- python path is now propagated to the tasks
- cli.py: Python path should be modified including when running the experiments
- python path and module in experiments
- pre-commit commitizen update
- when fabric is not there, use placeholders
- js vulnerabilities fix
- python 3.12 compatibility (but without ssh)
- no dependencies for loaded configurations
- fabric is now optional
- include jobs not in experiments
- include jobs not in experiments
- bug in setting workspace env
- andles better null process
- use a better way to load the XP file
- removed the launchers.yaml support
- better job interactions
- bug when no settings.yaml file
- set init tasks before job creation
- removed extra logging information
- wrong information reported
- wrong test for running/done tasks
- change the cli source dir
- bug in environment
- bugs in normal task exit
- change run-experiment parameters
- removed unused environments
- get back to sys.exit
- force exit with multiprocessing
- multiple launchers
- handles properly mp.spawn
- requirements can now include a disjunction
- environment can be defined in the settings
- slurm: Check if slurm process is alive before returning it
- really using the workspace directory
- uses job failure status
- HandledException to shorten error stack trace
- Deprecate YAML defined launchers
- added pre/post yaml options
- better error messages
- support python 3.12 pathlib
- typinutils: test for optional
- _collect_parameters is in Python 3.11
- types: accept generics as Param (but no real validation)
- ConfigurationBase: Documentation and new fields to describe an experiment
- process YAML file in the right order
- Configuration must be OmegaConf to be transformed
- Use XPMValue and XPMConfig to distinguish configurations and values
- set id as MISSING
- parent should be really optional
- decorator class method for experiment helpers
- representation of identifier as hex string
- load task from directory
- identifier is a method not a property
- run-experiment cli
- Saving/Loading from experiments
- access run mode from experiment
- Test for falsy documented
- utility function to format time for SLURM
- bad assert test
- launchers.py specification file
- Serialization for arbitrary data structures
- task identifier and reload
- task identifier and reload
- workspace settings
- init tasks
- undocumented produces more information
- undocumented in sphinx documentation
- Added decorator (NOOP) for config only methods
- Generic documentation checker (for use in automated tests)
- improved check documentation command
- List experiments
- documentation: Checks for undocumented configuration objects in a package
- configuration: Add access to pre-task and dependency copying
- more slurm fixes
- max duration QoS was ignored
- more fine-grained SLURM configuration
- blank line after job information
- slurm: Better SLURM launcher finder
- cleaned up the instance() mode
- cleaned up the instance() mode
- better instance()
- identifiers
- pre-task are properly handled
- exception thrown when adding pre-task to a sealed config
- pre task dependencies are taken into account
-
- subparam were removed (should be replaced by something more stable)
- serialiazed configurations were removed (too much trouble too)
- show dependencies when simulating
- removed config wrapper
- easier path LW task access
- Lightweight-pre-tasks: Lightweight pre-tasks allow code to be executed to modify objects
- bug in dependency tracking
- Expose the unwrap function
- Removes unnecessary server logs
- Fix submit hooks (and document them)
- serialized configurations
- requirement for fabric
- add gevent-websocket for supporting websockets
- Changed TaskOutput to ConfigWrapper
- submit hooks to allow e.g. changing the environment variables
- tags as immutable and hashable dicts
- corrected service status update for servers
- improved server
- When an experiment fails, display the path to stderr
- service proxying
- JS vulnerabilities fix
- Information message when locking experiment
- Improving slurm support
- Fix test bugs
- better handlign of services
- server: switched to flask and socketio for future evolutions of the server
- improvements for dry-run modes to show completed jobs
- more reliable identifier computation
- better identifier recomputation
- fix bugs with generate/dry-run modes
- allow using the old task identifier computation to fix params.json
- New identifiers will be different in all cases - use the deprecated command to recompute identifiers for old experiments
- For any task output which is different than the task itself, the identifier will change
- configuration: re-use computed sub-configuration identifiers
- server: fix some display bugs in the UI
- configuration: fixed more bugs with identifiers
- configuration: fixed bugs with identifiers
- configuration: serialize the task to recompute exactly the identifier
- removed jsonstreams dependency
- server: web services for experiment server
- wrong indent