Releases: bessagroup/f3dasm
Releases · bessagroup/f3dasm
v1.5.4
Bugfixes and minor improvements
- Change
__eq__
method of_Data
object topd.testing.assert_frame_equal
in order to omit differentfloat
types Domain.add_output
now has default argument ofto_disk=False
- Bugfix:
ExperimentSample.from_numpy
now gets the (optional) domain object when constructed ExperimentData.get_input_data
selects now only the input domain and removes the output space- Removing building JOSS paper workflow (paper has been published and draft is removed from repo)
- (Studies: fragile becomes supercompressible) update scripts config
What's Changed
- [JOSS] Paper by @RMeli in #277
- Pr/1.5.4 by @mpvanderschelling in #280
New Contributors
Full Changelog: v1.5.3...v1.5.4
v1.5.3
Resolved requests from reviewers JOSS:
- Fixes TOC header sphinx-gallery (#274)
paper.md
: Added DOI of Optuna reference in paper
What's Changed
- Sphinx Gallery problem with TOC headers by @mpvanderschelling in #275
- update version number to 1.5.3 by @mpvanderschelling in #276
Full Changelog: v1.5.2...v1.5.3
v1.5.2
Major changes
- Restricted
numpy<2.0.0
fromrequirements.txt
- Minor linting fixes
What's Changed
Full Changelog: v1.5.1...v1.5.2
v.1.5.1
Major changes
- Small bugfixes and typos
What's Changed
- Pr/1.5.1 by @mpvanderschelling in #267
Full Changelog: v.1.5.0...v1.5.1
v.1.5.0
Major changes
- Refactoring and removing of legacy code
- Change how hyperparameters are handled; they now need to be given as initialization key-word arguments
- Major rework on optimization functionality
- Rework on JOSS paper draft
What's Changed
- Pr/1.5 by @mpvanderschelling in #266
Full Changelog: v.1.4.8...v.1.5.0
v.1.4.8
Major Changes
- Dropped compatibility with Python 3.7
- Added Sphinx Gallery for documentation examples
- added
examples/
folder with code snippets explaining the functionality off3dasm
- added
Experimentdata.join
method - Change
_Store
class to public:StorageProtocol
- Added 2 benchmark studies: Fragile becomes supercompressible and Optimization benchmark
What's Changed
- Pull changes from 1.4.71 to branch 1.pr/4.8 by @mpvanderschelling in #264
- Pr/1.4.8 by @mpvanderschelling in #265
Full Changelog: v.1.4.71...v.1.4.8
v.1.4.71
Bugfix:
autosummary
documentation not build because of conflicting submodule name.
Full Changelog: v1.4.7...v.1.4.71
v.1.4.7
Features
- Added storage method for matplotlib figures
- Added GridSearch sampler*
- Added helper functions for datageneration with hydra configuration files
- Added overwrite function to
ExperimentData
- Added callback functionality for optimizers
- Allow addition of Domains
- Add differentiable tag to benchmark functions and optimizers
- Added
cluster_parallel
mode to `ExperimentData.evaluate()
* only working for categorical or discrete variables!
Bugfixes
What's Changed
- Fixes #248 by @mpvanderschelling in #249
- Mpvanderschelling/issue250 by @mpvanderschelling in #255
- Improving usability by implementing helper function to convert arbitrary functions to DataGenerators by @mpvanderschelling in #254
- Documentation is both hosted by GitHub Pages and ReadTheDocs by @mpvanderschelling in #256
- Implement new x0_selection strategy and add select_all_finished functionality by @mpvanderschelling in #260
- RETRY: refactoring optimization iterate and iterate_scipy by @mpvanderschelling in #262
- Pr/1.4.7 by @mpvanderschelling in #263
Full Changelog: v1.4.6...v1.4.7
v1.4.6
Major changes
- Added documentation on SLURM job scheduler. Added sphinx-tabs as dependency when building documentation.
- WIP: New back-end of
_Data
that relies on lists instead of pd.DataFrame. This has not yet replaced the pandas back-end but tests are passing with the new_newdata.py
module. - Removed unused method
Domain.from_yaml
. - Removed unused
_Data.fill_output
method - Output of benchmark functions is now a
float
instead ofnp.ndarray[float]
. Necessary to pass tests for new back-end, won't affect pandas back-end.
Minor changes
- Refactoring parts of the
ExperimentSample
code to be more robust to changes in the back-end - Moving factory methods to appropriate modules
- Removed
try_import
; import checks are moved to extension packages - Added
__version__.py
file - Updated docstrings to remove leading spaces
Bugfixes
- Correct equality operator of
Domain
to also compare foroutput_space
attribute - Closes #241
- Closes #191
- Closes #240
- Closes #243
What's Changed
- Newdata by @mpvanderschelling in #245
- Pr/1.4.6 by @mpvanderschelling in #246
Full Changelog: v1.4.5...v1.4.6
v1.4.5
This pull request will update the main
code to version 1.4.5
Major updates
Domain
and Parameter
classes
- All
Parameter
classes are now private. The suggested way to add parameters to a Domain will be with the publicadd_float
,add_int
,add_category
andadd_constant
methods. - Change the behaviour of the public
Domain.add
method to require the type and arguments of the (private) parameter to construct, instead of requiring the private Parameter object
ExperimentData
class
- Added a method that extracts all
ExperimentSample
objects or a selection ofExperimentSample
objects from anExperimentData
object - Removed the
.filename
attribute inExperimentData
: the name of the stored objects are hardcoded - Refactored the
.path
attribute to the.project_dir
attribute. When storing theExperimentData
object with the.store()
method, a subdirectory (/experiment_data/
) will be created and the input data, output data, domain and jobs will be stored there.
Optimizer
class
- Added functionality to select different strategies to determine the initial points (
$x_0$ ) before optimization.
_Data
class
- Decoupled the headers of the
_Data
class to a separate private_Columns
class. This makes it more flexible for future development on the generalization of different back-ends (currently pandas) for_Data
- Removed dependency of
Domain
to_Data
class
Storing output data
- Output that is stored to disk will no longer have the
path_
prefix before the parameter. This information is stored within the_Columns
object - The
ExperimentSample.output_data
will, if applicable, load the output parameters from disk automatically - The attribute
ExperimentSample.output_data_with_references
will give you the paths of the references - Added backwards compatibility for
ExperimentSample.output_data_loaded
(will redirect toExperimentSample.output_data
Documentation
- Made private methods of public objects visible in sphinx documentation API
Bessa Group Coding Guidelines
- Restored the original .flake8 file of the BRG coding guidelines; the maximum linewidth is now 79 instead of 119.
Miscellaneous
- Represent all private attributes and methods with an underscore (
_
).
Closed issues
Full Changelog: v1.4.4...v1.4.5