Changes from Version 0.1 to 0.1.1 [14-Apr-2014]
* Adding logging to track usage of TSDT.
* Removing missing response when computing subgroup mean response.
Changes from Version 0.1.1 to 0.1.2 [30-Apr-2014]
* Sorting split variables in collapse_redundant_split to get unique list
of splits -- i.e. X2<xxxxx & X1>=xxxxx should not be considered a
different subgroup from X1>=xxxxx & X2<xxxxxx.
Changes from Version 0.1.2 to 0.1.3 [09-May-2014]
* Change permute_method to either accept the character name of the
permute function or the name of the function directly (.i.e. not as
a string.)
* Remove Adjusted_Pvalue and associated Strength for Overall subgroup
since these have no meaningful interpretation.
Changes from Version 0.1.3 to 0.1.4 [12-May-2014]
* Remove Internal_Consistency and External_Consistency for Overall
subgroup because they have no meaningful interpretation.
Changes from Version 0.1.4 to 0.1.5 [16-May-2014]
* Embedded logging functionality in tryCatch block so user's do not get
errors or warnings if logging fails.
Changes from Version 0.1.5 to 0.1.6 [01-Jun-2014]
* Round variables to six decimal places when subsetting data to create
subgroups from trees built by ctree. The numerical precision of
ctree is apparently limited to six decimal places so if the
covariate data is not also rounded to six decimal places before
subsetting then subgroup sizes might differ from the ctree internal
representation of the tree. This could result in some subgroups
having fewer than the minimum required number of subjects.
Changes from Version 0.1.6 to 0.1.7 [10-Jun-2014]
* Change the covariate data passed into get_null_scores() from the
covariates data.frame to the source_data data.frame so the
WEIGHTS__ variables is also passed when present.
Changes from Version 0.1.7 to 0.1.8 [05-Jul-2014]
* Add trace parameter to TSDT to print progress of permutation
computation to console.
Changes from Version 0.1.8 to 0.1.9 [30-Jul-2014]
* Fix error in subsample. It was dropping the expermimental arm.
* Added code to transform a binary variable (taking two unique values,
but they could be any value) to a strictly {0,1} variable.
Changes from Version 0.1.9 to 0.1.10 [03-Aug-2014]
* Move httpRequest from Depends to Suggests
Changes from Version 0.1.10 to 0.1.11 [10-Nov-2014]
* Change whitespace in character variable values and factor levels to
an underscore.
Changes from Version 0.1.11 to 0.1.12 [13-Nov-2014]
* Change deprecated multicore package to parallel.
Changes from Version 0.1.12 to 0.1.13 [19-Nov-2014]
* Remove records with missing response.
Changes from Version 0.1.13 to 0.1.14 [24-Nov-2014]
* Add code to handle when quantile__ in TSDT_scoring_functions.R is of
class numeric.
Changes from Version 0.1.14 to 0.1.15 [27-Nov-2014]
* Add vector of NULL_SCORES to distributions slot of TSDT class.
Changes from Version 0.1.15 to 0.1.16 [02-Dec-2014]
* Add code to return NA for survival_time_quantile in
TSDT_scoring_functions.R if specified survival quantile is not
obtained.
Changes from Version 0.1.16 to 0.1.17 [19-Dec-2014]
* Add code to return compute weighted mean response in in-bag and oob
samples if case weights provided.
Changes from Version 0.1.17 to 0.1.18 [03-Jan-2015]
* Add code to print weighted mean response (where appropriate) in
summary results. Version 0.1.17 computed weighted means for
individual trees, but reported unweighted means in summary results.
Changes from Version 0.1.18 to 0.1.19 [06-Jan-2015]
* Add a dummy Logger function for when access to a web host for
processing the logging POST requests is unavailable.
* Subsetting a survival response destroys the Surv object and reduces it
to a two-column matrix. Added code to construct the survival object
after subsetting for both one-arm and two-arm samples.
Changes from Version 0.1.19 to 0.1.20 [24-Jan-2015]
* Add code to use ceiling of user-provided minimum subgroup sizes in case
non-integer values are given.
Changes from Version 0.1.20 to 0.1.21 [12-Feb-2015]
* Correct typo "surival" to "survival" in TSDT_helper_functions.R.
Changes from Version 0.1.21 to 0.1.22 [24-Aug-2015]
* Change definition of superior subgroup when scoring function is
proportion_desirable_response. A superior subgroup should *always*
have an Inbag_Score that is *greater* than overall. Do not change
to less than overall for a decreasing desirable response. This is
already accounted for by computing the Inbag_Score as the
proportion of records with response == 0 rather than response == 1.
Changes from Version 0.1.22 to 0.1.23 [26-Aug-2015]
* Compute permutation p-value as (k+1)/(n+1)
where
k is the number of null scores (strictly) greater than the observed
score (the observed score is the product of internal and external
consistency for a subgroup
n is the total number of perumuted scores.
Previous k was not computed with strict inequality.
Changes from Version 0.1.23 to 0.1.24 [30-Aug-2015]
* Allow user to specify different minimum subgroup sizes for in-bag and
and out-of-bag data.
* Allow user to specify minimum subgroup size as a proportion of the
sample. Previously, minumim subgroup sizes were specified as an
absolute number. Both methods are now available.
Changes from Version 0.1.24 to 0.1.25 [16-Sep-2015]
* Change function space2underscore() to operate on data columns where
any class in the column's class hierarchy is "factor" or
"character". This change accomodates ordered factors, which have
class '"ordered" "factor"'.
Changes from Version 0.1.25 to 0.1.26 [29-Sep-2015]
* Set permutation-adjusted p-value to 1 when the observed score is zero
i.e. ExternalConsistency is zero.
Changes from Version 0.1.26 to 0.1.27 [30-Sep-2015]
* Use RNGkind( kind = "L'Ecuyer-CMRG" ) and mc.reset.stream() to get
reproducible results in a multicore environment.
Changes from Version 0.1.27 to 0.1.28 [10-Feb-2016]
* Keep scoring_function_parameters as a list in final TSDT object --
i.e. remove call to unlist().
* Add restricted mean survival time as scoring function for surival
response.
Changes from Version 0.1.28 (BSID) to 0.0.1 (TSDT) [23-Feb-2016]
* The TSDT portion of the BSID package has been removed from BSID
and placed in a new package called TSDT.
Changes from Version 0.0.1 to 0.0.2 [29-Mar-2016]
* Change default minimum subgroup size from 20% of each treatment arm
to 10% of each treatment arm.
Changes from Version 0.0.2 to 0.0.3 [30-Mar-2016]
* Add code to Roxygen comments for summary method documentation to export
the summary() method and bind the method to the TSDT class.
Changes from Version 0.0.3 to 0.0.4 [05-Apr-2016]
* Add documentation for function get_suggested_subgroup().
Changes from Version 0.0.4 to 0.0.5 [10-Apr-2016]
* Add code to remove variables from tree_builder_parameters$cost for the
competitor trees as primary splitting variables are removed.
Changes from Version 0.0.5 to 0.0.6 [10-May-2016]
* Add cpart() from CustomPartition as a new tree-building algorithm.
* If user enters min_subgroup_n_* value that is greater than the total
count of trt or control subjects an error is returned.
* Make superior_subgroups$Subgroup character rather than factor
Changes from Version 0.0.6 to 0.0.7 [19-May-2016]
* Export all scoring functions so they are callable by the user.
* Do not use covariates in survival scoring functions. This often leads
to non-convergence. Instead, estimate survival models on intercept only.
* Reset all factor variables in covariates so no levels with zero records
remain.
* Reconstruct Surv object after subsetting to specific treatment arm in
survival scoring functions.
Changes from Version 0.0.7 to 0.0.8 [15-Jul-2016]
* Use double brakcets when populating a list of S4 objects: e.g.
mylist[[i]] <- S4_object. A recent version of R deprecates the single
bracket notation.
Changes from Version 0.0.8 to 0.0.9 [19-Jul-2016]
* Add hazard_ratio as a new scoring_function.
* Change sampling_proportion parameter in documentation to inbag_proportion
to be consistent with code.
Changes from Version 0.0.9 to 0.0.10 [19-Jul-2016]
* Remove elements from tree_builder_paramters$cost vector by position index
rather than name in case the vector elements are unnamed.
Changes from Version 0.0.10 to 0.0.11 [27-Jul-2016]
* Add documentation for TSDT::get_cutpoints accessor method and export to
user.
Changes from Version 0.0.11 to 0.0.12 [24-Aug-2016]
* Make sure R doesn't demote covariates data.frame to vector when removing
columns while growing growing competitor trees.
Changes from Version 0.0.12 to 0.0.13 [11-Nov-2016]
* The bootstrap function was only retaining unique rows in the in-bag data.
This made the in-bag data have fewer rows than the original data. The
bootstrap function now retains all resampled rows (including duplicate
rows).
Changes from Version 0.0.13 to 0.0.14 [16-Nov-2016]
* Remove commented-out code.
Changes from Version 0.0.14 to 0.0.15 [16-Nov-2016]
* Export cutpoints and distribution functions to user.
Changes from Version 0.0.15 to 0.0.16 [31-Jul-2017]
* Translate logical inequalities that appear in character variable values
so they will not be interpreted as numeric subgroups by parse_subgroups.
The translated values are converted back to their original values in the
final output.
The translations are as follows:
">=" --> "%%__GE__%%"
"<=" --> "%%__LE__%%"
">" --> "%%__GT__%%"
"<" --> "%%__LT__%%"
"=" --> "%%__EQ__%%"
"" --> "%%__EMPTY_STRING__%%"
Changes from Version 0.0.16 to 0.0.17 [08-Nov-2017]
* Prepare code for publication on CRAN. This involves removing references
to CustomPartition, removing logic for logging, and removing code for
sourcing other scripts.
Changes from Version 0.0.17 to 0.1.0 [29-Sep-2018]
* Version for initial release on CRAN. Implement changes to pass CRAN checks.
Changes from Version 0.0.17 to 1.0.2 [20-Aug-2019]
* Set trt_control and permute_arm to NULL when trt is NULL
* Correct desirable_response_proportion when desirable_response is
decreasing
Changes from Version 1.0.2 to 1.0.3 [21-Aug-2019]
* Correct desirable_response_proportion when desirable_response is
decreasing and trt_var does not exist
* Do not require subgroup mean response to be superior to overall mean
if scoring function is desirable_response_proportion.
Changes from Version 1.0.3 to 1.0.4 [22-Aug-2019]
* Make mean_response the default scoring function for a binary response
single-arm model, and treatment_effect the default scoring function for a
binary response two-arm model.
Changes from Version 1.0.4 to 1.0.5 [23-Aug-2019]
* Fix examples that are inconsistent with underlying code
Changes from Version 1.0.5 to 1.0.6 [18-May-2020]
* Allow user to set depth of competitor trees
Changes from Version 1.0.6 to 1.0.7 [06-Apr-2022]
* Update examples to fix errors in R CMD check. This involved enclosing
examples intended to demonstrate errors in \dontrun{...}.
* Fix unfactor() function. Should use single "&" inside any() function.
* Use is() function from methods library to test class of data objects
rather than comparing to a string.
Changes from Version 1.0.7 to 1.0.8 [07-Jan-2025]
* Escape curly braces in documentation to pass CRAN checks.
* Changed default maxdepth of trees from .Machine$integer.max to 30 to
satisfy CRAN checks.