CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

• Added 3 tutorials and embended existing colab tutorial in it.
• Support artifacts for each optimizer based on evaluated trials.
• Add a generic structure for the filesystem writers.
• Add live_plots flag for run() API to print the generic and optimizer-specific plots

Fixed

• fix device mismatch in the tensors created in gp acqisition process
• fix epsnet_selector for MO branckets to filter out pending trials

[0.16.0] - 2026-03-25

Added

• Add changelog workflow
• Add Clean command line for pruning trials
• support logarithm base for parameters
• Support individual discretization factors per hyperparameter in grid search
• Documentation and Example on PriMO

Changed

• Improve Acquisition Function optimization in discrete and mixed search spaces
• Update ifbo
• Clean command
• Update usage patterns and examples in documentation
• Update link to the tutorial

Fixed

• Fixes to fidelity resource calculation
• Trial id generation
• Docs for stopping criteria
• Fix link to examples in readme

[0.15.0] - 2025-12-15

Added

• Add import_trials method to NePSRandomSearch and NePSRegularizedEvolution for importing external evaluations as trials
• Add load_config function and update utility imports for better configuration handling
• Enhance NEPS API and examples: add create_config function, improve parameter handling, and update example scripts for better usability
• Add Regularized Evolution optimizer and integrate into NePS algorithms
• Add conversion logic for classic to NEPS search space in API
• Add string representation methods for Fidelity, Categorical, Float, Integer, Operation, and Resampled classes
• Add convert_classic_to_neps_search_space function to convert classic SearchSpace to NePS PipelineSpace
• Add inc_ratio parameter to _neps_bracket_optimizer and NePSPriorBandSampler for configurable incumbent ratio
• Add repeated resolvable type and dict resolving
• Add Lazy component to stop categoricals from eagerly resolving provided choices
• Add resolving for tuples and lists
• Add warning log for missing priors when use_priors is set to True in neps_random_search function
• Add location parameter and timing parameters to warmstart_neps function for configuration and performance tracking
• Add error handling and logging for pipeline space resolution in warmstart_neps function
• Add warnings for repeated warmstarting and compatibility checks for warmstarting in warmstart_neps function
• Add Gaussian priors to NePS Integer and Floats
• Add warmstarting functionality to the run method and create example script
• Add optimizer compatibility check for Pipeline in run function
• Add neps_algorithms as direct sub-import of neps
• Add documentation for NePS Spaces framework and usage examples
• Add Pytorch Neural Network example
• Add comprehensive tests for NEPS space functionality and conversion
• Add NOS like space
• Add bracket optimizers
• Add centering

Changed

• Introduce NePS-Spaces framework with enhanced pipeline space and parameter management
• Enhance optimizer implementations: improve bracket calculations, logging, and sampling strategies
• Refactor and enhance fidelity handling across optimizers with improved sampling and configuration normalization
• Add LocalAndIncumbent optimizer and integrate NePS Regularized Evolution with new sampling strategies
• Improve NePS space compatibility checks and enhance PipelineSpace serialization and persistence
• Consolidate resource tracking and enhance metric handling in state management
• Refactor documentation and examples for improved clarity and consistency across NePS spaces usage
• Simplify and enhance string representation methods for better readability of configuration objects
• Update ifbo dependency version in pyproject.toml

Fixed

• Remove unused parameter 'rung' from sample_config method in NePSLocalPriorIncumbentSampler
• Remove mf check
• Refactor bracket creation logic to group configurations by starting rung for improved sampling
• Remove fidelity value check and improve Hyperband bracket allocation logic
• Fix: Standardize ignore_fidelity parameter to use "highest_fidelity" across optimizers and related functions
• Add xfail for tests requiring additional setup in test_sample_trial
• Add additional setup requirement for neps_local_and_incumbent optimizer tests
• Update load_config to require either config or config_path and improve error handling
• Update _mutate_inc method to use fidelity attributes for environment values
• Update kernel_size in Conv2d operations from [3, 3] to [3, 4]
• Update kernel_size in test_list_as_arg for Conv2d operation
• Standardize ignore_fidelity parameter to use "highest_fidelity" instead of "highest fidelity"
• Improve error handling for args and kwargs in _format_operation function
• Improve error messages for incompatible optimizers in PipelineSpace
• Improve error messages and enhance user result filtering in evaluation
• Rename hyperparameter optimization example file for consistency
• Update ifbo dependency version in pyproject.toml
• Fix linter
• Update import_trials to use pipeline_space parameter and improve logging
• Enhance configuration handling by updating return types and adding prior checks in PipelineSpace
• Prevent resampling of Fidelity objects in Resampled class and add Operation to config_creation example
• Change status parameters
• Update load_config return type and improve operation string representation
• Simplify string representation in Operation class
• Improve string representation for Categorical and Operation classes
• Update string representation test to reflect BasicSpace instead of PipelineSpace
• Fix report loading
• Update docstring for NePSRegularizedEvolution to clarify functionality and parameters
• Update best score calculation to use actual best from trajectory and improve config selection logic
• Normalize weights in PriorBandSampler to handle floating-point precision issues and increase np.isclose tolerance
• Update optimizer names in tests for consistency and clarity
• Normalize weights in PriorBandSampler to handle floating-point precision issues
• Improve path formatting in tests and enhance readability of assertions
• Update root directory paths in tests for consistency and clarity
• Enhance configuration string formatting for hyperparameters and improve test assertions
• Improve formatting in wrap_config_into_string for better readability
• Update confidence level in MutatateUsingCentersSampler to MEDIUM for improved sampling accuracy
• Docs & remove .trace.lock
• Update contributors

Removed

• Remove write_summary_to_disk
• Remove obsolete Jupyter notebook for prior tests and update test handling for core examples
• Remove old warmstarting functionality, unused example notebooks and scripts
• Remove redundant check for 'auto' in compatibility verification

0.14.0 - 2025-09-16

Added

• Add ask and tell Documentations
• Add ask and tell example
• Add primo to algorithms
• Add primo
• Add primo
• Add primo
• Added trajectory and best incumbent. Solved warning in plot
• Added documentation and logging of incumbent
• Add requires_prior test case
• Add noqa comments to suppress linting warnings in algorithms and bayesian optimization modules
• Add tests fail comment

Changed

• Bump version: 0.13.0 → 0.14.0
• Update Contributors
• Async Saving Evaluation Result
• Change Worker ID
• Make defualt worker-id shorter
• Make worker id chosen by user
• Txt files fidelity stopping crit
• Fix Async and HB brackets
• Numpy>=2
• Modify tests for MOMF opts
• Allow None confidence centers for MO priors
• Update primo
• Add PriMO
• Change tests for MO
• Update api with MO and fix runtime issues
• Numpy version update
• Example fixed
• Fixed tests
• Fixed trace error in case of multiple runs
• Logging messages changed
• Merged branch with new txt files
• Merge remote-tracking branch 'origin/master' into trace_csv_file
• Neps.plot fixed
• Rename overwrite_working_directory => overwrite_root_directory
• Fixed multi-fidelity stopping criteria
• Optimizer - fidelities to spend
• Mf example
• Introduction of fidelities_to_spend
• Renaming max_cost_total to cost_to_spend
• Renaming max_evalutaions_total to evaluations_to_spend
• Multi-objective algorithms
• Upgrade to Python 3.13
• Fidelity and prior compatibility checks in optimization algorithms
• Update Documentation
• Enhances fidelity handling in optimization algorithms
• Assign trial_report to trial in tell and tell_custom
• Add Multi-objective Bayesian Optimization using qLNEHVI
• Add Multiobjective algorithms: syne-tune apache license
• Add MO_Hyperband
• Add MOASHA - add MOASHA to test neps state
• Add MOASHA - add optimizer in algorithms and modify bracketoptimizer
• Add MOASHA - add epsnet MO promotion strategy
• Add MOASHA - multi-objective brackets for promotion using MO strategies
• Clip decoded value oveflow after torch.exp
• Update LICENSE
• Update LICENSE

Fixed

• Fix Multi-fidelity algorithms
• HyperBand bracket priority
• Skip tests for PriMO for now
• Fix more ruff-formatting
• Fix ruff-format in status.py
• Disable info_dict logging
• Disable info_dict logging
• Fix CONTRIBUTING.md headers
• 123-optimizers-multifidelity-algorithms-should-check-for-min_budget-0
• Allow partial priors
• Allow partial priors in priors test
• Partial priors in space
• Merge branch 'master' into neps-mo
• Trial_report not being updated in Ask-Tell
• Clarify prior parameter requirements and error handling in documentation
• Simplify error message formatting in random_search function
• Improve error messages for prior usage in optimization functions
• Set default value for ignore_fidelity in grid_search function
• Insert pibo in requires prior in tests
• Tensor decoded value overflow for parameters with log
• Tensor decoded values overflow for parameters with log with torch.float64

Removed

• Remove upper bounds clip warning
• Removed info_dict from logging
• Removed comment
• Remove cli
• Remove roadmap and fix contributing docs
• Remove extras from prior test

0.13.0 - 2025-04-11

Added

• Adding only incumbent tblogger and docs
• Add mypy error ignore comments for portalocker lock module attributes, making pre-commit usable again
• Add image, removed code from README
• Added image-segmentation example
• Added image-segmentation pipeline example
• Added landing page
• Added FSDP examples with Neps
• Added FSDP examples with Neps
• Add warning if no stopping criterion are set
• Added native Pytorch DDP example with neps
• Added example for DDP with PyTorch Lightning
• Added support for PyTorch Lightning in the DDP backend.
• Added BOHB tips
• Add more details and links
• Add equation support
• Add more content and image captions
• Add images and SH explanation
• Added ifBO, changed graph
• Added diagram
• Added optimizer pages outline
• Add python 3.12 support
• Add ifbo
• Add cli to declarative usage docs + add declarative usage to README
• Add run-pipeline to neps run cli argument
• Add template of 'run_args' to neps init + add pipeline_space to neps run arguments

Changed

• Bump version: 0.12.2 → 0.13.0
• NePS tblogger feature refactoring
• Updating Docs
• Fix exception handling with tblogger
• Handle TensorBoard tracking errors and disable feature
• Running end trial callbacks under lock
• NePS tblogger feature refactoring
• Enhance optimizer documentation and structure
• Real world examples
• Updated optimizer.md to reflect recent changes to how optimizers are defined, passed and configured. Removed mentions of yaml-based customization.
• Finished restructuring algorithm pages
• Catch up optimizers page with changes to neps.run
• Landing page for algorithm section
• Merge remote-tracking branch 'origin/master' into optimizers_documentation
• TOC in Getting Started
• Cut down MF and Prior explanation
• Expandable boxes
• Algorithms as chapter point
• Whats next section in getting started page
• Convert pngs to jpgs
• Merge remote-tracking branch 'origin/master' into optimizers_documentation
• Lower the number of first stage samples for BO strat
• Model based bracket optimizers
• Simplify Prior, Sampler and ConfigEncoder construction
• Initial multi-objective support
• Added support for PyTorch Lightning in the DDP backend.
• Added native Pytorch DDP example with neps
• Merge master into ddp-pytorch-lightning
• Enable uv caching for github actions workflows
• Use env var to share config with higher rank workers
• Hacky fix for multiple DDP launches
• Reformatted algorithm names
• Filled MF and MF+Prior pages
• Small fixes and ideas
• Ask and tell
• Bump ytanikin/PRConventionalCommits from 1.2.0 to 1.3.0
• A lot
• Enable uv cache
• Renaming several functions
• Changed equation optics
• BO page
• Prior explanation
• Rename default to prior
• Restructuring optimizers page, inserting low-level-MF and prior explanations
• Newline in optimizers.md
• More detail and fixes for optimizer pages
• Switch from poetry to uv
• Enforcing conventional commit naming in PR titles
• Improve examples
• Update roadmap
• Update citations
• Improve conciseness of README
• Fix structuring of CONTRIBUTING docs
• Fixed priorband_template.py example
• Fixed pytest with new fidelity parameter checks
• Fixed priorband_template.py example
• New commands for CLI: sample-config, report-config, init, results ; Refactoring CLI
• Update roadmap
• Rename neps.XParameter to neps.X
• Rename neps.XParameter to neps.X
• Remove as many methods from SearchSpace as possible
• Use vulture to remove dead code
• Use ask() instead of two-stage load_optimization_state() and get_config_and_ids()
• BO and ifBO
• New cli commands help, info-config, errors, status, results + Documentation
• Refactor Declarative Usage (CLI) + Documentation
• Enable pipeline_space keyword in seperate yaml
• Change run-args to optional argument for neps run
• Provide a log message indicating summary data has been generated
• Improve message on how to deal with an error from a worker
• Modularize file state
• Introduce CLI
• Derive type(int, float, str) from str input of searcher-kwargs cli
• Introduce cli neps (init, run)
• Introduce Settings Class and Default() Arguments

Fixed

• Add mypy error ignore comments for portalocker lock module attributes
• Prevent prior definition for fidelity parameters
• Prevent prior value assignment for fidelity parameters in Float and Integer classes
• Spelling mistakes
• Making config ordering deterministic
• Explicit path print on successful 3D plotting
• Explicit path print on successful 3D plotting'
• Example and post-run plotting
• Ifbo plotting path and keys
• Example and post-run plotting
• Allowing None as constant HP in yamls
• Removing comment and unused import
• Making config ordering deterministic
• Allowing None as constant HP in yamls + minor fix to tblogger
• Use initial passed in value
• Fixup ruff pre-commit
• Categorical domain is is_categorical=True
• Populate prior
• Fixed bracket display
• Transformation of categoricals and handling of unusual fidelity bounds
• Fixed math display
• Fix headings
• Fix typo
• Filelock issues
• Fix typo in README
• Fix indendation in roadmap
• Debug env vars better
• Timeout set explicitly to 120 seconds, debug on env vars
• Fix np converting lists to np types
• Deprecation warning over concat with empty dataframe
• Use absolute path for previous config location

Removed

• Remove unused files
• Remove centers and priors
• Remove unused deps
• Removes a lot of methods from SearchSpace

0.12.2 - 2024-07-09

Added

• Add example for running neps on slurm scripts

Changed

• 0.12.2
• Update pyproject.toml
• Refactor declarative usage
• Merge of master
• Improve rendering of README and docs landing page
• Improve README.md and update docs landing page

0.12.1 - 2024-07-03

Added

• Add tests for checking optimizer loading via run_args
• Add python -m neps_examples utility to print examples
• Add torchvision to non-dev dependencies
• Add tests + fix docs
• Add searcher_args to searcher_info for custom class optimizer loaded via yaml
• Add providing arguments for loaded class BaseOptimizer via yaml + fix pre-commit related errors
• Add docstring rm solved ToDos
• Add mixed usage functionality for run_args and neps func arguments

Changed

• Bump version from v0.12.0 to v0.12.1
• Allow use of neps.Float instead of neps.FloatParameter
• Only log searcher as info level, log searcher_alg to debug
• Set post_run_summary to True always
• Fix incumbent tblogger bug
• Minor code fix + comment
• Minor change in test
• Minor change in test
• Update to master
• Readd sleep to hyperparameters example
• Maintain roadmap
• Move examples lists from test to neps_examples
• Improve readme, fix broken link, make slimmer
• Maintain roadmap
• Improve README writing, fix links, remove dead weight
• Minor bug fix
• Bug fix
• Have PiBO as an explicit searcher
• Change path of yaml test directories to tests_tmpdir
• Change path of yaml test directories to tests_tmpdir
• Change boolean values in yamls from [True, False] to [true, false]
• Update declarative example
• Code clean up + add notes to docs
• Clean up code and docsctrings
• Adapt SearcherConfigs to the new dict design of optimizers
• Update docs
• Update rm searcher_kwargs key from yaml for user
• Update rm searcher_kwargs key from yaml for user
• Change algorithm to strategy
• Change searcher key algorithm to strategy + rm searcher_kwargs argument
• Define dict in run_args yaml
• Update docs to the new design
• Merge master + fix loading neps-searchers from yaml
• Rm neps argument searcher_path
• Change design for searcher config
• Simplify code
• Change design for pre-load-hooks
• Align the tests for declarative usage to the ones that are in the docs + try fix test
• Update pyproject.toml for PyTorch versions
• Fix AttributeError: 'int' object has no attribute 'nodes'
• Cast Path to str
• Update pyproject.toml
• Update pyproject.toml
• Update pyproject.toml for PyTorch versions
• Update precision of best loss in logs
• Enable mixed usage for run_args and neps func arguments + new design for pre-load-hooks + new design for searcher config (yaml)
• Integrate pipeline space into run args yaml + new design for defining constant parameter
• Minor optimizations to examples using graph parameters
• Switch to clone() for search space.
• Declarative usage examples + simplified yaml usage
• Partial update of state
• Just two seperate classes
• Partial update of state
• Use binarized formats for de/serialization of rng
• Remove sleep from examples
• No arbitrary object loading torch.load
• Use binarized formats for de/serialization
• Directly link to API in docs, remove repetitions
• Move neps.run() to be the first recommended page
• Use function API instead of duplicating docs
• Directly link to API docs in search_space docs

Fixed

• Fix yaml dumping and default args selection for optimizer loaded via run_args
• Fix double logging and move post_eval_hook to runtime
• Save seed state by default
• Save seed state by default
• Fix path reference
• Fix test
• Fix pre-commit
• Fix pre-commit error
• Fix pipeline_space example
• Fix errors in tests
• Explicitly load previous results
• Allow for a worker not to have evaluated
• Boolean check for cost
• Lock by default for
• Ensure new state is saved to path
• Use 3.8 compatibly type aliases
• Fix test relying on boolean positionals

Removed

• Remove arg from examples
• Remove unnecessary torchvision requirement
• Remove neps argument 'searcher_path' and yaml argument searcher_kwargs + New loading design for optimizer

0.12.0 - 2024-05-02

Added

• Add ruff as dev dep
• Add ruff config
• Add valid type check for choices in categorical parameter
• Add logging info for parameter that uses e notation to inform them that the parameter gets deduced as float
• Add global variable to handle path location
• Add comment to check_arg_defaults
• Add type hints for arguments and returnn for all run_args functions
• Add cli neps(init, run) + modify custom optimizer usage, providing just a class
• Add tests to check functionality within neps.run() when using run_args.yaml to provide arguments + comment code
• Adding sleep time to inaccessible locks for summary_csv and sampler_info

Changed

• Version 0.12.0
• Big documentation cleanup
• Use relative links instead of https ones
• Big documentation cleanup
• Re-enable pre-commit
• Limit to just src dir
• Add sections regarding linting
• Update Contributing
• Fix up some files, ignore others
• Remove refernces to black
• Remove references to pylint
• Cleanup common.py
• Cleanup pyproject.toml
• Update pre-commit yaml
• Add action to run pre-commit
• Some progress towards neps.utils
• Fix up linting errors with status dir
• Document neps.runtime
• Ruff pass on types.py
• Ignore all files
• Improve validation and ==
• Improved error feedback for wrong path
• Change search_space: to pipeline_space: for tests
• Update to master
• Log boolean check
• Update eq function for parameters to use it to cover test cases
• First implementation of using yaml from run_args=
• Switch to relative import for module utils
• Implement strict argument usage check; raise errors as needed
• Minor fix
• Reorganizing doc structure for running NePS
• Update neps_run.md
• Update neps_run.md
• Doc fix for run_arg usage
• Rm cli command
• Delete cli
• Enhanced usability by supporting module paths with and without '.py' extension + general improved error feedback
• Rework of the yaml_usage examples + adding readme for guidance
• Documentation for run_args
• Rework of some of the code + comment and doc string changes
• Increase code maintenance + enabled strict usage of run_args, overwritting all other arguments that are provided via neps.run(...) + Enable loading the pipeline_space provided as a dictionary from YAML and the loading of the BaseOptimizer as searcher from YAML
• Implementation of run_args, loading and checking settings from yaml + tests
• Integrate metahyper as neps.runtime
• Type fixes
• Use Dict in TypeAlias for <=3.8 compat
• Use Union for <=3.9 compatibility
• Use poll everywhere
• Integrate metahyper closely
• Merge pull request #72 from automl/ci-concurreny-groups
• Run tests on any push/PR to master
• Add concurrency group to save resources on github action tests
• Update author list

Fixed

• Fix import
• No need to pass logger anymore
• Fix ruff versioning
• Fixup neps.plot
• Fix up neps.init
• Fix link in docs
• Fix issue #35 + boolean checks for log and is_fidelity
• Fix issue #35
• Fix wrong key name for example search space yaml
• Fix type hints for 3.8
• Fix type hints for 3.8
• Fix test + add run_args: constraint structure for yaml file
• Fix python environment call in windows for test cases
• Fix module import BaseOptimizer
• Fix provided arguments check, special case searcher-pipeline_space
• Fix provided arguments check
• Fix provided arguments check
• Broken import
• Fix example

Removed

• Removed unfinished declerative docs to another branch + added missing usage information for providing pipeline_space to neps.run()

0.11.1 - 2024-02-19

Added

• Adding first iteration to the run_pipeline documentation
• Add Examples to Readme
• Added citation + analyse.md iteration
• Add comment in code example
• Add comment in code example

Changed

• Bump version from v0.11.0 to v0.11.1
• Updating run_pipeline docs
• Updating docs for citations
• Doc link fixes + some minor text changes
• Merge pull request #56 from automl/configspace-patch-1
• Update pyproject.toml
• Documentation iteration + minor tblogger example change
• Merge pull request #50 from automl/next_doc_iteration
• Adjusting summary_csv with task_id and dev_stage_id args
• Big Documentation Update + align naming of configspace for yaml usage
• Closes #52: summary_csv compatible with development_stage_id
• Merge pull request #34 from automl/readme_changes_danrgll
• Updating parallelization documentation and README links
• Update branch to current master
• Fixing merge conflicts with master
• Enhancing README: Adding paper links as insights to key features
• Landing page README iteration + installation guide
• Update README.md
• Iterating analyse.md; format + doc changes
• Update README.md with additional key features and improved structure
• Update README.md
• Update citations.md
• Update README.md
• Tblogger doc iteration + adding images
• Merge remote-tracking branch 'origin/master' into readme_changes_danrgll
• Tblogger doc iteration + light/dark mode doc switch
• Iterating through examples README + tblogger doc
• Fixing conflicts from main and README iteration
• Change format
• Change format
• More compact format
• Change order of documentation section
• Adapt Indentation
• Changes for code usage example
• Changes for code usage example
• Merge pull request #48 from automl/dependabot/github_actions/actions/setup-python-5
• Bump actions/setup-python from 4 to 5
• Merge pull request #46 from automl/yaml_config_space
• Update to current master
• Changing root README.md Citations link to docs

Fixed

• Fix issue #44

0.11.0 - 2023-12-08

Added

• Add citations to the documentation
• Adding edits from #34 from @danrgll and @TarekAbouChakra
• Add type specification for arguments + add more detailed DocStrings for paramter validation functions
• Add examples + test adaptation to new functionalities + outsorcing utils for yaml search space in own file
• Add usage of e and 10^ for exponent format as input for yaml file + new tests for optional type argument
• Add documentation for config_space
• Adding callable function for post-run csv analysis
• Add tests, adapt pipeline_space_from_yaml(), fix some errors raised by pre-commit in search space
• Add yaml to dict function and make it callable in api run
• Adding .optimizer_info.yaml checks for catching user errors
• Adding post-run summary files + testing
• Added whoami function and a test case for searcher as BaseOptimizer
• Added return values

Changed

• Bump version from v0.10.0 to v0.11.0
• Fixing roadmap version
• Updating roadmap pre-release
• Merge pull request #47 from automl/Neeratyoy-patch-1
• Update README.md
• Enable usage of Path object for yaml_file config_space
• Merge master into branch
• Fixing flattening merge issues between src/ and neps/
• Merge pull request #45 from automl/repo-flattening
• Merge remote-tracking branch 'origin/master' into repo-flattening
• Minor file import changes for clarity and consistency
• Updating pyproject.toml with new repo directories location
• Repo flattening + adjustments
• Merge pull request #36 from automl/typing-export-types
• Export py.typed for neps + metahyper
• Resolves #26; resolving deprecation warning
• Made code more readable for validate parameter inputs + add tests + comment functions
• Changes in yaml_search_space examples
• Resolve merge conflicts
• Change tests to marker neps-api and resolve merge conflicts
• Merge pull request #33 from automl/api-checks
• Better testing + api checks before starting the search
• Fixing merge conflicts with master
• API refactoring + API tests and checks
• Clean up tests
• Summary_csv test fix
• Iterate over summary_csv, fix warning, and convert strings to lowercase
• Adapt tests and documentation for yaml_search_space
• Adapt tests and documentation for yaml_search_space
• Merge pull request #17 from automl/dependabot/github_actions/actions/checkout-4
• Bump actions/checkout from 3 to 4
• Minor tblogger image logging bug fix
• Fix merge error and move MissingDependencyError into metahyper
• Fixes #24 for the next release
• Merge pull request #38 from automl/deps-remove-pandas-upperbound
• Remove overly restrictive dependancies
• Enable to run tests for yaml
• Merge pull request #22 from automl/csv_summary
• Refactoring unit test + stochastic failure debug
• Minor update to neps/status/status.py
• Minor updates to neps/api.py
• Resolving merge conflicts with master
• Minor updates
• Fixing multi-thread .optimizer_info.yaml using decision_locker
• Tblogger example iteration
• Tblogger example iteration
• Merge pull request #21 from automl/tabular-space-support
• Merge conflict resolution
• Removing neps api test from GitHub actions
• Minor cleanup
• Allowing full table sampling in acquisition
• Merging updates
• Budget id >= 0
• Minor clean up, and suggested changes
• Increase pytest timeout to 15mins
• Minor logic updates
• Removing stray iimport
• Making PFN conditioned on local availability
• Merge resolution + removing timing code
• Merging with master
• Tabular support + DyHPO - joint commit with @karibbov
• Merge remote-tracking branch 'refs/remotes/origin/master'
• Merge remote
• Set normalize_y to False
• Utility functions update + example templates iteration
• Reducing tblogger example runtime
• Update releasing.md

Fixed

• Fix test cases
• Fix format of Pipeline Space Documentation for mkdocs
• Fix naming of parameters in test
• Fix issue regarding tests and search space from yaml file

Removed

• Remove 10^ notation + introduce key checking for parameters + enable e notation usage for all parameters and arguments
• Remove extra pop

0.10.0 - 2023-11-06

Added

• Adding warning filter for tensorboard in pytest
• Adding Py3.11 test on Mac
• Add support for python 3.11
• Add cloning to docs of contributor install
• Adding utility functions for checkpoints and the lightning template
• Adding priorband template
• Adding template for basic neps usage
• Add input-output scaling to DeepGP, changes from Heri
• Add functionality to MFObservedData
• Added ignore_missing_defaults option to sample_default_configuration()

Changed

• Bump version from v0.9.0 to v0.10.0
• Change license from MIT to Apache-2.0
• Increasing GitHub pytest timeout from 10 to 15 minutes
• Minor typo fix
• Removing argparse from tblogger example
• Shortening tblogger example runtime
• Removing argument parsing in Tensorboard example
• Template update + torchvision in deps
• Including new examples for testing
• Adjusting tblogger tutorial
• Adjustments for Tensorboard and saving configuration data from Metahyper
• Fixing PyTorch2.0 issue with pyproject
• Updating dependency, torch version
• Exclude macos python 3.11 test for now
• Update author order in CITATION
• Update author order
• Reformatting optimizer_info yaml with fewer restrictions on Metahyper
• Merge pull request #18 from automl/yaml-api
• Stricter conditions on the optimizer
• Iterating yaml api PR
• Neps with pytorch lightning example
• Making new markers for the yaml_api tests
• Handle Failure mode of the DeepGP: use untrained model
• Tabular SearchSpace for MFEIBO
• Merge remote-tracking branch 'refs/remotes/origin/master'
• DeepGP checkpoint and refine
• Minibatch DeepGP training
• Yaml test cases and minor fixes
• Api yaml new feature
• Merge remote-tracking branch 'origin/master' into tensorboard-neps
• Minor tblogger example changes
• Minor tblogger example changes
• Record learning curves explicitly
• Expose DeepGP parameters, make early stopping optional, disable default normalize_y, add TODOs
• Avoid duplicates when sampling new configs
• Merge pull request #14 from automl/tensorboard-neps
• Tblogger cleanup
• Merge remote-tracking branch 'origin/master' into tensorboard-neps
• Merge remote-tracking branch 'origin/master' into tensorboard-neps
• Tblogger changes in class and example
• Merge remote-tracking branch 'origin/master' into tensorboard-neps
• Fixing issues in tblogger and example
• Tensorboard example and class minor changes
• Tensorboard functionality
• future.annotations
• Deep gp with MFEIBO
• Merge remote-tracking branch 'refs/remotes/origin/master'
• Update init design for MF-EI-BO
• Bug fix to MFI + budget-based initial design
• Initial deep gaussian process
• Merge fix
• Simplify MFEIBO get_config_and_ids, anf mf_ei.py
• Fixing issues with MFEIBO
• Update init.py
• Dyhpo Acquisition Function MF-EI implementation
• Minor cleanup
• Cleaning changes to MF-EI
• Synthetic Regression Tests added
• Initial MF-EI changes
• Regularized Evolution default assisted = False
• Resolve python version version based issues
• DyHPOBase draft version, with a random acq
• Bug fix in set_hyperparameters_from_dict() for ConstantParameter
• Bug fix in set_hyperparameters_from_dict()
• Modified set_hyperparameters_from_dict() to allow setting it from another search space as well
• Changed data_loading utility functions to be more flexible and use json instead of yaml
• Tensorboard Evaluation File
• Removing deprecated import
• Removing deprecated import
• Removing deprecated files
• Bug fix: log transform with log as False

Fixed

• Fixing optimizer yaml file
• Fixing tblogger example

Removed

• Remove support for python 3.7 (end of life)
• Remove metahyper doc stubs
• Remove the budget_id column from observed data

0.8.4 - 2023-05-03

Changed

• Bump version from v0.8.3 to v0.8.4
• Updating PriorBand defaults
• Docstring update
• Simplifying PB decay
• Un-exclude macos tests
• Adjust numpy version
• Un-exclude some windows / macos tests
• Adjust torch version for macOS, add OS classifiers
• Bound torch version to try fix macOS issue
• Allowing prior-based MF algos to sample default at max budget
• Fix poetry version for github actions
• Bump poetry version for github actions
• Merge pull request #9 from automl/dependabot/github_actions/abatilo/actions-poetry-2.3.0
• Bump abatilo/actions-poetry from 2.1.6 to 2.3.0
• Run pre-commit on all files

Removed

• Remove continuation of crashed configs from metahyper
• Remove torch version constraint

0.8.3 - 2023-03-21

Added

• Add -u flag to publish documentation

Changed

• Bump version from v0.8.2 to v0.8.3
• Bug fix in SH
• Fixing failing test case for mutation on IntParam
• Replace prints with logger

0.8.2 - 2023-03-18

Added

• Add eddie to CITATION.cff
• Add Eddie as author
• Adding mutation as incumbent sampler
• Add utils and prototype optimizer for meta_neps
• Adding PriorBand+BO
• Add result hallucination for pending

Changed

• Bump version from v0.8.1 to v0.8.2
• Adjust sleep in example
• Updating PB decay method
• Exposing prior confidence param in BO
• Typo fix
• Bug fix: IntParam mutable object not updating
• Ignore *.DS_Store in .gitignore
• Bug fix in set_hyperparameters_from_dict() preventing None as default with has_prior=True
• Updating mutation function flow per HP + local mutation func. in PB utils
• Manual merge of changes from hnas branch
• Fixing MF-BO joint 2-step AF
• Bug fix: PB BO joint
• Revamping MF BO
• Minor update to PB
• Fixing log transform trigger for float HPs
• Major update to PriorBand family + Bug fixes
• Async PriorBand param update
• Setting default log prior weights to False
• MFBO data subset bug fix + PB decay_t
• PriorBand design changes for ablation
• Fixing baseline logic
• Updating HB+inc baseline
• Minor update to dynamic inc sampling
• Changes to mobster
• Minor logic update in inc activation (PriorBand)
• Code refining and cleaning (mf-priors)
• Minor update to SH sampling default
• Fixing import from merge conflict
• Merging to master a highly diverged commit

Fixed

• Fix missing pyyaml dependency
• Fix example links

0.8.1 - 2023-01-08

Added

• Add small fixes for repetitive architecture spaces

Changed

• Bump version from 0.8.0 to 0.8.1

0.8.0 - 2022-12-29

Changed

• Bump version from 0.7.0 to 0.8.0
• Clean architecture and hps example
• Clean architecture example
• Clean up NAS api further
• Ignore grakel warning
• Rework NAS api
• Update roadmap

Fixed

• Fix init structure

0.7.0 - 2022-12-18

Added

• Add mobster
• Added delete_previous_defaults anf delete_previous_values options to set_hyperparameters_from_dict()
• Added function set_default_confidence_score to float, integer and categorical
• Add get_summary_dict
• Add budget stop criterion to multi_fidelity
• Add fixing of autograd warning to roadmap
• Add rename proposal to roadmap
• Adding async HB baseline
• Adding async PB exp
• Adding PB to async SH/HB
• Add incumbent plotting
• Adding HB + inc sampling
• Adding sampling from hypersphere
• Add convenience examples and fix example links
• Add parallelization example and remove pipeline_dir from basic_usage
• Add metahyper to this repository directly
• Adding HB,ASHA,ASHA-HB based PriorBand versions
• Adding NePS HB with PriorBand ensembling
• Adding Ensemble Policy and custom HB default

Changed

• Bump version from v0.6.1 to v0.7.0
• Merge pull request #4 from automl/remove_fnctl_dependancy
• Update roadmap
• Exclude macos 3.7 and 3.8 tests
• Bug fix in set_hyperparameters_from_dict() for categorical parameters
• Improve plotting sciprts
• Update roadmap
• Deactivate macOS 3.8 and 3.7 tests
• Do not create graph for nlml backward pass
• Exclude windows with python 3.10 as a gh-actions test
• Fix mark selection for pytest
• Trigger on update to workflows too
• Enable metahyper tests
• Add macos and windows to tests
• Update actions
• Add dependabot for github actions
• Updating PriorBand with inc sampling type and style + misc changes
• Minor cleanup
• Rename budget to max_cost_total, add to example
• Updating asyncPriorBand
• Fixes for parallel/async MF algos
• Update roadmap
• Improve error message for loss value on error
• Minor param update to async PB
• Fixing metahyper import
• Do not print summary when plotting
• Save previous_config_id to metadata.yaml
• Update old docstring
• Clean up metahyper, add analysis test
• Improve README text
• Minor change to plot structure
• Improve README
• Improve analyse example
• Improve documentation on analysis
• Improve examples readme
• Rework examples with respect to analysis
• Parallelization example as .md
• Improve plotting (now through neps.plot)
• Move AttrDict to common utils
• Revert example
• PriorBand exps and ablations
• Removing deprecated files
• Making new additions restructure compatible
• Make root_directory an arg not a kwarg
• Constrain the torch version due to installation issues
• Refactor examples
• Use pipeline_directory instead of working_directory in examples
• Improve speed of poetry install a bit
• Update to poetry 1.2 syntax
• Updating logs and minor formats
• Update roadmap

Fixed

• Fix metahyper test
• Fix roadmap formating
• os.setsid not available on windows
• Try unlock instead of lock for windows segfaults
• Add test for filelocking
• Change to cross-plat portalocker
• Fix root dir name in example
• Fix examples readme
• Fix python compatibility issues for plotting
• Fix poetry version in github workflow
• Fix verison in citation.cff

Removed

• Remove dill dependency
• Remove unnecessary metahyper files

0.6.1 - 2022-10-25

Added

• Add python versions to test matrix
• Support python3.8/9/10

Changed

• Bump version from v0.6.0 to v0.6.1
• Setting multi-fidelity default to model-free HB
• Merge fix
• Minor fixes to HB and asyncHB
• Making sync. SH and HB parallel
• Initial SH update
• Adjusted to 0.6.0
• Merge https://github.com/automl/neps into HEAD

Fixed

• Fix gh actions semantics for python 3.10
• Fix install for 3.7 and 3.10
• Fix version constraint

0.6.0 - 2022-10-22

Added

• Add ability to ignore errors
• Added ranomized PriorBand Policy
• Added random sparse promotion
• Added default sampling and made peaky prior a non-default setting
• Adding v6 with v5 variations
• Added ablation study functionality - regular sampling and promotion
• Added args for allowing ablation studies in raceband
• Add disable_prior flag to BO optimizer
• Adding variants to v5
• Added raceband - temporary name for possible Hyperband competitor
• Adding v5 to mf-priors
• Adding confidence and random sample HPs in mf-prior opts
• Added EI, toyed with bandwidths to emphasize prior
• Added Expected Improvement-like weighting to the TPE
• Adding random_interleave_prob to multi_fidelity
• Adding new optimizer to decay prior to uniform
• Adding new ASHA-AsyncHB opt
• Adding new opt with small update to Async HB
• Added adaptive correlation between fidelities for MF-TPE - only lacking
• Adding first few trial versions of mf-prior algos
• Added ability to remove samples that have been queried at multiple
• Added async support - registering pending configurations as bad in the
• Add Nils
• Adding func to track incumbent in SH
• Add mike to documentation for versioning support
• Add parallelization example to roadmap
• Add alternatives documenation and improve toc of docs
• Add Note section to readme
• Add carl to CITATION.cff
• Added KDE & Vanilla TPE along with example. KDE comes with additional
• Add constant to pipeline_space_from_confgispace
• Add comments explaining core vs all tests
• Adding Asynch HB with and w/o priors

Changed

• Bump version from v0.5.1 to v0.6.0
• Jitter=1e-6 for gp.py
• Modified defaults and adjusted sampling according to discussion
• Making BO support sampling of default first
• Appending default as first sample param in MF algos
• Changed so that constants have default values, and so that
• Changed to enable vanilla Raceband, no priors
• Fixed minor budgeting bug
• Changed sampling procedure tolerance
• Changed Raceband to accept max budget and utilize sharper priors
• Fixed constant support and categorical distance bug
• Omit ConstantParameter in GP-opt
• Updating v4 with sync SH/HB changes
• Simplified RaceBand according to suggestions
• Fixed sampling tolerance in neighborhood sampling
• Fixed integer parameter bug in sobol -> config conversion
• Updating synchronous MF algos
• Updating categorical choice confidences
• Introduced learning between brackets
• Merging master
• Option for ultra confidence in mf-prior algos
• Changed some HPs of the KDE, and re-added some plotting utilities (that
• Minor readme edit
• Minor comment
• Updating async HB sampling of brackets/rungs
• Fixing/tuning v3_2
• Merge https://github.com/automl/neps into HEAD
• New opt with prior decay to RS
• Updates to dense graph allowing te be trained. small fixes in the API for graphs
• Now supporting Constant values (and ordinals, but those are considered
• For Prior-based MF algos defaulting to NePS confidence
• Reverted the prior strengths in SH to adhere to the convention
• Making SH priors peaky
• Regression tests
• Regression test's metrics changed
• Ensured to properly differentiate between rung and fidelity
• Fixing configspace mapping and RS parameterization
• Bug fix in compute_prior() for a ConstantParameter
• Categorical, Integer and log variables now all work. V1, should be ready
• Async soft promotions now work as intended. First version of algo done,
• Changed the mf_tpe example to not include categoricals (not quite
• Multi-fidelity promotions incorporated - still need to remove
• Ensured multifidelity splitting is done correctly in TPE, and modelled
• KDE now weights each point according to fidelity - still need to sort
• Option for prior confidence in SH/HB
• Syncing with master
• Started incorporating multifidelity funcitonality
• Update roadmap
• Document mike
• Improve documentation on testing and checking protocols
• Refine roadmap
• Refine roadmap
• Unbold documentation / examples
• Improve readme text organization
• Refactor search spaces sub directories
• Renamed KDE, fixed weighting, added some defaults
• Adjusted some parameters of the KDE, allowed for fitting of fixed
• KDE now runnable
• KDE now runs - fixed imports
• Accidentally changed in old optimizers.init.py file, reverted now
• Restructure neps_examples and simplifiy test_examples.py

Fixed

• Fix docs
• Fix docs
• Fix links in docs
• Fix links in README
• Fixed cost-cooling when estimated costs are zero
• Fix typo in README
• Fix links in documenation
• Fix pyproject and add carl

Removed

• Removed print statements
• Removed print statements
• Remove deprecated documentation part
• Remove torch installation utils
• Remove old plotting utils

0.5.1 - 2022-08-29

Added

• Add changing of CITATION.cff to docs on version releases

Changed

• Bump version from v0.5.0 to v0.5.1
• Initial HB commit
• Testing submodule + minor code cleanup
• Fixing failing example
• Fix failing test
• Bump version to 0.5.0 also in CITATION.cff
• Restructuring multi-fidelity - adding SH, ASHA with and w/o priors
• Restructuring multi-fidelity - adding SH, ASHA with and w/o priors
• Improve poetry documentation

0.5.0 - 2022-08-26

Fixed

• Fix version from 0.4.10 to 0.5.0

0.4.10 - 2022-08-26

Added

• Add roadmap to documentation
• Add function set_defaults_to_current_values() to search_space.py
• Add Samir
• Add collection of todos
• Add documentation link to README
• Add dev and task id to run() and pass them to metahyper.run()

Changed

• Bump version from 0.4.9 to 0.4.10
• Move jahs_bench to dev dependency
• Deprecate working_directory in favor of root_directory for neps.run
• Move roadmap into contributing folder
• Improve readme and documentation
• Improve shields
• Change license
• Update link to documentation in README
• Improve contributing documentation and add releasing howto
• Regression tests
• Bug fix in set_defaults_to_current_values()
• Refactor CategoricalParameter outside numerical
• Working_directory -> root_directory
• Update documentation structure
• Updated poetry installation instructions
• Making error the default behavior for optimizer

Fixed

• Fix configspace + pipelinespace suppor
• Fix docstrings
• Fix pyproject license
• Fix link to example

0.4.9 - 2022-05-31

Added

• Add install from pypi
• Add authors and info
• Add random interleave prob in MF BO
• Add unparsing
• Add back handling of a new graph represenation (hierarchical)
• Add prior for multiple repetitive grammars

Changed

• Ignore dist
• Rename project for pypi
• Small updates to take last combo op
• Create CITATION.cff
• Letting MF BO switch to BO after eta evals on max fidelity
• Implement loss_value_on_error and cost_value_on_error base_optimizer.py and all optimizer classes. Values can be passed to api.run() with **searcher_kwargs. status() and _post_evaluation_hook() always use default value inf.
• Introduce parameters loss_value_on_error and cost_value_on_error in result_utils.py and base_optimizer.py

Fixed

• Fix pip install
• Fix date

0.4.8 - 2022-05-12

Added

• Add handling of grid search exhaustion
• Add grid search to optimizers
• Add grid generation for search space
• Add prior to dense graph

Changed

• Bump version
• Update RE config id
• True fix
• Bug fix
• Neps search space code improvements
• Lower distance for neighbours in crossover
• Enable crossover for HPs

Fixed

• Fixed cost_cooling and updated test case

0.4.7 - 2022-05-07

Changed

• Bump version, fix async MF
• Bump version, fix async MF
• Fixing async failure modes for MF BO
• Init design fix to MF + error handling if AF sample fails

0.4.6 - 2022-05-05

Changed

• Bump version, fix asynch MF
• Set to max fidelity in AF optimization
• Small fixes
• Change prior decay and initial sampling, MF-BO bug fix
• Update constrained grammar
• Handle hierarchical graph extraction
• Implemented cost cooling
• More flexible towards constraints
• Extract graphs from new representation
• Fix matplotlib version
• Fix call to parent constructor in graph Parameters
• Raise NotImplementedError when not implemented
• Update regularized evolution

Fixed

• Fix a bug in extract config

Removed

• Remove prints
• Remove change

0.4.5 - 2022-05-02

Added

• Adding toggle option for model-search and RS in MF BO

Changed

• Bump version to 0.4.5
• Check for attribute existing

0.4.4 - 2022-05-02

Added

• Add pibo functionality for constrained grammar
• Add fault tolerance example
• Add demonstration of working_directory and previous_working_directory

Changed

• Bump version to 0.4.4
• Fix error with values not sampled when not using fidelities
• Put realistic lr search space in multi fidelity example
• Allow down to pytorch 1.7.0 + fix torch.linalg.cholesky
• Allow down to pytorch 1.8.0
• Debugging multi-worker setup
• Small fixes
• Debugging multi-worker setup
• MF BO fixed to handle pending evals
• Improve repr lisibility
• Improve code for normalizing, include log in normalizing, add sample_batch to mutation_sampler
• Debugging MF BO again
• Debugging MF BO
• MF BO default performance as inf
• Fixing MF BO restarts
• Another quick fix
• Quick fix
• Remember where zero op is placed
• Disable fault_tolerance example test as it makes no sense
• Update constrained grammar

Fixed

• Fix MF ASHA

0.4.3 - 2022-04-28

Added

• Add cost cooling structure
• Adding get_cost utility
• Add handling of fidelity parameters
• Add HP normalization for the GP input
• Adding GP-UCB (with TODOs)
• Add set fidelity to maximum value for the acquisition function optimization
• Add *args to forward call

Changed

• Bump version
• Trick to access the method from child
• Use searcher kwargs
• Commenting GP-UCB until tested locally
• MF BO allows optimization restarts
• Determine rungs wrt max_budget, add todos
• Do not apply decayingpriorweightedacq everytime
• Move kernel construction out of optimizer class
• Update BO optimizer
• Move NOTICE to source code

Removed

• Remove surrogate model from acquisition functions init when still present (bug fix)

0.4.2 - 2022-04-22

Added

• Add matplotlib
• Add id property to graph grammar

Changed

• Use log priors per default, bump version
• Update repetitive graph grammar
• Compose functions directly from string descriptor
• Move to method to utils
• Update example

Fixed

• Fix user priors for categorical hyperparameters

Removed

• Remove setup method

0.4.1 - 2022-04-21

Added

• Add Literal types
• Add wrapper class and remove edge_data argument from forward method
• Add surrogate_model_fit_args to GP constructor
• Add example for PiBO with graphs and HPs
• Add shift to avoid negative log values
• Add log-PiBO backend for graphs from grammars
• Add merge operation to functional backend

Changed

• Bump version
• Use mf_bo per default when space has fidelities
• Move multi fidelity to own fil
• Updating function docstring
• Clip likelihood before saving to list
• Update example
• Merging MF BO with master
• Fixing AF sampling for MF BO
• Edits to make MF BO runnable
• Initial ASHA implementation (not tested)
• Upload example using hierarchical GP
• Also update hierarchical GP accordingly
• Rename work_on to set_state
• Refacot load_results
• Also avoid copy in random_sampler acq_sampler
• Rename use_user_priors to user_priors
• Also avoid copy() in rs
• Move copy() into sample()
• Clean up dependencies
• Only run core tests per default
• Make log parameterizable for PiBO
• Start GP fitting of kernel weights froalways m scratch
• Fix NaN in _unscaled_distance
• Move likelihood clipping into optimization

Removed

• Remove init
• Remove print from base_optimizer
• Remove last graph since it's equal to the full graph
• Remove file

0.4.0 - 2022-04-13

Added

• Add TF backend
• Added required graph generation for sampling to check if graph is valid
• Add support for intermediate build API for hierarchical NAS
• Add set_recursive_attribute to set channels
• Add dict structure as production
• Support for OrdinalHyperparameter in pipeline_space_from_configspace
• Add connection to search space for debugging
• Add connection to hierarchical GP
• Add explanation for hallucination
• Add some TODOs
• Add surrogate from map

Changed

• Bump version
• Set gradients to zero to avoid memory leak
• Speed up examples
• Rename arguments to function parameter
• Move stem / head out of grammar
• Change the node attribute to coarse label and debugged on all datasets
• Debug on changing the order of the graphs: [hierarchical graphs] + [final_graph] to [final_graph] + [hierarchical graphs]
• Replace fit_on_model with set_state
• Change arguments in examples
• Refactore some and add graphs over hierarchies backend
• Also remove id_parse_tree in other examples
• Refactor hierarchical NAS api
• Debug the running of hierarchical kernels:
• Refactor optimizer
• Basic CArBO implementation
• Move BO specific parts out of base_optimizer
• Make terminal_to_graph_repr map optional
• Fix small bugs

Fixed

• Categorical parameters: fix infinite loop and value type
• Fix BO docstring

Removed

• Remove unpacking
• Remove setup
• Remove id_parse argument
• Remove terminal_to_graph_repr

0.3.3 - 2022-03-31

Changed

• Bump version

Removed

• Remove large parts of graph_utils/utils.py

0.3.2 - 2022-03-30

Changed

• Bump version
• Typing + small refactors

0.3.1 - 2022-03-28

Changed

• Update metahyper and bump version
• Increase default patience

Removed

• Remove whitespace

0.3.0 - 2022-03-24

Added

• Adding test for user prior
• Add equal operator for graphs
• Add get_edge_list function for variable topologies
• Add missing files
• Add cost-aware skeleton/api
• Added in robin's additive wl surrogate into the bo framework in neps updates
• Added in robin's additive wl surrogate into the bo framework in neps updates
• Add has_prior flag and compute_prior method
• Add sceleton for utilizing prior in acq
• Add prior weighted acq stub and has_prior attr to search space
• Add link to mkdocs documentation
• Add documentation structure
• Added in robin's additive wl surrogate into the bo framework in neps
• Add multi fidelity BO skeleton
• Add SearchSpace.fidelity and SearchSpace.has_fidelity()
• Add logging of working directory
• Add overwrite option
• Add hint to basic example
• Add nullhandler to neps logger
• Add documentation for showing status repeatedly
• Add documentation for specific torch versions
• Add float userpriors
• Add user_prior flag and categorical user prior
• Add patience from BO class to sampling
• Add documentation to status
• Add a comment for integer parameter sampling
• Support singular result
• Add user prior api
• Add multi fidelity API
• Add primitive operations
• Add get_state and load_state methods for optimizers
• Add cost aware setup
• Add status script
• Add improved logging
• Add python support to soon-to-come
• Add features to readme
• Add documentation to api
• Add arglint hook for the api
• Add stopping control for distributed setting
• Add run_pipeline_args and kwargs
• Add new run_pipeline API
• Add Mapping magic to search space
• Add expected hypervolume improvement AF in develop
• Add from_configspace construction to neps searchspace

Changed

• Deprecate several args in optimizer API and run_pipipeline
• Change default for the watch
• Fix example
• Ensure loss is a float
• Passing budget argument to metahyper
• Upgrade metahyper version
• Fix floats types
• When model building fails, log error and sample randomly
• Fail tests in case of error logged
• Put gp loggers to debug
• Bump metahyper versio
• Check if cuda seed exists in state before using it
• Bump metahyper version
• Bump metahyper version
• Use dev version of metahyper
• Bump metahyper version
• Also run tests when pyproject changed
• Refactor more code
• Json serialization
• Change default value in example
• Complete pibo implementation
• Apply prior weighted when prior exists
• Shorten docstring
• Compute ei incumbent in update step
• Rename reset_surrogate to update
• Extract propose_location from acq function
• Improve comment
• Install metahyper from pypi
• Refactor docs structure and add explanations to CONTRIB.md
• Get 2 versions of gpwl in
• Small refactors and doctring on BayesianOptimization
• Use original error message again
• Improve variable name in README
• Shorten variable names in readme
• Shorten comment in README
• Shorten comment in README
• Improve basic example
• Improve documentation
• Feature pointer to examples more prominently
• Improve parallelization documentation
• Use validation_error in documentation
• Allow more torchvision versions
• Update license in README
• Update LICENSE
• Tweak categorical prior
• Enable user priors
• Use strings for default confidence
• Enhance support for configspace
• Replace random sampler with searchspace.sample_new()
• Simplify randomsampler
• Improve examples
• Improve example
• Mention examples not supported
• Make status importable
• Ignore usage_example
• Improve readme
• Run pre commit
• Update black
• Update to new metahyper version
• Improve post evaluation logging
• Count config_idsfrom 1
• Make hp_kernels and graph_kernels optional
• Update README.md
• Improve usage part in readme
• Deprecate search space api of run
• Update source adaptation for ehvi
• Bug fixes, update hp example to include integer hps
• Make categorical type more general

Fixed

• Fix ei docstring
• Fix ucb still being important
• Fix log determinant adding
• Fix invalid patience sampling implementation
• Fix error when raising ValueError
• Fix typo
• Fix warning with python -m neps.status
• Fix wrong mypy override
• Fix readme
• Fix multi fidelity example
• Fix typo
• Fix metahyper version
• Fix example error in README
• Fix typing
• Fix type errors numerical & search space

Removed

• Remove ids whenever possible. Should also be done for graphs
• Remove ucb and make ehvi private
• Remove dead function
• Remove n_iterations
• Remove unneeded mypy override
• Remove unneeded mypy override
• Remove old api from optimizers
• Remove evaluation and utils.util
• Remove old metahyper
• Remove print statement
• Remove read_and_write
• Remove dead code

[0.2.0] - 2022-01-24

Added

• Added crossover to search space and updated BO loop to handle pending evaluations
• Add print in eval mode
• Add graph grammar multiple repetitive
• Add linear n node topology
• Add typing packages to dependencies
• Add deprecate decorator
• Add comments about metahyper API status
• Add empty documentation
• Add note to mypy ignoring
• Add mypy to pre-commit hooks and CONTRIBUTING
• Add tests badge
• Add license and copyright notice for nasbowl
• Add github workflow for ci
• Add mutation on a level of dense graph (representation of child was missing)
• Add information on test_tmpdir
• Add instructions for failing tests
• Add poetry tip to CONTRIBUTING
• Add basic badges
• Add absolufy-imports for automated relative imports
• Add infos on utils.install_torch and add to CONTRIBUTING
• Add option to return dict with subgraphs
• Add log-EI in acquisition function mapping
• Add conv-bn-relu to basic primitives
• Add log-EI
• Add all examples to tests
• Add tests_tmpdir to pytests and teardown
• Add API functions for BO
• Add initial population size
• Add FMNIST augmentations/normalization
• Add new metahyper api
• Add pytest mention to CONTRIBUTING
• Add mdformat
• Add random search for the new api
• Add numerical parameter
• Add GraphGrammarCell
• Added all missing modules and reduced number of iterations in hierarchical_architecture example to speed up testing
• Add --py37 plus flag for pyupgrade
• Add hierarchical architecture example
• Add graph_dense implementation based on grammars
• Added repetitive graph grammar and added common utility functions for graph grammars
• Add convenient dense n-node DAG topology
• Add more basic primitives and topologies
• Add simple graph grammar search space computation
• Add incumbent printing after a successful run
• Add FashionMNIST
• Add pending evaluations for BO
• Add get_dictionary
• Add HPO example (counting ones) using metahyper
• Add get_dictionary and create_from_id for metahyper communication on search space and hyperparameter level
• Add isomorphism checking
• Add repetitive crossover operation
• Add resolution constrained CFG
• Add unparsing of parse trees
• Add tests for examples
• Add commented pytest to pre-commit for later
• Add example folders
• Add python coding guidelines
• Add local_search mutation for hyperparameters
• Add get_hp_by_name for search_space object by holding all hps in a dictionary
• Add example script for running cnas with new search_space object
• Add objective with api
• Add compatability of BO for metahyper
• Add graph_dense sampling
• Add checkpoint path to exception
• Add graph_dense sampling
• Add hyperparameter class (old parameter) with sampling for child classes
• Add deletion of tensors and mmore workers for data loading
• Add RE
• Add hyperparameter class (old parameter) with sampling for child classes
• Add node subgraph
• Add checkpoint reader
• Add simple API objective base class
• Add seed handling to base class
• Add inv transform for scores
• Add base objective class
• Add checkpointer and id
• Add crossover for constrained grammar
• Add repetitive crossover
• Add depth constrained & constrained grammars. Moved mutation & crossover to grammars
• Add grad clipping
• Add training utilities for DL-based experiments
• Add argument to control if we return opt details, default false.
• Add search space class
• Add metahyper dependency
• Add tabulate dependency
• Added other parameter types
• Added wrapper for graph grammar implementing mutation & crossover
• Add parameter types and pipeline space
• Add direct import of api
• Add initial implementation for evolution acq optimizer
• Add pyupgrade to pre-commit
• Add examples as package to pyproject.toml
• Add NOTICE
• Add simon to license
• Add initial api and dummy example
• Added surrogate model fit args to BO optimizer
• Added random interleaving
• Added seed setting for cuda
• Added incumbent plotting
• Added helper libraries for other helpers
• Added functionality to also check for isomorphisms in history
• Added log file
• Added more logging
• Added error handling for random sampler and removed some deepcopying
• Added some special cases
• Added mutation acq optimizer & moved misc to utils
• Added sum kernel to GP
• Add metahyper to pyproject.toml

Changed

• Update v0.2.0
• Bug fix
• Update training pipeline
• Fix bug
• Return two booleans in crossover
• Bug fix
• Move theta_vector related stuff to combine kernel, apply SMAC lengthscale
• Check whether pipeline_space object supports dimensionality split, check whether loaded result is failed
• Dynamic programming for search space computation
• Fixed remaining mypy errors in graph grammar
• Bug fix
• Fixed more mypy errors
• Resolved merge conflict
• Try to fix poetry install in actions
• Type api
• Rename base_optimizer
• Allow manual trigger
• Fixed some mypy errors
• Bug fix
• Fix pylint warnings
• Move from automl to automl-private
• Move to automl from automl-private
• Change to automl from automl-private
• Update mypy
• Move build system toml part
• Set copyright to general contributors
• Document ci in CONTRIBUTING
• Activate tests on push
• Rename ci
• Rename ci
• Rename ci
• Improve editorconfig documentation
• Save & read seed states
• Small bug fixes
• Improve readme formulations
• Improve pylint ignore warnings section
• Improve dev practices and tooling section
• Improve developer installation section
• Improve tooling tips in CONTRIBUTING
• Apply absolufiy to force relative imports
• Improve content of CONTRIBUTING.md
• Improve content of CONTRIBUTING.md
• Subgraphs is not unrolled anymore
• Also support list as x input for graphs
• Change default value
• Bug fix for log-EI
• Update CONTRIBUTING format and examples guideline
• Change default mutation to simple
• Move example descriptions to example README
• In readme use new style of searchspace construction
• Update README.md
• Bug fixes for RE
• Adjust example for hierarchical NAS for the new metahyper
• Adjust example for NASHPO for the new metahyper
• Make hyperparameters attribute public
• Let pipeline space be defined as dict
• Apply pre-commit
• Update example to use new metahyper
• Make methods private
• Make function calls private
• Update README
• Slim down readme
• Update readme, add example readme
• Update gitignore
• Delete old optimizer abc
• Use list for pylint config
• Update pylint
• Update pyupgrade
• Update isort
• Apply new black version
• Update black version
• Reorder pre-commit hooks
• Change name to neps
• Exclude hyperparameters_architecture from tests for now
• Adjust graph_dense to the changes in graph_grammar
• Merge
• Update examples script, add hpo+nas
• Last naming changes
• Change of search_space initalization and other changes that follow
• Changes in graphs according to the changes of parent classes
• Adjust numerical parameters to the changes in parent classes
• Naming change hyperparameter to parameter (of a search space)
• Upload augmentation lib
• Relative path
• To f-string
• Sync graph_dense with the new graphgrammer implementation
• Switch to value
• Update training pipeline
• Only allow up to python 3.8
• Fix deprecation warning of cholesky
• Copy ResNet Down Block over to graph dense
• Move constrained CFG
• Bug fix
• Bump version for metahyper
• Update README
• Bump metahyper version
• Allow for plotting of temporary results and added axis argument
• Fix evolution bug
• Fix bugs
• Change metahyper branch to master
• Cleanup
• Live_logging does not work for now
• Also rename RS pool_size -> n_candidates
• Fix evolution sampler fitness standardization
• Adjusted BO including renamings.
• Update contributing guidelines
• Work towards new api
• More options for initial samples of evolution acq optimizer
• Make graph flattening optional and fix to_pytorch bug
• Renaming get_graph -> get_graphs
• Change parsing for search_space
• Base hpo benchmarks on new api
• Check back-compatibility, fix name missmatch
• Cleanup
• Fixes of mutation and crossover
• Fix crossover bug
• Also handle concat comb op and subgraphs
• Small fix
• Moved utility class to hierarchical benchmarks
• More consistent with non-api case
• Rename folder
• Bug fix
• Plot shaded region instead of bars
• Changed name of objective to search space
• More refactoring of benchmarks
• Change BO from max to min
• Update benchmark sampling method name
• Update gitignore
• Refactor of hpo benchamrks
• Refactor of hpo benchamrks
• Small change
• Small fixes
• Small adjustments to new design and minor bug fix
• Fixed bugs in pytorch model generation
• Refactor hartmann6 for seperate objective_fn
• Small check if depth constraints match to nonterminals
• Bug fixes in pytorch model generation
• Refactor evaluation of seed in its own fn
• Update benchmarks for refactoring
• Update benchmarks before refactor
• Small fixes
• Moved graph grammar core here and changed directory structure
• Quick fix
• Changed computational flow to be compatible with new evolutionary acqusition optimizers etc.
• Avoid side effect on save_path
• Renamed bo->bayesia_optimization and rs->random_search
• Sort keys for plotting
• Delete empty init
• Move torch install utils to comp_nas/utils
• Make return opt_details truly optional
• Changed python version
• Changed python version
• Moved import to top
• Small bug fix when there are not model proposals
• Small fix if no model-based proposal is used
• Use std error instead of std dev
• Always write out csv
• Online csv writing
• Small changes
• Reuse kernels while kernel fitting and addedparallelized version oof EI
• Shorten argument help, since choices are provided
• Fix typo
• New default values
• Do not use property annotations for getter
• Avoid computing inc in EI over and over again
• Bug fix in mutation acq opt
• Updated README and added error messages
• Bug fix of mutation acq optimizer and minor changes in GP
• Changed default values
• Run formating on torch_install.py
• Avoid error log for HPO kernels and return more information from BO optimizer
• FIX naming bug
• Check if we have list/tuple containing multiple graphs
• Replaced Union with Iterable
• Quick bug fix
• Small changes and fixed minor issue in GP
• Renamed folder & removed initial_design folder
• Move plotting out of BO
• Return opt details dict & added arguments
• ADD nasbenches in NASLib style, query accepts API now.
• Merge
• Move utils to surrogate models and added acqusition optimizer base class
• Renamed kernel folder
• HPO benchmarks follow NASLib now
• HPO benchmarks follow NASLib now
• FIX gp bug
• Complete changes, joint representation for a configuration (before switching to NASLib object). Framework tested on all the benchmarks (nb301 gives an error for the check weights.is_leaf == True, commented out for now). ADD nbconfigfiles for running nasbenches (.pth file needed for nb201, .model file needed for nb301).
• Complete changes, joint representation for a configuration (before switching to NASLib object). Framework tested on all the benchmarks (nb301 gives an error for the check weights.is_leaf == True, commented out for now). ADD nbconfigfiles for running nasbenches (.pth file needed for nb201, .model file needed for nb301).
• Use black profile for isort
• Changed folder name
• New software design for NAS optimizers
• More clean-ups
• ADD nasbenches
• Initial commit
• Initial commit

Fixed

• Fix access to theta_vector in case of no vectorial features
• Fix test bug
• Fix lengthscale parameter for stationary kernels, add split dimensionality for search space and kernels (for now only default categorical/continuous split
• Fix for-else unnecessary
• Fix mypy for pre-commit run -a
• Fix typo in acquisition
• Fix pylint warning
• Fix bug in the search space mutation
• Fix old name
• Fix typo
• Fix readme indents
• Fix imports in examples
• Fix new_result and get_config for metahyper communication
• Fix seeding for benchmarks
• Fix typo
• Fix isort not_skip depreciation

Removed

• Remove unnecessary args in README
• Remove unnecessary args in examples
• Remove worker init fn argument
• Remove license section in README
• Remove more legacy torch import guards
• Remove install_dev_utils add commands to CONTRIBUTING
• Remove pytest comments in pre commit
• Remove torch import guard and add it to api.py
• Remove latex from editorconfig
• Remove attr
• Remove query from graph
• Remove _id for hps
• Remove all imports to hierarchical_nas_benchmarks
• Remove hierarchical_nas_benchmarks dependencies in hierarchical architecture example
• Remove other hierarchical_nas_benchmarks dependencies in graph_dense
• Remove duplication of BO class
• Remove rng passing, add simple mutation for hps
• Remove rng passing, add simple mutation for hps
• Remove results from repo
• Removed unnecessary arguments
• Remove previously used argument
• Removed deepcopying config over and over again
• Remove torch packages from pyproject, add install util
• Remove pylint comments
• Removed wildcard import
• Removed logs and results