mirror of https://github.com/microsoft/autogen.git
1 Commits
| Author | SHA1 | Message | Date |
|---|---|---|---|
EgorKraevTransferwise
|
5245efbd2c
|
Factor out time series-related functionality into a time series Task object (#989)
* Refactor into automl subpackage Moved some of the packages into an automl subpackage to tidy before the task-based refactor. This is in response to discussions with the group and a comment on the first task-based PR. Only changes here are moving subpackages and modules into the new automl, fixing imports to work with this structure and fixing some dependencies in setup.py. * Fix doc building post automl subpackage refactor * Fix broken links in website post automl subpackage refactor * Fix broken links in website post automl subpackage refactor * Remove vw from test deps as this is breaking the build * Move default back to the top-level I'd moved this to automl as that's where it's used internally, but had missed that this is actually part of the public interface so makes sense to live where it was. * Re-add top level modules with deprecation warnings flaml.data, flaml.ml and flaml.model are re-added to the top level, being re-exported from flaml.automl for backwards compatability. Adding a deprecation warning so that we can have a planned removal later. * Fix model.py line-endings * WIP * WIP - Notes below Got to the point where the methods from AutoML are pulled to GenericTask. Started removing private markers and removing the passing of automl to these methods. Done with decide_split_type, started on prepare_data. Need to do the others after * Re-add generic_task * Most of the merge done, test_forecast_automl fit succeeds, fails at predict() * Remaining fixes - test_forecast.py passes * Comment out holidays-related code as it's not currently used * Further holidays cleanup * Fix imports in a test * tidy up validate_data in time series task * Test fixes * Fix tests: add Task.__str__ * Fix tests: test for ray.ObjectRef * Hotwire TS_Sklearn wrapper to fix test fail * Attempt at test fix * Fix test where val_pred_y is a list * Attempt to fix remaining tests * Push to retrigger tests * Push to retrigger tests * Push to retrigger tests * Push to retrigger tests * Remove plots from automl/test_forecast * Remove unused data size field from Task * Fix import for CLASSIFICATION in notebook * Monkey patch TFT to avoid plotting, to fix tests on MacOS * Monkey patch TFT to avoid plotting v2, to fix tests on MacOS * Monkey patch TFT to avoid plotting v2, to fix tests on MacOS * Fix circular import * remove redundant code in task.py post-merge * Fix test: set svd_solver="full" in PCA * Update flaml/automl/data.py Co-authored-by: Chi Wang <wang.chi@microsoft.com> * Fix review comments * Fix task -> str in custom learner constructor * Remove unused CLASSIFICATION imports * Hotwire TS_Sklearn wrapper to fix test fail by setting optimizer_for_horizon == False * Revert changes to the automl_classification and pin FLAML version * Fix imports in reverted notebook * Fix FLAML version in automl notebooks * Fix ml.py line endings * Fix CLASSIFICATION task import in automl_classification notebook * Uncomment pip install in notebook and revert import Not convinced this will work because of installing an older version of the package into the environment in which we're running the tests, but let's see. * Revert c6a5dd1a0 * Fix get_classification_objective import in suggest.py * Remove hcrystallball docs reference in TS_Sklearn * Merge markharley:extract-task-class-from-automl into this * Fix import, remove smooth.py * Fix dependencies to fix TFT fail on Windows Python 3.8 and 3.9 * Add tensorboardX dependency to fix TFT fail on Windows Python 3.8 and 3.9 * Set pytorch-lightning==1.9.0 to fix TFT fail on Windows Python 3.8 and 3.9 * Set pytorch-lightning==1.9.0 to fix TFT fail on Windows Python 3.8 and 3.9 * Disable PCA reduction of lagged features for now, to fix svd convervence fail * Merge flaml/main into time_series_task * Attempt to fix formatting * Attempt to fix formatting * tentatively implement holt-winters-no covariates * fix forecast method, clean class * checking external regressors too * update test forecast * remove duplicated test file, re-add sarimax, search space cleanup * Update flaml/automl/model.py removed links. Most important one probably was: https://robjhyndman.com/hyndsight/ets-regressors/ Co-authored-by: Chi Wang <wang.chi@microsoft.com> * prevent short series * add docs * First attempt at merging Holt-Winters * Linter fix * Add holt-winters to TimeSeriesTask.estimators * Fix spark test fail * Attempt to fix another spark test fail * Attempt to fix another spark test fail * Change Black max line length to 127 * Change Black max line length to 120 * Add logging for ARIMA params, clean up time series models inheritance * Add more logging for missing ARIMA params * Remove a meaningless test causing a fail, add stricter check on ARIMA params * Fix a bug in HoltWinters * A pointless change to hopefully trigger the on and off KeyError in ARIMA.fit() * Fix formatting * Attempt to fix formatting * Attempt to fix formatting * Attempt to fix formatting * Attempt to fix formatting * Add type annotations to _train_with_config() in state.py * Add type annotations to prepare_sample_train_data() in state.py * Add docstring for time_col argument of AutoML.fit() * Address @sonichi's comments on PR * Fix formatting * Fix formatting * Reduce test time budget * Reduce test time budget * Increase time budget for the test to pass * Remove redundant imports * Remove more redundant imports * Minor fixes of points raised by Qingyun * Try to fix pandas import fail * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Try to fix pandas import fail, again * Formatting fixes * More formatting fixes * Added test that loops over TS models to ensure coverage * Fix formatting issues * Fix more formatting issues * Fix random fail in check * Put back in tests for ARIMA predict without fit * Put back in tests for lgbm * Update test/test_model.py cover dedup * Match target length to X length in missing test --------- Co-authored-by: Mark Harley <mark.harley@transferwise.com> Co-authored-by: Mark Harley <mharley.code@gmail.com> Co-authored-by: Qingyun Wu <qingyun.wu@psu.edu> Co-authored-by: Chi Wang <wang.chi@microsoft.com> Co-authored-by: Andrea W <a.ruggerini@ammagamma.com> Co-authored-by: Andrea Ruggerini <nescio.adv@gmail.com> Co-authored-by: Egor Kraev <Egor.Kraev@tw.com> Co-authored-by: Li Jiang <bnujli@gmail.com> |