v1.1.2

News

• AAAI-23 tutorial: We will be giving a tutorial about FLAML in AAAI-23 on Feb 08, 2023. Please find the tutorial agenda in this page.
• Latest research to appear at ICLR-23: Our latest research on multiobjective HPO is accepted as one of the notable-top-5% research papers in ICLR 2023:
  • Please find more details about the method in our paper: Targeted Hyperparameter Optimization with Lexicographic Preferences Over Multiple Objectives. Shaokun Zhang, Feiran Jia, Chi Wang, Qingyun Wu. ICLR 2023 (notable-top-5%).
  • Please find detailed documentation about this new functionality in this page: https://microsoft.github.io/FLAML/docs/Use-Cases/Tune-User-Defined-Function#lexicographic-objectives.
• OpenAI support is added in this release. A notebook example can be found at: https://github.com/microsoft/FLAML/blob/main/notebook/integrate_openai.ipynb

What's Changed

• handle num_samples=-1 by @sonichi in #879
• Bump json5 from 2.2.1 to 2.2.3 in /website by @dependabot in #877
• fix the doc error of customized metrics in automl by @skzhang1 in #883
• Bump ua-parser-js from 0.7.32 to 0.7.33 in /website by @dependabot in #900
• display data head in notebook; exclude None by @sonichi in #885
• Support percentage tolerance for lexicographic optimization by @skzhang1 in #875
• Document how to use the group k-fold by @coffepowered in #894
• update doc for research papers by @qingyun-wu in #912
• stratified group kfold splitter by @sonichi in #899
• add cost budget; move loc of make_dir by @sonichi in #888
• Add bibtex entries for research publications by @Animaholic in #904
• Openai by @sonichi in #905

New Contributors

• @coffepowered made their first contribution in #894
• @Animaholic made their first contribution in #904

Full Changelog: v1.1.1...v1.1.2

v1.1.1

What's Changed

• create dir for log file name by @sonichi in #867
• Do not persist entire AutoMLState in Searcher by @Yard1 in #870
• fix #871: call check_spark only when necessary by @thinkall in #872
• notebook test; spark warning message; reproducibility bug; sequential tuning stop condition by @sonichi in #869

Full Changelog: v1.1.0...v1.1.1

v1.1.0

Highlights

• Spark is now supported as a new parallel tuning backend.
• New tuning capability: targeted tuning with multiple lexicographic objectives. Check out documentation and an example for this new tuning capability.
• New metrics: roc_auc_weighted, roc_auc_ovr_weighted, roc_auc_ovo_weighted.
• New reproducible learner selection method when time_budget is not specified.
• AutoML-related functionaility is moved into a new automl subpackage.

Thanks to all contributors who contributed to this release!

What's Changed

• Bump actions/checkout from 2 to 3 by @dependabot in #699
• fix dependably alert by @skzhang1 in #818
• fix typo by @skzhang1 in #823
• install editable package in codespace by @sonichi in #826
• skip test_hf_data in py 3.6 by @sonichi in #832
• fix typo of output directory by @thinkall in #828
• catch TFT logger bugs by @int-chaos in #833
• roc_auc_weighted metric addition by @shreyas36 in #827
• make performance test reproducible by @sonichi in #837
• Refactor into automl subpackage by @markharley in #809
• Edit the announcement of AAAI-23 tutorial and the KDD tutorial announcement. by @HangHouCheong in #820
• Use get to avoid KeyError by @sonichi in #824
• Update doc by @skzhang1 in #843
• fix bug related to choice by @sonichi in #848
• FAQ about OOM by @sonichi in #849
• Update .NET documentation links by @luisquintanilla in #847
• Added an info reminding user that if no time_budget and no max_iter is specified, then effectively zero-shot AutoML is used by @jingdong00 in #850
• Fix example tune-pytorch where the checkpoint path may be named differently by @jingdong00 in #853
• Format errors on the web. by @skzhang1 in #855
• Add supporting using Spark as the backend of parallel training by @thinkall in #846
• Info and naming by @sonichi in #864

New Contributors

• @thinkall made their first contribution in #828
• @markharley made their first contribution in #809
• @HangHouCheong made their first contribution in #820

Full Changelog: v1.0.14...v1.1.0

v1.0.14

Highlights

• Preparing alpha release of multi-objective hyperparameter tuning with lexicographic preference.
• Fixed issues related to zero-shot automl.
• Multiple improvements to documentation.

What's Changed

• Discord Badge Added by @royninja in #760
• fix bug in current nlp documentation by @liususan091219 in #763
• Multiple objectives hyperparameter tuning with lexicographic preference by @Anonymous-submission-repo in #752
• Indentation corrected by @Kirito-Excalibur in #778
• Included hint to escape brackets for pip setup by @evensure in #786
• Docs by @velezbeltran in #765
• Bump actions/setup-python from 2 to 4 by @dependabot in #700
• Bump codecov/codecov-action from 1 to 3 by @dependabot in #697
• Removed extra | in documentation by @satya-vinay in #790
• fix_alert by @skzhang1 in #793
• Fixed typo by @ElinaAndreeva in #797
• fix_alerts by @skzhang1 in #799
• Documentation about classification/regression task #753 by @royninja in #802
• Added a link to documentation webpage in notebook time_series_forcast by @jingdong00 in #791
• Fix issues related to zero-shot automl by @sonichi in #783
• added the models used for forecasting in documentation by @shreyas36 in #811
• Add performance test for LexiFlow by @Anonymous-submission-repo in #812

New Contributors

• @royninja made their first contribution in #760
• @Anonymous-submission-repo made their first contribution in #752
• @Kirito-Excalibur made their first contribution in #778
• @evensure made their first contribution in #786
• @velezbeltran made their first contribution in #765
• @satya-vinay made their first contribution in #790
• @ElinaAndreeva made their first contribution in #797
• @jingdong00 made their first contribution in #791
• @shreyas36 made their first contribution in #811

Full Changelog: v1.0.13...v1.0.14

v1.0.13

Highlights

• Logging the search_state.config directly to MLflow instead of key-dictionary pair
• Move searcher and scheduler into tune
• Move import location for Ray 2
• Fix NLP dimension mismatch bug

What's Changed

• Dockerfile building problem by @skzhang1 in #719
• Update Contribute.md by @vijaya-lakshmi-venkatraman in #716
• Move import location for Ray 2 by @sonichi in #721
• Fix issue 728 add hyperlink to GitHub location by @Libens-bufo in #731
• Update model.py by @vijaya-lakshmi-venkatraman in #739
• Issue724 by @liususan091219 in #745
• log search_state.config directly instead of under tag config by @prithvikannan in #747
• move searcher and scheduler into tune by @sonichi in #746
• updating the data collator for seq-regression to handle the dim mismatch problem by @liususan091219 in #751
• Update Contribute by @sonichi in #741
• Remove NLP classification head by @liususan091219 in #756

New Contributors

• @vijaya-lakshmi-venkatraman made their first contribution in #716
• @Libens-bufo made their first contribution in #731
• @prithvikannan made their first contribution in #747

Full Changelog: v1.0.12...v1.0.13

v1.0.12

Highlights

• Fix MLFlow bug to support the case where search.state.metric_for_logging is None
• Support customized cross-validation strategy
• Fix SARIMAX seasonal_order parameter name in the wrapper

Thanks to all the contributors for this release!

What's Changed

• chore: Auto update github actions with dependabot by @iemejia in #688
• talks and tutorials by @qingyun-wu in #694
• updating nlp notebook by @liususan091219 in #693
• "intermediate_results" TypeError: argument of type 'NoneType' is not iterable by @liususan091219 in #695
• Update Research.md by @sonichi in #701
• Bump actions/setup-node from 2 to 3 by @dependabot in #698
• Bump actions/cache from 1 to 3 by @dependabot in #696
• Support customized cross-validation strategy by @skzhang1 in #669
• Add $schema to cgmanifest.json by @JamieMagee in #708
• Fix SARIMAX seasonal_order parameter name in the wrapper by @EgorKraevTransferwise in #711

New Contributors

• @iemejia made their first contribution in #688
• @JamieMagee made their first contribution in #708
• @EgorKraevTransferwise made their first contribution in #711

Full Changelog: v1.0.11...v1.0.12

v1.0.11

Highlights

• Preserve the checkpoint when deleting AutoML objects.
• Create no eval set when setting use_best_model to False for catboost.

What's Changed

• add guideline collection by @qingyun-wu in #687
• LightGBM notebook update by @sonichi in #690
• Add preserve_checkpoint to preserve the checkpoint after del by @liususan091219 in #692
• use_best_model for catboost by @sonichi in #679

Full Changelog: v1.0.10...v1.0.11

v1.0.10

This release contains several new features to highlight:

• A major new feature is to support multiple time series in one dataset with a new task named "ts_forecast_panel" and a neural network estimator from pytorch-forecast.
• Allow disabling shuffle for custom splitter.
• Allow explicit specification of whether the choices of a hp have an inherent order.
• Allow skipping data transformation to avoid overhead.
• Support AzureML pipeline tuning.
• Allow log file name to be specified in tune.run and perform logging when ray is used.

There are other improvements for the transformer estimator and bug fixes for config constraints.

What's Changed

• Fixing the issue that FLAML trial number is significantly smaller than Transformers.hyperparameter_search by @liususan091219 in #657
• make test result more stable by @sonichi in #646
• Add pipeline tuner component and dependencies. by @ruizhuanguw in #671
• Skip transform by @jmrichardson in #665
• pull request template by @sonichi in #668
• Update Research.md by @liususan091219 in #672
• Documentation on search space and parallel/sequential tuning by @qingyun-wu in #675
• time series forecasting with panel datasets by @int-chaos in #541
• categorical choice can be ordered or unordered by @sonichi in #677
• Disable shuffle for custom CV by @jmrichardson in #659
• update time series forecast notebook by @int-chaos in #682
• check config constraints for the initial config by @sonichi in #685
• log_file_name in tune.run() by @sonichi in #681
• updating nlp notebook by @liususan091219 in #683
• VW version requirement and documentation on config_constraints vs metric_constraints by @qingyun-wu in #686

New Contributors

• @jmrichardson made their first contribution in #665

Full Changelog: v1.0.9...v1.0.10

v1.0.9

Highlight

• Add the feature names and importance in AutoML
• Update NLP search space and fix several bugs in NLP tasks
• Respect kwargs in AutoML.predict()

What's Changed

• Feature names and importances by @sonichi in #621
• fix NER roberta bug by @liususan091219 in #632
• updating search space by @liususan091219 in #633
• Bump terser from 5.10.0 to 5.14.2 in /website by @dependabot in #642
• This PR fixes the frequent NLP bugs in the other PRs by @liususan091219 in #647
• added "**kwargs" to "predict" by @zzheng93 in #641
• Fix alerts by @skzhang1 in #644
• Update .NET documentation by @luisquintanilla in #643
• Fix HPO evaluation bug by @liususan091219 in #645

New Contributors

• @dependabot made their first contribution in #642
• @zzheng93 made their first contribution in #641
• @luisquintanilla made their first contribution in #643

Full Changelog: v1.0.8...v1.0.9

v1.0.8

• Support latest xgboost version
• Reproducibility improvement for blendsearch
• Allow custom GroupKFold object as split_type
• Bug fix in token classification tasks such as NER
• Allow FLAML_sample_size in starting_points

What's Changed

• log msg about ensemble by @sonichi in #597
• support latest xgboost version by @sonichi in #599
• Fix automl settings in scikit-learn pipeline integration example by @ZviBaratz in #602
• update got version by @sonichi in #607
• min eci depends on cost_attr; cost_attr in ls by @sonichi in #612
• Replaced !pip calls with %pip magic command by @ZviBaratz in #604
• cath URLError by @sonichi in #613
• Updated pre-commit hooks by @ZviBaratz in #609
• Py36 by @sonichi in #614
• Allow custom GroupKFold object as split_type by @sonichi in #616
• Typo fix by @ZviBaratz in #618
• use relative url in doc by @sonichi in #620
• This PR will solve issue, code example format in the doc #622 by @31Sanskrati in #623
• fix ner bug; refactor post processing of TransformersEstimator prediction by @liususan091219 in #615
• isinstance(x, int) -> isinstance(x, (int, np.integer)) by @liususan091219 in #627
• Allow FLAML_sample_size in starting_points by @qingyun-wu in #619
• disable max_len for ner by @liususan091219 in #629
• fix #630 by @adi611 in #631

New Contributors

• @ZviBaratz made their first contribution in #602
• @31Sanskrati made their first contribution in #623
• @adi611 made their first contribution in #631

Full Changelog: v1.0.7...v1.0.8