dream-faster/fold issues and pull requests

#533 - update from internal repo

Pull Request - State: closed - Opened by almostintuitive 9 months ago - 1 comment

#532 - feature(Loop): added `trim_initial_period_after_preprocessing` option, set disable_memory to `True` by default

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#531 - fix(Loop): also added `project_hyperparameters` to `TrainedPipelineCard`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#530 - feature(Loop): added & propagate new `project_hyperparameters` property to Composite.Metadata

Pull Request - State: closed - Opened by almostintuitive about 1 year ago

#529 - feature(Models): omit feature importances in Artifact during `.fit()`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#528 - fix(Inference): Metadata's name is unified across training/inference

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#527 - fix(Inference): set Composite.Metadata correctly for preprocessing pipeline

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#526 - fix(Events): added `FixedForwardWindowIndexerNoTruncation` class, make sure `truncate_if_timeframe_is_smaller==True` outputs numbers instead of nans for the right tail of the time series

Pull Request - State: closed - Opened by almostintuitive about 1 year ago

#525 - feature(Events): `truncate_if_timeframe_is_smaller` to `FixedForwardHorizon`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#524 - chore(Composites): remove unsupported `Cache` Composite

Pull Request - State: closed - Opened by almostintuitive about 1 year ago

#523 - fix(Loop): remove standalone sample_weights parameter, rely exclusively on events' sample weights

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#522 - fix(Wrappers): use get_params(deep=False), reduce number of redundant code in WrapLGBM & WrapXGB

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#521 - feature(Transformations): added self.input_columns* to `Test`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#520 - feature(Composite): propagate `original_artifact` to `postprocess_result_primary()`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#519 - fix(Utils): `get_column_names()` should be matching columns correctly

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#518 - feature(Composites): Ensemble will print out avg correlation of probabilities if verbose is True

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#517 - fix(Utils): should handle `*` in both end and start of string when filtering columns

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#516 - feature(Ensemble): added `reconstruct_predictions_from_probabilities`

Pull Request - State: closed - Opened by almostintuitive about 1 year ago

#515 - fix(Composites): added `inference` to Metadata, Cache is now deployment-compatible

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#514 - feature(Weighting): simplified code, overridden str() in subclasses

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#513 - feature(Loop): added `project_name` to PipelineCard, using it in Cache

Pull Request - State: closed - Opened by almostintuitive about 1 year ago - 1 comment

#512 - fix(Labeling): added assertions, logging in failure modes

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#511 - feature(Transformation): Adding holiday features based on exchanges.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#510 - feature(Labeling): added optional transformation_function to FixedForwardHorizon

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#509 - feature(Labeling): optional `shift_by` parameter added to `FixedForwardHorizon`, it's default value changed to `-(period / 2)` from the hardcoded `-1`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#508 - fix(Types): TrainedPipelineCard.preprocessing is a `TrainedPipelines`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#507 - feature(Utils): added get_last_trained_pipeline()

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#506 - feature(Training): Return processed predictions.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#505 - feature(Scoring): Return insample `ScoreCard` from `train_evaluate`

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#504 - fix(Utils): replacing via traversal should work with Clonables

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#503 - fix(Loop): carry state of Composites when cloning a nested pipeline

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#502 - fix(Composites): cloning should carry over `id`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#501 - fix(Base): Tunable should not inherit from Block

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#500 - feature(Loop): don't trim initial NaNs, RemoveLowVarianceFeatures' business logic

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#499 - feature(Tests): integration tests for inference

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#498 - fix(Inference): extracting the last pipeline correctly

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#497 - fix(Loop): inference is working for preprocessing

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#496 - fix(Loop): inference / updating now uses TrainedPipelineCard

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#495 - feature(Loop): throw exception if there's a single unique value in `y` for the window

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#494 - fix(Loop): Joblib backend

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#493 - feature(Loop): added joblib backend

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#492 - chore(Events): remove UsePredefinedEvents class

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#491 - feature(Models): added wrapper classes, tests

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#490 - feature(Events): added `flip_sign` option to `FixedForwardHorizon`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#489 - feature(Models): added all models from fold_models package

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#488 - feature(Loop): added event_labeler and event_filter to PipelineCard, remove CreateEvents

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#487 - feature(Loop): display which transformation is currently processed with tqdm for the preprocessing step

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#486 - fix(Loop): trained preprocessing pipelines should be stored the same way as normal pipelines

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#485 - fix(Loop): remove `disable_memory` option, added it to `Transformation.Properties`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#484 - feature(Loop): added `PipelineCard` type, optional preprocessing pipeline step

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#483 - refactor(Types): remove old `Transformations` type

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#482 - fix(Transformations): `WrapSKLearnFeatureSelector` should return a fully indexed Artifacts

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#481 - fix(Loop): UsePredefinedEvents now works as expected

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#480 - refactor(Transformations): removed ability for `transform()` to return an optional Artifact

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#479 - refactor(Transformations): removed ability for `transform()` to return an optional Artifact

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#478 - fix(Utils): loop should not return larger artifacts than necessary, _get_maximum_memory_size should not fail on empty pipeline

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#477 - feature(Loop): added `disable_memory` option

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#476 - feature(Loop): added `updateable` property

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#475 - fix(Utils): apply_function_batched wasn't working correctly

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#474 - fix(Training): Path should not default on sequential if there is only one `Fold`.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#473 - fix(Transformations): AddFeatures should apply suffix before calling the function

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#472 - feature(Splitters): moved from using only_last_fold to from_cutoff, only allow pd.Timestamp

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#471 - feature(Splitters): moved from using `only_last_fold` to `from_cutoff`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#470 - feature(Splitters): added `only_last_fold` property

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#469 - feature(Transformations): added `batch_columns` parameter to all transformations that apply a function, to enable reduced memory usage

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#468 - refactor(Loop): Backend is now a separate class, enum was renamed to BackendType

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#467 - fix(Dependencies): upgrade imb-learn, remove scikit-learn constraint

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#466 - refactor(Composites): postprocess_results() renamed to average_results()

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#465 - feature(Transformations): added `output_dtype` parameter to relevant classes

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#464 - fix(Composites): use feather format for faster IO, retaining special characters in column names

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#463 - fix(Composites): `Cache` should treat in-sample and out-of-sample results separately

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#462 - chore(Transformations): removed MakeStationary

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#461 - fix(Loop): enabled copy-on-write, second round of copy=False for all relevant concat methods

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#460 - fix(Loop): concat with copy=False everywhere except handling of memory

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#459 - fix(Transformation): naming output columns of AddWindowFeatures

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#458 - feature(Splitter): `SlidingWindowSplitter`.initial_window allows smaller first train_window

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#457 - chore(Composite): Remove Threshold Finder.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#456 - feature(Transformations): added "keep_original" to all relevant classes, to be able to remove extra concatenation with the original data

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#455 - feature(Loop): ray should be using shared memory storage again

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#454 - fix(Backend): stop using `ray.put()` until we find a solution to the memory leak

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#453 - fix(Loop): also free futures when calling ray.internal.free()

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#452 - fix(Loop): ray backend should release objects it put in the shared memory

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#451 - feature(Tests): better tests for `Cache` Composite, don't copy trained_pipeline when backtesting with `ray`

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#450 - fix(Utils): `concat_on_columns_with_duplicates` returns multiple selected_features columns

Issue - State: open - Opened by almostintuitive over 1 year ago

#449 - feature(Cache): cache per target (now part of Composite.Metadata)

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#448 - feature(Composite): Artifacts are now evaluated after results for Composite primaries.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#447 - feature(Artifacts): start using sparse dtypes

Issue - State: open - Opened by almostintuitive over 1 year ago

#446 - feature(Loop): make Composites aware of fold_index via `metadata` property, introducing `Clonable` interface

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#445 - fix(Composites): FindThreshold should wrap pipeline into a list automatically

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#444 - chore(Types): Exporting all files to get linting and suggestions.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#443 - fix(Linting): Pipeline (Composite) ranamed to `Sequence` to avoid clash with type `Pipeline` and `Pipelines`.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#442 - fix(Composite): Imbalance pipeline export and artifact naming.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#441 - fix(Loop): ray backend shouldn't parallelize training of child_transformations

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#440 - feature(Composite): `FindThreshold` Composite.

Pull Request - State: closed - Opened by szemyd over 1 year ago - 1 comment

#439 - fix(Loop): handle Tuples as Lists, make sure we replace all child transformations when cloning

Pull Request - State: closed - Opened by almostintuitive over 1 year ago

#438 - fix(Transformations): default properties were overwriting unpickled properties, preventing transformation from used in multiprocessing, now parallelizing everything

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#437 - feature(Loop): parallelize backtesting

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#436 - feature(Transformations): use `~` as separator for feature engineering Transformations, add ability to use `*` to select columns, unified column handling logic

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#435 - feature(Concat): allow custom merge function to be passed in

Pull Request - State: closed - Opened by almostintuitive over 1 year ago - 1 comment

#434 - feature(Composites): made `Pipeline` and `Concat` class Tunable

Pull Request - State: open - Opened by almostintuitive over 1 year ago - 1 comment

GitHub / dream-faster/fold issues and pull requests