Merge branch 'develop' of github.com:interpretml/interpret into develop

R/R/binning.R

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 # Author: Paul Koch <[email protected]>
 
-generate_quantile_cuts <- function(
+cut_quantile <- function(
    feature_values, 
    count_samples_per_bin_min, 
    is_humanized, 
@@ -14,14 +14,14 @@ generate_quantile_cuts <- function(
    count_cuts <- as.double(count_cuts)
 
    cuts_lower_bound_inclusive <- .Call(
-      GenerateQuantileCuts_R, 
+      CutQuantile_R, 
       feature_values, 
       count_samples_per_bin_min, 
       is_humanized, 
       count_cuts
    )
    if(is.null(cuts_lower_bound_inclusive)) {
-      stop("error in GenerateQuantileCuts_R")
+      stop("error in CutQuantile_R")
    }
    return(cuts_lower_bound_inclusive)
 }

R/R/booster.R

@@ -153,21 +153,21 @@ create_regression_booster <- function(
 }
 
 generate_model_update <- function(
-   thread_state_boosting_handle, 
+   booster_handle, 
    index_feature_group, 
    learning_rate, 
    count_samples_required_for_child_split_min, 
    max_leaves
 ) {
-   stopifnot(class(thread_state_boosting_handle) == "externalptr")
+   stopifnot(class(booster_handle) == "externalptr")
    index_feature_group <- as.double(index_feature_group)
    learning_rate <- as.double(learning_rate)
    count_samples_required_for_child_split_min <- as.double(count_samples_required_for_child_split_min)
    max_leaves <- as.double(max_leaves)
 
    gain <- .Call(
       GenerateModelUpdate_R, 
-      thread_state_boosting_handle, 
+      booster_handle, 
       index_feature_group, 
       learning_rate, 
       count_samples_required_for_child_split_min, 
@@ -180,13 +180,13 @@ generate_model_update <- function(
 }
 
 apply_model_update <- function(
-   thread_state_boosting_handle
+   booster_handle
 ) {
-   stopifnot(class(thread_state_boosting_handle) == "externalptr")
+   stopifnot(class(booster_handle) == "externalptr")
 
    validation_metric <- .Call(
       ApplyModelUpdate_R, 
-      thread_state_boosting_handle
+      booster_handle
    )
    if(is.null(validation_metric)) {
       stop("error in BoostingStep_R")
@@ -221,19 +221,6 @@ free_boosting <- function(booster_handle) {
    return(NULL)
 }
 
-create_thread_state_boosting <- function(booster_handle) {
-   thread_state_boosting_handle <- .Call(CreateThreadStateBoosting_R, booster_handle)
-   if(is.null(thread_state_boosting_handle)) {
-      stop("Error in CreateThreadStateBoosting")
-   }
-   return(thread_state_boosting_handle)
-}
-
-free_thread_state_boosting <- function(thread_state_boosting_handle) {
-   .Call(FreeThreadStateBoosting_R, thread_state_boosting_handle)
-   return(NULL)
-}
-
 get_best_model <- function(native_ebm_booster) {
    model <- lapply(
       seq_along(native_ebm_booster$feature_groups), 
@@ -308,14 +295,11 @@ native_ebm_booster <- function(
       stop("Unrecognized model_type")
    }
 
-   thread_state_boosting_handle <- create_thread_state_boosting(booster_handle)
-
    self <- structure(list(
       model_type = model_type, 
       n_classes = n_classes, 
       feature_groups = feature_groups, 
-      booster_handle = booster_handle,
-      thread_state_boosting_handle = thread_state_boosting_handle
+      booster_handle = booster_handle
    ), class = "native_ebm_booster")
    return(self)
 }
@@ -370,14 +354,14 @@ cyclic_gradient_boost <- function(
       for(episode_index in 1:max_rounds) {
          for(feature_group_index in seq_along(feature_groups)) {
             gain <- generate_model_update(
-               ebm_booster$thread_state_boosting_handle, 
+               ebm_booster$booster_handle, 
                feature_group_index - 1, 
                learning_rate, 
                min_samples_leaf, 
                max_leaves
             )
 
-            validation_metric <- apply_model_update(ebm_booster$thread_state_boosting_handle)
+            validation_metric <- apply_model_update(ebm_booster$booster_handle)
 
             if(validation_metric < min_metric) {
                min_metric <- validation_metric
@@ -402,7 +386,6 @@ cyclic_gradient_boost <- function(
 
       return(list(model_update = model_update, min_metric = min_metric, episode_index = episode_index))
    }, finally = {
-      free_thread_state_boosting(ebm_booster$thread_state_boosting_handle)
       free_boosting(ebm_booster$booster_handle)
    })
    return(result_list)

R/R/ebm.R

@@ -79,7 +79,7 @@ ebm_classify <- function(
    discretized <- vector("numeric", length(y))
    for(i_feature in 1:n_features) {
       X_feature <- X[, i_feature] # if our originator X matrix is byrow, pay the transpose cost once
-      feature_cuts <- generate_quantile_cuts(
+      feature_cuts <- cut_quantile(
          X_feature, 
          min_samples_bin, 
          humanized, 

R/build_R.bat

@@ -4,15 +4,15 @@ SETLOCAL
 SET root_path=%~dp0
 
 REM exclude the ebm_native directory so that it doesn't delete our existing source files which we mirror separately
-robocopy "%root_path%\" "%root_path%..\tmp\R" /S /PURGE /R:2 /NP /XF build_R.bat /XF cran_format_licence.txt /XF LICENSE /XF Makevars /XF Makevars.interpret /XD ebm_native
+robocopy "%root_path%\" "%root_path%..\tmp\R" /S /PURGE /R:2 /NP /XF build_R.bat /XF cran_formatted_licence.txt /XF LICENSE /XF Makevars /XF Makevars.interpret /XD ebm_native
 ECHO robocopy returned error code %ERRORLEVEL%
 IF %ERRORLEVEL% GEQ 8 (
    EXIT /B %ERRORLEVEL%
 )
 
 copy /Y "%root_path%src\Makevars.interpret" "%root_path%..\tmp\R\src\Makevars"
 
-copy /Y "%root_path%cran_format_licence.txt" "%root_path%..\tmp\R\LICENSE"
+copy /Y "%root_path%cran_formatted_licence.txt" "%root_path%..\tmp\R\LICENSE"
 
 robocopy "%root_path%..\shared\ebm_native" "%root_path%..\tmp\R\src\ebm_native" /S /PURGE /R:2 /NP /XF *.sln /XF *.vcxproj.* /XF ebm_native_exports.* /XF *.md /XD .vs /XD ebm_native_test /XD TestResults /XD special /XD avx512_ebm /XD cuda_ebm /XD cuda_missing_ebm /XF Directory.Build.targets /XF sse2_32.cpp /XF sse2_64.cpp
 ECHO robocopy returned error code %ERRORLEVEL%

R/src/Makevars

@@ -15,33 +15,33 @@ OBJECTS = \
    $(NATIVEDIR)/ApplyModelUpdateValidation.o \
    $(NATIVEDIR)/BinBoosting.o \
    $(NATIVEDIR)/BinInteraction.o \
-   $(NATIVEDIR)/Booster.o \
+   $(NATIVEDIR)/BoosterCore.o \
+   $(NATIVEDIR)/BoosterShell.o \
    $(NATIVEDIR)/CalculateInteractionScore.o \
-   $(NATIVEDIR)/CutRandom.o \
-   $(NATIVEDIR)/DataFrameBoosting.o \
-   $(NATIVEDIR)/DataFrameInteraction.o \
-   $(NATIVEDIR)/DataFrameShared.o \
-   $(NATIVEDIR)/DebugEbm.o \
+   $(NATIVEDIR)/CutQuantile.o \
+   $(NATIVEDIR)/CutUniform.o \
+   $(NATIVEDIR)/CutWinsorized.o \
+   $(NATIVEDIR)/data_set_shared.o \
+   $(NATIVEDIR)/DataSetBoosting.o \
+   $(NATIVEDIR)/DataSetInteraction.o \
+   $(NATIVEDIR)/debug_ebm.o \
    $(NATIVEDIR)/Discretize.o \
    $(NATIVEDIR)/FeatureGroup.o \
-   $(NATIVEDIR)/FindBestBoostingSplitsPairs.o \
-   $(NATIVEDIR)/FindBestInteractionScorePairs.o \
    $(NATIVEDIR)/GenerateModelUpdate.o \
-   $(NATIVEDIR)/GenerateQuantileCuts.o \
-   $(NATIVEDIR)/GenerateUniformCuts.o \
-   $(NATIVEDIR)/GenerateWinsorizedCuts.o \
-   $(NATIVEDIR)/GrowDecisionTree.o \
-   $(NATIVEDIR)/InitializeResiduals.o \
-   $(NATIVEDIR)/InteractionDetector.o \
-   $(NATIVEDIR)/InterpretableNumerics.o \
-   $(NATIVEDIR)/RandomExternal.o \
+   $(NATIVEDIR)/InitializeGradientsAndHessians.o \
+   $(NATIVEDIR)/InteractionCore.o \
+   $(NATIVEDIR)/InteractionShell.o \
+   $(NATIVEDIR)/interpretable_numerics.o \
+   $(NATIVEDIR)/PartitionOneDimensionalBoosting.o \
+   $(NATIVEDIR)/PartitionRandomBoosting.o \
+   $(NATIVEDIR)/PartitionTwoDimensionalBoosting.o \
+   $(NATIVEDIR)/PartitionTwoDimensionalInteraction.o \
    $(NATIVEDIR)/RandomStream.o \
+   $(NATIVEDIR)/sampling.o \
    $(NATIVEDIR)/SamplingSet.o \
    $(NATIVEDIR)/SegmentedTensor.o \
    $(NATIVEDIR)/SumHistogramBuckets.o \
    $(NATIVEDIR)/TensorTotalsBuild.o \
-   $(NATIVEDIR)/ThreadStateBoosting.o \
-   $(NATIVEDIR)/ThreadStateInteraction.o \
    $(NATIVEDIR)/common_c/common_c.o \
    $(NATIVEDIR)/common_c/logging.o \
    $(NATIVEDIR)/compute/Loss.o \

R/src/Makevars.interpret

@@ -15,33 +15,33 @@ OBJECTS = \
    $(NATIVEDIR)/ApplyModelUpdateValidation.o \
    $(NATIVEDIR)/BinBoosting.o \
    $(NATIVEDIR)/BinInteraction.o \
-   $(NATIVEDIR)/Booster.o \
+   $(NATIVEDIR)/BoosterCore.o \
+   $(NATIVEDIR)/BoosterShell.o \
    $(NATIVEDIR)/CalculateInteractionScore.o \
-   $(NATIVEDIR)/CutRandom.o \
-   $(NATIVEDIR)/DataFrameBoosting.o \
-   $(NATIVEDIR)/DataFrameInteraction.o \
-   $(NATIVEDIR)/DataFrameShared.o \
-   $(NATIVEDIR)/DebugEbm.o \
+   $(NATIVEDIR)/CutQuantile.o \
+   $(NATIVEDIR)/CutUniform.o \
+   $(NATIVEDIR)/CutWinsorized.o \
+   $(NATIVEDIR)/data_set_shared.o \
+   $(NATIVEDIR)/DataSetBoosting.o \
+   $(NATIVEDIR)/DataSetInteraction.o \
+   $(NATIVEDIR)/debug_ebm.o \
    $(NATIVEDIR)/Discretize.o \
    $(NATIVEDIR)/FeatureGroup.o \
-   $(NATIVEDIR)/FindBestBoostingSplitsPairs.o \
-   $(NATIVEDIR)/FindBestInteractionScorePairs.o \
    $(NATIVEDIR)/GenerateModelUpdate.o \
-   $(NATIVEDIR)/GenerateQuantileCuts.o \
-   $(NATIVEDIR)/GenerateUniformCuts.o \
-   $(NATIVEDIR)/GenerateWinsorizedCuts.o \
-   $(NATIVEDIR)/GrowDecisionTree.o \
-   $(NATIVEDIR)/InitializeResiduals.o \
-   $(NATIVEDIR)/InteractionDetector.o \
-   $(NATIVEDIR)/InterpretableNumerics.o \
-   $(NATIVEDIR)/RandomExternal.o \
+   $(NATIVEDIR)/InitializeGradientsAndHessians.o \
+   $(NATIVEDIR)/InteractionCore.o \
+   $(NATIVEDIR)/InteractionShell.o \
+   $(NATIVEDIR)/interpretable_numerics.o \
+   $(NATIVEDIR)/PartitionOneDimensionalBoosting.o \
+   $(NATIVEDIR)/PartitionRandomBoosting.o \
+   $(NATIVEDIR)/PartitionTwoDimensionalBoosting.o \
+   $(NATIVEDIR)/PartitionTwoDimensionalInteraction.o \
    $(NATIVEDIR)/RandomStream.o \
+   $(NATIVEDIR)/sampling.o \
    $(NATIVEDIR)/SamplingSet.o \
    $(NATIVEDIR)/SegmentedTensor.o \
    $(NATIVEDIR)/SumHistogramBuckets.o \
    $(NATIVEDIR)/TensorTotalsBuild.o \
-   $(NATIVEDIR)/ThreadStateBoosting.o \
-   $(NATIVEDIR)/ThreadStateInteraction.o \
    $(NATIVEDIR)/common_c/common_c.o \
    $(NATIVEDIR)/common_c/logging.o \
    $(NATIVEDIR)/compute/Loss.o \