- Moved plotting utils to the
cumulative
viz project
- Added
utils/plotting.py
module to facilitate the rendering of matplotlib visualizations
- Added Python and MLtraq version to experiment's metadata
- Added
utils/sysenv.py
to track environment stats (system, architecture, locale, time, ...) - Added
utils/reproducibility.py
, providing a temporary seed context - Added
utils/plotting.py
, providing a temporary plot context for a single axes - Removed
utils/plot.py
due to lack of usage (and unlikely usage in the future) - Improved
BaseOptions.get(...)
withprefer
andotherwise
parameters to make it more flexible
- Increased tests to
87%
coverage
- Using
Union[x,y]
instead ofx|y
for type definitions for increased Python 3.9 compatibility - Added a read-only mode for
BunchStore
, preventing an unnecessary write check - Added support for local files in
fetch
module
- Fixed a race condition in
BunchStore
- Added support for Python 3.9
- Verified tests on latest versions of dependencies
- Added missing
__init__.py
file
- Cleanup and update of dependencies
- Improved
BunchStore
class overloading also__delitem__
and__iter__
- Added SVG optimizer (smaller SVGs, no timestamps that cause git changes)
- Default value for
create_tables
ofDatabase
class constructor changed to False, allowing the usage of this class also for use cases beyond managing MLtraq DBs. - Added class
BunchStore
to handle aBunch
class transparently on memory and filesystem. - Added blog post on how to use
return_as=generator_unordered
and fail fast (blog/posts/011) - Added blog post on how to handle secrets with dynamic
Options
(blog/posts/012) - Added blog post on
BunchStore
to handle caches and more (blog/posts/013)
- Fixed some typos in tests and docs
- Added blog post on
BunchEvent
(blog/posts/010)
- Added
BunchEvent
type, a Bunch with function triggers for setters and getters
- Added utility steps
drop_fields
andnothing
- Fixed some quoting typos in the docs and code
- Added blog post on serializing a
Bunch
of things (blog/posts/009) - Cleanup visibility of class methods and attributes (
_
and__
prefix) - Added PNG logo files with transparency for talks etc.
- Added slides from Munich MLOps Community Meetup #7, linked in the benchmarks/speed page
- Added notebooks/11 to experiment with the performance of different data storage options
- Added blog post on upstreaming experimental results (blog/posts/008)
- Renamed
session.load
tosession.load_experiment
andsession.persist
tosession.persist_experiment
for consistency
- Dropped
include_hidden
option support for archives, not supported byglob.glob
in Python 3.10