Version 0.13
Trains is now ClearML
This documentation applies to the legacy Trains versions. For the latest documentation, see ClearML.
Version 0.13.3
Trains
Features and Bug Fixes
- Add a binding for
tensorboard.summarywriter.addscalars
- Add the
tensorboard_single_series_per_graph
method which supports separate plots for each TensorBoard scalar. - Add the
Task.set_base_docker
andTask.get_base_docker
methods for the base Docker image used by Trains Agent. -
Add support for the standard OS environment variables to obtain default credentials for:
- AWS:
AWS_ACCESS_KEY_ID
,AWS_SECRET_ACCESS_KEY
, andAWS_DEFAULT_REGION
. - Azure Storage:
AZURE_STORAGE_ACCOUNT
andAZURE_STORAGE_KEY
. - Google Cloud Storage:
GOOGLE_APPLICATION_CREDENTIALS
.
- AWS:
-
Add the
Task.get_parameters_as_dict
andTask.set_parameters_as_dict
methods supporting get / set of parameters from referenced Tasks (use theTask.get_task
to get a reference). - Make sure
Task.connect
always returns the connected instance passed to it. tensorflow_gpu
takes precedence overtensorflow
when Trains detects installed packages to record experiment dependencies.- Remove title and series naming restrictions (allow
$
and.
) when reporting metrics. - Fix incorrect printouts in initialization wizard and upgrade notifications
- Fix debug images URL for uploaded files with
%
in their name
Trains Agent
Features and Bug Fixes
- Allow providing queue names instead of queue IDs in daemon mode.
-
Docker mode improvements:
- Support running as a specific user inside a docker using the
TRAINS_AGENT_EXEC_USER
environment flag. - Pass the correct GPU limit when skipping gpus flag.
- Add the
--force-current-version
daemon command-line flag.
- Support running as a specific user inside a docker using the
-
Add K8s/trains glue service example
-
Add K8s support in daemon mode
- Running inside a K8s pod
- Mounting dockerized experiment folders to host
- Allow a specific network for the docker
-
Add default storage environment vars (for AWS, GS and Azure) to generated agent configuration
- Improve Unicode/UTF stdout handling
Version 0.13.2
Trains
Features and Bug Fixes
- Allow reporting a pre-uploaded image url in (Logger.report_image() using the (url parameter.
- Add support for Git repositories without a
.git
suffix, for example Azure Repos. - Improve conda support.
- Improve hyper-parameters (argparser integration.
- Fix (savefig() patching in matplotlib binding.
- Fix logs, events and Jupyter Notebook flushing on exit.
Version 0.13.1
Trains
Features and Bug Fixes
- Add support for
pyplot.savefig
andpylab.savefig
in matplotlib binding. - Add support for SageMaker.
- Improve configuration wizard.
- Try to make sure TensorBoard is available when using torch.
- Do not store keras model network design if it cannot be serialized (GitHub Issue #72).
- Fix matplotlib binding support.
Version 0.13.0
Trains
Features and Bug Fixes
- Add support for (trains-server v0.13.0.
- Add support for nested (non-main) tasks.
- Add warning when automatic argument parser binding cannot be turned off.
- Add
Task.upload_artifact
support for external URLs (pre-uploaded). - Add support for special characters in hyper-parameter keys (white-spaces,
.
and$
) (GitHub Issue #69). - Add support for PyTorch
.pt
model files. - Calculate data-audit artifact uniqueness by user-criteria (GitHub Issue #45).
- Use an environment variable for setting a default docker image (GitHub Issue #58).
- Improve
trains-init
configuration wizard. - Update examples for new joblib versions.
- Update jupyter example to TensorFlow 2.
- Fix task clone to copy only input artifacts.
- Fix matplotlib import binding when using
Agg
backend. - Fix
ProxyDictPreWrite
andProxyDictPostWrite
so they can be pickled correctly (GitHub Issue #72). - Fix requests issue in Python 2.7 that can cause a deadlock when importing netrc.
- Fix argparser binding sub-parser and type casting support (GitHub Issue #74).
- Fix argparser binding Python 2.7 unicode handling.
- Fix unsynced connected hyper parameters overridden during remote execution.
Trains Server
Features and Bug Fixes
- Add parallel coordinates hyper-parameter comparison, available under Compare Experiments -> HYPER PARAMETERS -> Parallel Coordinates (in the drop-down) (GitHub Issue #53).
- Add encoding of experiment table view settings in URL to allow sharing using browser URL copy/paste.
- Add loguru (ANSI color) support (GitHub Issue #29).
- Add support for special characters in hyper-parameter keys (white-spaces,
.
and$
) (GitHub Issue #69). - Add optional anonymous daily usage statistics (help us improve Trains Server):
- Disabled by default.
- Requires user opt-in.
- Single averages report per day.
- Reports average load metrics per day (CPU/memory).
- Reports average workload per day (amount and average duration of queues, agents and experiments).
- Improve experiment table filtering indication.
- Improve model view to allow navigating to its generating experiment.
- Fix experiment comparison to distinguish between experiments with the same name (GitHub Issue #52).
- Fix Web UI compare plots bug (GitHub Issue #55), (GitHub Issue #73).
Trains Agent
Features
- Add support for Docker pre-installed pytorch versions that do not exist on PyPI/PyTorch.org.
- Add AWS dynamic cluster management service.
- Add support for various event query endpoints in APIClient.
- Improve the configuration wizard.