* Fix help line to pass precommit.
* support for deterministic inference in onnx (#5593)
* Init: actor.forward outputs separate deterministic actions
* changelog
* Renaming
* Add more tests
* Package changes to support deterministic inference (#5599)
* Init: actor.forward outputs separate deterministic actions
* fix tensor shape for discrete actions
* Add test and editor flag
- Add tests for deterministic sampling
- update editor and tooltips
* Reverting to "Deterministic Inference"
* dissect tests
* Update docs
* Update CHANGELOG.md
Co-authored-by: Chingiz Mardanov <chingiz.mardanov@unity3d.com>
Co-authored-by: cmard <87716492+cmard@users.noreply.github.com>
* Added normalization to critic during training for poca, ppo, and sac.
* Fixed critic normalization bug. Added new tests to cover.
* Updated CHANGELOG.
* Fixed typo in test_trainers.py
* Fixing the analytics side-channel for curriculum learning.
* Made a more robust test.
* Update the changelog.
* Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
Co-authored-by: Maryam Honari <honari.m94@gmail.com>
* Added training area replicator to com.unity.ml-agents package.
* Added num_areas to Unity RL Initialization proto. Added cli and config file support for num_areas.
* Changed training area replicator to size grid automatically from number of areas.
* Added tests for the training area replicator.
* Added setup for tests for the training area replicator.
* Added comments and updated create tutorial for training area replicator.
* Updated CHANGELOG.
* Fixed some failing tests.
* Update com.unity.ml-agents/CHANGELOG.md
Co-authored-by: Henry Peteet <henry.peteet@unity3d.com>
* Update docs/Learning-Environment-Create-New.md
Co-authored-by: Henry Peteet <henry.peteet@unity3d.com>
* Update com.unity.ml-agents/Runtime/Areas/TrainingAreaReplicator.cs
Co-authored-by: Henry Peteet <henry.peteet@unity3d.com>
* Addressed CR comments.
Co-authored-by: Miguel Alonso Jr <miguelalonsojr>
Co-authored-by: Henry Peteet <henry.peteet@unity3d.com>
### Change(s)
Update the SubprocessEnvManager to restart workers when the underlying Unity environments crash.
When a worker receives an ENV_EXITED signal it will now:
1. Record all failures coming through the step queue and drop all other messages.
2. Purge any pending trajectories as they may belong to a crashed worker or be corrupted.
3. Restart all failed workers (up to a configurable limit)
This behavior can be limited via a rate limit, max lifetime limit, or both. The configuration options for both are shown below with their default values.
⚠️ Each of these options applies to a single environment, if num_envs > 1 then the limit will apply separately to each replica (num_envs = 2 will spawn 2 Unity environments which can each be restarted 10 times).
```yaml
env_settings:
# Can restart 10 times over the lifetime of the experiment.
max_lifetime_restarts: 10
# Rate limit of 1 failure per 60s
restarts_rate_limit_n: 1
restarts_rate_limit_period_s: 60
```
They can of course be passed via CLI arguments as well
```bash
--max-lifetime-restarts
--restarts-rate-limit-n
--restarts-rate-limit-period-s
```
### Disabling this feature
* Rate limiting can be turned off by setting `--restarts-rate-limit-n=-1`
* Lifetime limiting can be turned off by setting `--max-lifetime-restarts=-1`
* Hash128 is not a cryptographic hash, replace with HMAC-SHA256.
* Extend TrainingAnalytics side channel to expose configuration details
* Change member function scopes and hash demo_paths
* Extract tbiEvent hashing method and add test coverage
* Added initial (untested) support for separating schedule for beta, LR, and epsilon.
* Added support for beta and epsilon schedule in test config conversion
* Updated docs. Set epsilon to be linear by default.
* Updated configs to add beta and epsilon schedule where appropriate.
* Addressed CR change requests.
* Minor update.
* Updated docs.
* Added changes to CHANGELOG.
* Updated CHANGELOG.
* Fixed typo in Training-Configuration-File.md
* Reverted config changes to remove beta and epsilon schedules.
* mypy error fixed from pre-commit.
* Fixed a bug that was causing a settings test to fail.
* Addressed CR comments.
Co-authored-by: Miguel Alonso Jr <miguel.alonso@miguel.alonso-00MM>
Co-authored-by: Miguel Alonso Jr <miguelalonsojr>
Pre-commit is using a version of dotnet and dotnet-format and has picked up a newer version of the utility that is not compatible with .NET 3.1 or 5.x -- it installs it's packages using dotnet tool install. This version's default entrypoint is also incompatible with our project structure, as the style and analyzer subcommands are now run by default and do not support the --folder option.
The entrypoint can be over-ridden to specify the whitespace module, which should match previous behavior.
Miguel Alonso Jr
Miguel Alonso Jr
Xun Cao
Jason Rupert
milant
Henry Peteet
Maryam Honari
Cassandra Lacombe
cmard
Marwan Mattar
Jason Bowman
Vincent-Pierre BERGES
andrewcoh
Metin Celik
Ruo-Ping Dong
jmercado1985
Ervin T