December 8, 2022

- randomSelection function added. - crissCrossValidate and crissCrossPlot contributions by Harry Robertson added and harmonised to ClassifyR code style. - selectionMethod and classifier defaults become "auto" for crossValidate. Previously only documented but not implemented in code. - predict method for standalone use now finds the correct prediction method for each trained model.

Dario Strbenac authored on 08/12/2022 06:15:07
November 22, 2022

- .doSelection modified so that feature selection doesn't happen twice for selectMulti function. - selectMulti fixed so correct indices are returned relative to the whole training data table, not to an individual assay table.

Dario Strbenac authored on 22/11/2022 04:52:30
October 17, 2022

- Minor fixes.

Dario Strbenac authored on 17/10/2022 03:30:05
October 14, 2022

- Minor fixes to standalone train and predict functions for list of inputs.

Dario Strbenac authored on 14/10/2022 06:45:03

- Typo in train.DataFrame variable fixed. - .predict for DLDA made renamed to DLDA to enable easy dispatch by predict method. - randomForest wrapper now uses ranger as the underlying package instead of randomForest.

Dario Strbenac authored on 14/10/2022 00:30:12
October 12, 2022

- Extreme gradient boosting wrapper added in interfaceXGB.R. - crossValidate gains performanceType parameter that defaults to "auto" which chooses depending on classification or survival task or can be set to any user-specified performance metric. - SelectParams default performanceType is now balanced accuracy, for consistency with other functions. - Vignette text updated to refer to balanced accuracy and to add k-NN and XGB classifiers to classifier table. - NEWS file updated for upcoming Bioconductor release.

Dario Strbenac authored on 12/10/2022 13:05:23
September 7, 2022

- Restored utilities.R.

Dario Strbenac authored on 07/09/2022 00:48:49
September 6, 2022

export colCoxTests

Ellis Patrick authored on 06/09/2022 22:57:36
September 5, 2022

Update utilities.R

Ellis Patrick authored on 05/09/2022 06:03:37 • GitHub committed on 05/09/2022 06:03:37
September 2, 2022

- train methods now exported and can be used. train generic imported from generics package. - simple params now have performanceType specified. - Fixes to training with multiple views but no aggregation. - Fix to parameter tuning in .doTrain.

Dario Strbenac authored on 02/09/2022 06:10:10
August 31, 2022

- train and predict functions created to allow training and prediction to be done independently. But they have the capabilities of crossValidate, such as multi-view methods. Implemented as S3 methods to work well with existing generics defined by R. - runTest, runTests and crossValidate now all utilise prepareData and allow passing in extra parameters for it. - getLocationsAndScales becomes private function. - All reference to targets parameter converted to useFeatures, which allows both the assay and the feature to be specified, rather than only the assay. - .MAEtoWideTable data flattening function is gone. Its functionality is incorporated into prepareData. - Parameter tuning in .doSelection of resubstitution mode now correctly uses training data as the test data instead of accidentally using the testing data.

Dario Strbenac authored on 31/08/2022 11:45:02
August 25, 2022

- getFeatures functions added to simple params settings to extract selected features from within trained model where relevant. - Nearest Shrunken Centroid added as a simple params function and a classifier keyword option.

Dario Strbenac authored on 25/08/2022 05:15:03
August 21, 2022

- Restored runTest, runTests, ModellingParams, CrossValParams as public documented functions. Vignette also restored to explain them. - Constructors for params now expect a character keyword which is then converted into a function internally.

Dario Strbenac authored on 21/08/2022 16:07:38
August 17, 2022

- All references to runTest and runTests in examples and vignette converted to crossValidate. End users should always use crossValidate from now on. - Minor fixes to code mistakes. - Performance tuning of training method parameters chosen within feature selection is now faithfully used in the model training.

Dario Strbenac authored on 17/08/2022 13:55:15
August 14, 2022

- Classifiers and feature selection functions no longer have multiple signaures and are private. - prepareData function to filter and subset input data using common ways, such as missingness and variability. - The variable renaming and storage in Original Feature and Renamed Feature reverted back to column metadata and assay / feature colums. - sampleInfo now reverted back to clinical.

Dario Strbenac authored on 14/08/2022 23:45:28