Replace sprintf with snprintf.
Drop the dependency on rattle.data since it's not available.
Replace JSS URLs with DOIs.
Correct invalid links in NEWS.
Correct the uninitialized variable pos in
src/c4_5_var_selector.cpp.
Suggest markdown and rmarkdown as dependency (See Move the markdown package from Imports to Suggests #1864).
Add authors' ORCID identifiers.
Change http to https.
Use rattle.data for datasets instead of rattle.
Remove deprecated Make variables with prefix CXX1X
(CHANGES
IN R 3.4.0).
Increase the responsiveness when user interrupt.
Add registration for native routines.
Segment fault and memory not mapped when user interrupt.
Add description for installation from GitHub.
Add citation info.
Change the output of predict.wsrf to be a list of
predictions of different types.
Fix bug introduced in version 1.6.25 when adding
wsrf.default for separate inputs and response.
Throw more informative error when variable (attribute) not
found in newdata of predict.wsrf.
Add interrupt check in predict.wsrf.
Add a new argument nodesize to change the minimum
node size.
Add a method wsrf.default to deal with separate
inputs (x) and response (y), besides
wsrf.formula with formula.
Drop support of Boost for multithreading.
Speed up data loading in wsrf and predict.
Add subject classifications for the content of the package.
Add badges in README.
Fix undefined column error when column name is a non-syntactic name, such as names contain whitespace.
Fix failure to switch to the code for the right compiler version.
Add URL and BugReports into DESCRIPTION.
Fix out-of-bound access bug triggered by passing test data without the column of target variable.
Add missing imports to correct the NOTEs from the newly added check item of R-devel on 2015-06-29.
Correct the URLs of CRAN package pages to their canonical form in the documentation.
Add more examples in the manual.
Update documentation.
Change vignette from Rnw to Rmd.
Negative information gain when splitting nodes leads to
-nan when normalized by sqrt().
No-matched predicted results when predict type is
"prob" (Thank Erich Studerus for reporting this
bug).
Fix a bug when assessing variable importance: Vector out of range access.
Update documentation: Make it clear that the internal algorithm used.
Move all predict functionality from R into C++
code.
Correct all compilation warnings on Windows, Linux, and OS X and errors from memtest.
Add new function subset to get a sub-forest from a
wsrf model.
Add new function combine to merge multiple
wsrf models into bigger one.
Add confusion, oot.times, predicted,
useweights, mtry in the wsrf model list.
Add clusterlogfile in the parameter list of function
wsrf for debug.
Function predict.wsrf returns NAs when there are
missing values in newdata.
Remove function summary.wsrf, moving its
functionality into print.wsrf, so that the default outputs
of print.wsrf gives a summary of the model.
Tree structure printed by function print.wsrf is
slightly different from previous version: Decisions from the same
test node (internal node) have the same item number.
Function oobErrorRate is renamed into
oob.error.rate and made as generic.
Refactor code, making rooms for future modifications: Remove unused code; Redefine some classes.
Operate on R objects directly in C++ code instead of copied ones.
Improve time performance.
Correct weighted random sampling.
Replace srand() with R RNG functions.
Add new function wsrfParallelInfo for querying the
underlying parallel implementation (parallel or not).
The predict method requires a dummy outcome column to be in the new data (Thank Max Kuhn for reporting this bug).
Adapt R's make variable CXX_STD to configure installation.
List packages used by vignettes code in DESCRIPTION.
Let R to build the vignettes so that it can be visible on wsrf.
added documents:
vignettes
README.Rd
NEWS.md
provided Windows support (no multi-threading functionality, but can run in distributed way)
bug fixed: when used in distributed environment, variable name info is missing, therefore, "Error: Training Set is empty" occurs.