Breaking Changes
- Arguments order changed in
Logger.report_line_plot()
,Logger.report_plotly()
andLogger.report_matplotlib_figure()
- please use keywords instead of positional arguments
Features
- Add OS environemt variable
CLEARML_DEFAULT_OUTPUT_URI
to override defaultoutput_uri
for automatic models upload #328 - Add
Task.force_requirements_env_freeze()
forcingpip freeze
instead of package analysis #304 - Improve pipeline support
- Add Task parameters override in
ClearmlJob
#214 - Add PipelineController node skip visualization and pre/post-execution callback
- Add Task parameters override in
- Allow enabling/disabling tensorboard auto-connect using
Task.init(auto_connect_frameworks={'tensorboard': True/False})
(default:True
) - Allow to continue running from previous execution iteration
- Support stopping instead of resetting in
Task.execute_remotely()
when server supports enqueueing stopped tasks - Add Model query interface using
Model.query_models()
- Support storing dict with
"."
in the keys using Task.connect_configuration - Add
force
argument toTask.mark_failed()
to allow setting Task status tofailed
regardless of Task state - Add
docker_args
anddocker_bash_setup_script
arguments toTask.create()
- Change Logger plot iteration argument default to 0 (or
None
) - Add task filtering in
Task.get_task()
using theallow_archived
andtask_filter
arguments - Add
Task.get_configuration_objects()
for reteriving configuration object as a blob of text in automation use-cases (instead of usingTask.connect_configuration()
) - Improved pipeline support
- Add pipeline step caching
- Add Pipeline controller caching
- Improve pipeline plot reporting
- Add
PipelineController
into main namespace (usefrom clearml import PipelineController
)
- Add jupyter notebook preview update time stamp
- Change default metric report flushing to every 5 seconds
- Improve
Task.artifacts
for safer Task multi node use cases - Support ClearML server API v2.13
- Add Python 3.9 to the support table
- Improve documentation and examples
- Improve documentation for clearml-data, clearml-task
- Add
multiple_tasks_single_process
example script for multiple Tasks created in same code - Add
using_artifacts_example
andexecute_remotely_example
example scripts - Update
requirements.txt
forKerasTuner
example
Bug Fixes
- Fix
\r
and\n
in debug samples title/series #323 - Fix Google API credentials without storage section raises exception on import #331
- Fix Matplotlib automagic export legend not showing series names #337
- Fix
Task.set_comment/name()
passingNone
should be equivalent to''
- Fix
Task.set_credentials()
uses incorrect config file when bothtrains.conf
andclearml.conf
are found - Fix
Task.set_credetnials()
should override settings from config file - Fix Keras h5 model storage
- Fix
Task.execute_remotely()
with sub process - Fix Task requirements should conform to strings
- Fix reporting after task closing
- Fix remove deprecation warning with matplotlib 3.1
- Fix dataset create single root plot
- Fix
Model.task
should return task ID - Fix code hangs when leaving process
- Fix flushing on CR timeout (default 10sec)
- Wait for subprocess when terminating
- Improve flush behaviour on
KeyboardInterrupt
exception - Wait for subprocesses to flush when terminating
- Fix
Task.set_credentials()
withstore_conf_file=True