The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
orsf_vs now returns a column that contains non-reference coded variable names (see https://github.com/ropensci/aorsf/pull/52).
orsf_vs no longer throws an error when n_predictor_min = 1 is used (see https://github.com/ropensci/aorsf/pull/58).
orsf_summarize_uni now allows specification of a class to summarize for oblique classification forests (see https://github.com/ropensci/aorsf/pull/57).
fixed an issue where orsf would throw an uninformative error when all predictors were categorical (see https://github.com/ropensci/aorsf/pull/56)
oblique random forests can now compute out-of-bag predictions on modified versions of their training data (see https://github.com/ropensci/aorsf/pull/54)
Setting oobag_pred_type to 'none' when growing a forest no longer necessitates the specification of pred_type when calling predict later (see https://github.com/ropensci/aorsf/pull/48).
Setting sample_fraction to 1 will no longer result in empty oobag_rows in the forest object (this would cause R to crash when the forest was passed to C++; see https://github.com/ropensci/aorsf/pull/48)
Re-worked the creation and maintenance of oobag_denom in C++ routines (see https://github.com/ropensci/aorsf/pull/48).
Restricted mean survival time is now used for pred_type = 'time' instead of median survival time (See https://github.com/ropensci/aorsf/pull/46).
Allowed option "time" for pred_type in predict and partial dependence to predict survival time (see https://github.com/ropensci/aorsf/issues/37).
Added pred_spec_auto() for more convenient specification of variables for partial dependence.
Partial dependence now runs much faster with multiple threads.
Added orsf_vint() to compute variable interaction scores using partial dependence.
Added orsf_update(), which can copy and modify an obliqueForest or modify it in place.
Added orsf_control functions for classification, regression, and survival (https://github.com/ropensci/aorsf/pull/25).
optimization implemented for matrix multiplication during prediction (https://github.com/ropensci/aorsf/pull/20)
Fixed an uninitialized value for pd_type
Fixed various issues related to memory leaks
Re-worked internal C++ routines following the design of ranger.
Re-worked how progress is printed to console when verbose_progress is TRUE, following the design of ranger. Messages now indicate the action being taken, the % complete, and the approximate time until finishing the action.
Improved variable importance, following the design of ranger. Importance is now computed tree-by-tree instead of by aggregate. Additionally, mortality is the type of prediction used for importance with survival trees, since mortality does not depend on pred_horizon.
Allowed multi-threading to be performed in orsf(), predict.orsf_fit(), and functions in the orsf_vi() and orsf_pd() family.
Allowed sampling without replacement and sampling a specific fraction of observations in orsf()
Included Harrell’s C-statistic as an option for assessing goodness of splits while growing trees.
Fixed an issue where an uninformative error message would occur when pred_horizon was > max(time) for orsf_summarize_uni. Thanks to @JyHao1 and @DustinMLong for finding this!
orsf() no longer throws errors or warnings when you try to give it a single predictor. A note was added to the documentation in the details of ?orsf that explains why using a single predictor with orsf() is somewhat useless. This was done to resolve https://github.com/mlr-org/mlr3extralearners/issues/259.
predict.orsf_fit now accepts pred_horizon = 0 and returns sensible values. Thanks to @mattwarkentin for the feature request.
added a function to perform variable selection, orsf_vs().
Made variable importance consistent with respect to group_factors. Originally, the output from orsf would have ungrouped VI values while orsf_vi would have grouped values. With this update, orsf defaults to grouped values. The ungrouped values can still be recovered.
Fixed an issue in orsf_pd functions where output data were not being returned on the original scale.
orsf formulas now accepts Surv objects (see https://github.com/ropensci/aorsf/issues/11)
Added verbose_progress input to orsf, which prints messages to console indicating progress.
Allowance of missing values for orsf. Mean and mode imputation is performed for observations with missing data. These values can also be used to impute new data with missing values.
Centering and scaling of predictors is now done prior to growing the forest.
Included rOpenSci reviewers Christopher Jackson, Marvin N Wright, and Lukas Burk in DESCRIPTION as reviewers. Thank you!
Added clarification to docs about pros/cons of different variable importance techniques
Added regression tests for aorsf versus obliqueRSF (they should be similar)
Additional support and tests for functions with long right hand sides
Updated out-of-bag vignette with more appropriate custom functions.
Allow status values in input data to be more general, i.e., not just 0 and 1.
Allow missing values in predict functions, including partial dependence.
Added orsf_control_custom(), which allows users to submit custom functions for identifying linear combinations of inputs while growing oblique decision trees.
Added weights input to orsf, allowing users to over or under fit orsf to specific data in their training set.
Added chf and mort options to predict.orsf_fit(). Mortality predictions are not fully implemented yet - they are not supported in partial dependence or out-of-bag error estimates. These features will be added in a future update.
Core features implemented: fit, interpret, and predict using oblique random survival forests.
Vignettes + Readme covering usage of core features.
Website hosted through GitHub pages, managed with pkgdown.
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.