outliertree 0.1.0

Sign up to get free protection for your applications and to get access to all the features.
Files changed (25) hide show
  1. checksums.yaml +7 -0
  2. data/CHANGELOG.md +3 -0
  3. data/LICENSE.txt +674 -0
  4. data/NOTICE.txt +14 -0
  5. data/README.md +107 -0
  6. data/ext/outliertree/ext.cpp +260 -0
  7. data/ext/outliertree/extconf.rb +21 -0
  8. data/lib/outliertree.rb +17 -0
  9. data/lib/outliertree/dataset.rb +35 -0
  10. data/lib/outliertree/model.rb +128 -0
  11. data/lib/outliertree/result.rb +190 -0
  12. data/lib/outliertree/version.rb +3 -0
  13. data/vendor/outliertree/LICENSE +674 -0
  14. data/vendor/outliertree/README.md +155 -0
  15. data/vendor/outliertree/src/Makevars +3 -0
  16. data/vendor/outliertree/src/RcppExports.cpp +123 -0
  17. data/vendor/outliertree/src/Rwrapper.cpp +1225 -0
  18. data/vendor/outliertree/src/cat_outlier.cpp +328 -0
  19. data/vendor/outliertree/src/clusters.cpp +972 -0
  20. data/vendor/outliertree/src/fit_model.cpp +1932 -0
  21. data/vendor/outliertree/src/misc.cpp +685 -0
  22. data/vendor/outliertree/src/outlier_tree.hpp +758 -0
  23. data/vendor/outliertree/src/predict.cpp +706 -0
  24. data/vendor/outliertree/src/split.cpp +1098 -0
  25. metadata +150 -0
data/vendor/outliertree/README.md ADDED
@@ -0,0 +1,155 @@
1
+ # OutlierTree
2
+
3
+ Explainable outlier/anomaly detection based on smart decision tree grouping, similar in spirit to the GritBot software developed by RuleQuest research. Written in C++ with interfaces for R and Python. Supports columns of types numeric, categorical, binary/boolean, and ordinal, and can handle missing values in all of them. Ideal as a sanity checker in exploratory data analysis.
4
+
5
+ # How it works
6
+
7
+ Will try to fit decision trees that try to "predict" values for each column based on the values of each other column. Along the way, each time a split is evaluated, it will take the observations that fall into each branch as a homogeneous cluster in which it will search for outliers in the 1-d distribution of the column being predicted. Outliers are determined according to confidence intervals on this 1-d distribution, and need to have a large gap with respect to the next observation in sorted order to be flagged as outliers. Since outliers are searched for in a decision tree branch, it will know the conditions that make it a rare observation compared to others that meet the same conditions, and the conditions will always be correlated with the target variable (as it's being predicted from them).
8
+
9
+ As such, it will only be able to detect outliers that can be described through a decision tree logic, and unlike other methods such as [Isolation Forests](https://github.com/david-cortes/isotree), will not be able to assign an outlier score to each observation, nor to detect outliers that are just overall rare, but will always provide a human-readable justification when it flags an outlier.
10
+
11
+ Procedure is described in more detail in [Explainable outlier detection through decision tree conditioning](http://arxiv.org/abs/2001.00636).
12
+
13
+ # Example outputs
14
+
15
+ Example outliers from [hypothyroid dataset](http://archive.ics.uci.edu/ml/datasets/thyroid+disease):
16
+ ```
17
+ row [1137] - suspicious column: [age] - suspicious value: [75.000]
18
+ distribution: 95.122% <= 42.000 - [mean: 31.462] - [sd: 5.281] - [norm. obs: 39]
19
+ given:
20
+ [pregnant] = [t]
21
+
22
+
23
+ row [2229] - suspicious column: [T3] - suspicious vale: [10.600]
24
+ distribution: 99.951% <= 7.100 - [mean: 1.984] - [sd: 0.750] - [norm. obs: 2050]
25
+ given:
26
+ [query hyperthyroid] = [f]
27
+ ```
28
+ (i.e. it's saying that it's abnormal to be pregnant at the age of 75, or to not be classified as hyperthyroidal when having very high thyroid hormone levels)
29
+ (this dataset is also bundled into the R package - e.g. `data(hypothyroid)`)
30
+
31
+
32
+ Example outlier from [Titanic dataset](https://www.kaggle.com/c/titanic):
33
+ ```
34
+ row [885] - suspicious column: [Fare] - suspicious value: [29.125]
35
+ distribution: 97.849% <= 15.500 - [mean: 7.887] - [sd: 1.173] - [norm. obs: 91]
36
+ given:
37
+ [Pclass] = [3]
38
+ [SibSp] = [0]
39
+ [Embarked] = [Q]
40
+ ```
41
+ (i.e. it's saying that the this person paid too much for the kind of accomodation he had)
42
+
43
+ _Note that it can also produce other types of conditions such as 'between' (for numeric intervals) or 'in' (for categorical subsets)_
44
+
45
+ # Installation
46
+
47
+ * For R:
48
+ ```r
49
+ install.packages("outliertree")
50
+ ```
51
+
52
+
53
+ * For Python:
54
+ ```
55
+ pip install outliertree
56
+ ```
57
+ (Package has only been tested in Python 3)
58
+
59
+ **Note for macOS users:** on macOS, the Python version of this package will compile **without** multi-threading capabilities. This is due to default apple's redistribution of `clang` not providing OpenMP modules, and aliasing it to `gcc` which causes confusions in build scripts. If you have a non-apple version of `clang` with the OpenMP modules, or if you have `gcc` installed, you can compile this package with multi-threading enabled by setting up an environment variable `ENABLE_OMP=1`:
60
+ ```
61
+ export ENABLE_OMP=1
62
+ pip install outliertree
63
+ ```
64
+ (Alternatively, can also pass argument `enable-omp` to the `setup.py` file: `python setup.py install enable-omp`)
65
+
66
+
67
+ * For C++: package doesn't have a build system, nor a `main` function that can produce an executable, but can be built as a shared object and wrapped into other languages with any C++11-compliant compiler (`std=c++11` in most compilers, `/std:c++14` in MSVC). For parallelization, needs OpenMP linkage (`-fopenmp` in most compilers, `/openmp` in MSVC). Package should *not* be built with optimization higher than `O3` (i.e. don't use `-Ofast`). Needs linkage to the `math` library, which should be enabled by default in most C++ compilers, but otherwise would require `-lm` argument. No external dependencies are required.
68
+
69
+
70
+ # Sample usage
71
+
72
+ * For R:
73
+ ```r
74
+ library(outliertree)
75
+
76
+ ### random data frame with an obvious outlier
77
+ nrows = 100
78
+ set.seed(1)
79
+ df = data.frame(
80
+ numeric_col1 = c(rnorm(nrows - 1), 1e6),
81
+ numeric_col2 = rgamma(nrows, 1),
82
+ categ_col = sample(c('categA', 'categB', 'categC'), size = nrows, replace = TRUE)
83
+ )
84
+
85
+ ### test data frame with another obvious outlier
86
+ nrows_test = 50
87
+ df_test = data.frame(
88
+ numeric_col1 = rnorm(nrows_test),
89
+ numeric_col2 = c(-1e6, rgamma(nrows_test - 1, 1)),
90
+ categ_col = sample(c('categA', 'categB', 'categC'), size = nrows_test, replace = TRUE)
91
+ )
92
+
93
+ ### fit model
94
+ outliers_model = outliertree::outlier.tree(df, outliers_print = 10, save_outliers = TRUE)
95
+
96
+ ### find outliers in new data
97
+ new_outliers = predict(outliers_model, df_test, outliers_print = 10, return_outliers = TRUE)
98
+
99
+ ### print outliers in readable format
100
+ summary(new_outliers)
101
+ ```
102
+ (see documentation for more examples)
103
+
104
+ Example [RMarkdown](http://htmlpreview.github.io/?https://github.com/david-cortes/outliertree/blob/master/example/titanic_outliertree_r.html) using the Titanic dataset.
105
+
106
+
107
+ * For Python:
108
+ ```python
109
+ import numpy as np, pandas as pd
110
+ from outliertree import OutlierTree
111
+
112
+ ### random data frame with an obvious outlier
113
+ nrows = 100
114
+ np.random.seed(1)
115
+ df = pd.DataFrame({
116
+ "numeric_col1" : np.r_[np.random.normal(size = nrows - 1), np.array([float(1e6)])],
117
+ "numeric_col2" : np.random.gamma(1, 1, size = nrows),
118
+ "categ_col" : np.random.choice(['categA', 'categB', 'categC'], size = nrows)
119
+ })
120
+
121
+ ### test data frame with another obvious outlier
122
+ df_test = pd.DataFrame({
123
+ "numeric_col1" : np.random.normal(size = nrows),
124
+ "numeric_col2" : np.r_[np.array([float(-1e6)]), np.random.gamma(1, 1, size = nrows - 1)],
125
+ "categ_col" : np.random.choice(['categA', 'categB', 'categC'], size = nrows)
126
+ })
127
+
128
+ ### fit model
129
+ outliers_model = OutlierTree()
130
+ outliers_df = outliers_model.fit(df, outliers_print = 10, return_outliers = True)
131
+
132
+ ### find outliers in new data
133
+ new_outliers = outliers_model.predict(df_test)
134
+
135
+ ### print outliers in readable format
136
+ outliers_model.print_outliers(new_outliers)
137
+ ```
138
+
139
+ Example [IPython notebook](http://nbviewer.ipython.org/github/david-cortes/outliertree/blob/master/example/titanic_outliertree_python.ipynb) using the Titanic dataset.
140
+
141
+ * For C++: see functions `fit_outliers_models` and `find_new_outliers` in header `outlier_tree.hpp`.
142
+
143
+ # Documentation
144
+
145
+ * For R : documentation is built-in in the package (e.g. `help(outliertree::outlier.tree)`) - PDF can be downloaded in [CRAN](https://cran.r-project.org/web/packages/outliertree/index.html).
146
+
147
+ * For Python: documentation is available at [ReadTheDocs](http://outliertree.readthedocs.io/en/latest/) (and it's also built-in in the package as docstrings, e.g. `help(outliertree.OutlierTree.fit)`).
148
+
149
+ * For C++: documentation is available in the source files (not in the header).
150
+
151
+ # References
152
+
153
+ * Cortes, David. "Explainable outlier detection through decision tree conditioning." arXiv preprint arXiv:2001.00636 (2020).
154
+ * [GritBot software](https://www.rulequest.com/gritbot-info.html) .
155
+
data/vendor/outliertree/src/Makevars ADDED
@@ -0,0 +1,3 @@
1
+ PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
2
+ PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS)
3
+ CXX_STD = CXX11
data/vendor/outliertree/src/RcppExports.cpp ADDED
@@ -0,0 +1,123 @@
1
+ // Generated by using Rcpp::compileAttributes() -> do not edit by hand
2
+ // Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
3
+
4
+ #include <Rcpp.h>
5
+
6
+ using namespace Rcpp;
7
+
8
+ // deserialize_OutlierTree
9
+ SEXP deserialize_OutlierTree(Rcpp::RawVector src);
10
+ RcppExport SEXP _outliertree_deserialize_OutlierTree(SEXP srcSEXP) {
11
+ BEGIN_RCPP
12
+ Rcpp::RObject rcpp_result_gen;
13
+ Rcpp::RNGScope rcpp_rngScope_gen;
14
+ Rcpp::traits::input_parameter< Rcpp::RawVector >::type src(srcSEXP);
15
+ rcpp_result_gen = Rcpp::wrap(deserialize_OutlierTree(src));
16
+ return rcpp_result_gen;
17
+ END_RCPP
18
+ }
19
+ // check_null_ptr_model
20
+ Rcpp::LogicalVector check_null_ptr_model(SEXP ptr_model);
21
+ RcppExport SEXP _outliertree_check_null_ptr_model(SEXP ptr_modelSEXP) {
22
+ BEGIN_RCPP
23
+ Rcpp::RObject rcpp_result_gen;
24
+ Rcpp::RNGScope rcpp_rngScope_gen;
25
+ Rcpp::traits::input_parameter< SEXP >::type ptr_model(ptr_modelSEXP);
26
+ rcpp_result_gen = Rcpp::wrap(check_null_ptr_model(ptr_model));
27
+ return rcpp_result_gen;
28
+ END_RCPP
29
+ }
30
+ // fit_OutlierTree
31
+ Rcpp::List fit_OutlierTree(Rcpp::NumericVector arr_num, size_t ncols_numeric, Rcpp::IntegerVector arr_cat, size_t ncols_categ, Rcpp::IntegerVector ncat, Rcpp::IntegerVector arr_ord, size_t ncols_ord, Rcpp::IntegerVector ncat_ord, size_t nrows, Rcpp::LogicalVector cols_ignore_r, int nthreads, bool categ_as_bin, bool ord_as_bin, bool cat_bruteforce_subset, bool categ_from_maj, bool take_mid, size_t max_depth, double max_perc_outliers, size_t min_size_numeric, size_t min_size_categ, double min_gain, bool follow_all, bool gain_as_pct, double z_norm, double z_outlier, bool return_outliers, Rcpp::ListOf<Rcpp::StringVector> cat_levels, Rcpp::ListOf<Rcpp::StringVector> ord_levels, Rcpp::StringVector colnames_num, Rcpp::StringVector colnames_cat, Rcpp::StringVector colnames_ord, Rcpp::NumericVector min_date, Rcpp::NumericVector min_ts);
32
+ RcppExport SEXP _outliertree_fit_OutlierTree(SEXP arr_numSEXP, SEXP ncols_numericSEXP, SEXP arr_catSEXP, SEXP ncols_categSEXP, SEXP ncatSEXP, SEXP arr_ordSEXP, SEXP ncols_ordSEXP, SEXP ncat_ordSEXP, SEXP nrowsSEXP, SEXP cols_ignore_rSEXP, SEXP nthreadsSEXP, SEXP categ_as_binSEXP, SEXP ord_as_binSEXP, SEXP cat_bruteforce_subsetSEXP, SEXP categ_from_majSEXP, SEXP take_midSEXP, SEXP max_depthSEXP, SEXP max_perc_outliersSEXP, SEXP min_size_numericSEXP, SEXP min_size_categSEXP, SEXP min_gainSEXP, SEXP follow_allSEXP, SEXP gain_as_pctSEXP, SEXP z_normSEXP, SEXP z_outlierSEXP, SEXP return_outliersSEXP, SEXP cat_levelsSEXP, SEXP ord_levelsSEXP, SEXP colnames_numSEXP, SEXP colnames_catSEXP, SEXP colnames_ordSEXP, SEXP min_dateSEXP, SEXP min_tsSEXP) {
33
+ BEGIN_RCPP
34
+ Rcpp::RObject rcpp_result_gen;
35
+ Rcpp::RNGScope rcpp_rngScope_gen;
36
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type arr_num(arr_numSEXP);
37
+ Rcpp::traits::input_parameter< size_t >::type ncols_numeric(ncols_numericSEXP);
38
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type arr_cat(arr_catSEXP);
39
+ Rcpp::traits::input_parameter< size_t >::type ncols_categ(ncols_categSEXP);
40
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type ncat(ncatSEXP);
41
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type arr_ord(arr_ordSEXP);
42
+ Rcpp::traits::input_parameter< size_t >::type ncols_ord(ncols_ordSEXP);
43
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type ncat_ord(ncat_ordSEXP);
44
+ Rcpp::traits::input_parameter< size_t >::type nrows(nrowsSEXP);
45
+ Rcpp::traits::input_parameter< Rcpp::LogicalVector >::type cols_ignore_r(cols_ignore_rSEXP);
46
+ Rcpp::traits::input_parameter< int >::type nthreads(nthreadsSEXP);
47
+ Rcpp::traits::input_parameter< bool >::type categ_as_bin(categ_as_binSEXP);
48
+ Rcpp::traits::input_parameter< bool >::type ord_as_bin(ord_as_binSEXP);
49
+ Rcpp::traits::input_parameter< bool >::type cat_bruteforce_subset(cat_bruteforce_subsetSEXP);
50
+ Rcpp::traits::input_parameter< bool >::type categ_from_maj(categ_from_majSEXP);
51
+ Rcpp::traits::input_parameter< bool >::type take_mid(take_midSEXP);
52
+ Rcpp::traits::input_parameter< size_t >::type max_depth(max_depthSEXP);
53
+ Rcpp::traits::input_parameter< double >::type max_perc_outliers(max_perc_outliersSEXP);
54
+ Rcpp::traits::input_parameter< size_t >::type min_size_numeric(min_size_numericSEXP);
55
+ Rcpp::traits::input_parameter< size_t >::type min_size_categ(min_size_categSEXP);
56
+ Rcpp::traits::input_parameter< double >::type min_gain(min_gainSEXP);
57
+ Rcpp::traits::input_parameter< bool >::type follow_all(follow_allSEXP);
58
+ Rcpp::traits::input_parameter< bool >::type gain_as_pct(gain_as_pctSEXP);
59
+ Rcpp::traits::input_parameter< double >::type z_norm(z_normSEXP);
60
+ Rcpp::traits::input_parameter< double >::type z_outlier(z_outlierSEXP);
61
+ Rcpp::traits::input_parameter< bool >::type return_outliers(return_outliersSEXP);
62
+ Rcpp::traits::input_parameter< Rcpp::ListOf<Rcpp::StringVector> >::type cat_levels(cat_levelsSEXP);
63
+ Rcpp::traits::input_parameter< Rcpp::ListOf<Rcpp::StringVector> >::type ord_levels(ord_levelsSEXP);
64
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_num(colnames_numSEXP);
65
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_cat(colnames_catSEXP);
66
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_ord(colnames_ordSEXP);
67
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type min_date(min_dateSEXP);
68
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type min_ts(min_tsSEXP);
69
+ rcpp_result_gen = Rcpp::wrap(fit_OutlierTree(arr_num, ncols_numeric, arr_cat, ncols_categ, ncat, arr_ord, ncols_ord, ncat_ord, nrows, cols_ignore_r, nthreads, categ_as_bin, ord_as_bin, cat_bruteforce_subset, categ_from_maj, take_mid, max_depth, max_perc_outliers, min_size_numeric, min_size_categ, min_gain, follow_all, gain_as_pct, z_norm, z_outlier, return_outliers, cat_levels, ord_levels, colnames_num, colnames_cat, colnames_ord, min_date, min_ts));
70
+ return rcpp_result_gen;
71
+ END_RCPP
72
+ }
73
+ // predict_OutlierTree
74
+ Rcpp::List predict_OutlierTree(SEXP ptr_model, size_t nrows, int nthreads, Rcpp::NumericVector arr_num, Rcpp::IntegerVector arr_cat, Rcpp::IntegerVector arr_ord, Rcpp::ListOf<Rcpp::StringVector> cat_levels, Rcpp::ListOf<Rcpp::StringVector> ord_levels, Rcpp::StringVector colnames_num, Rcpp::StringVector colnames_cat, Rcpp::StringVector colnames_ord, Rcpp::NumericVector min_date, Rcpp::NumericVector min_ts);
75
+ RcppExport SEXP _outliertree_predict_OutlierTree(SEXP ptr_modelSEXP, SEXP nrowsSEXP, SEXP nthreadsSEXP, SEXP arr_numSEXP, SEXP arr_catSEXP, SEXP arr_ordSEXP, SEXP cat_levelsSEXP, SEXP ord_levelsSEXP, SEXP colnames_numSEXP, SEXP colnames_catSEXP, SEXP colnames_ordSEXP, SEXP min_dateSEXP, SEXP min_tsSEXP) {
76
+ BEGIN_RCPP
77
+ Rcpp::RObject rcpp_result_gen;
78
+ Rcpp::RNGScope rcpp_rngScope_gen;
79
+ Rcpp::traits::input_parameter< SEXP >::type ptr_model(ptr_modelSEXP);
80
+ Rcpp::traits::input_parameter< size_t >::type nrows(nrowsSEXP);
81
+ Rcpp::traits::input_parameter< int >::type nthreads(nthreadsSEXP);
82
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type arr_num(arr_numSEXP);
83
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type arr_cat(arr_catSEXP);
84
+ Rcpp::traits::input_parameter< Rcpp::IntegerVector >::type arr_ord(arr_ordSEXP);
85
+ Rcpp::traits::input_parameter< Rcpp::ListOf<Rcpp::StringVector> >::type cat_levels(cat_levelsSEXP);
86
+ Rcpp::traits::input_parameter< Rcpp::ListOf<Rcpp::StringVector> >::type ord_levels(ord_levelsSEXP);
87
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_num(colnames_numSEXP);
88
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_cat(colnames_catSEXP);
89
+ Rcpp::traits::input_parameter< Rcpp::StringVector >::type colnames_ord(colnames_ordSEXP);
90
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type min_date(min_dateSEXP);
91
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type min_ts(min_tsSEXP);
92
+ rcpp_result_gen = Rcpp::wrap(predict_OutlierTree(ptr_model, nrows, nthreads, arr_num, arr_cat, arr_ord, cat_levels, ord_levels, colnames_num, colnames_cat, colnames_ord, min_date, min_ts));
93
+ return rcpp_result_gen;
94
+ END_RCPP
95
+ }
96
+ // check_few_values
97
+ Rcpp::LogicalVector check_few_values(Rcpp::NumericVector arr_num, size_t nrows, size_t ncols, int nthreads);
98
+ RcppExport SEXP _outliertree_check_few_values(SEXP arr_numSEXP, SEXP nrowsSEXP, SEXP ncolsSEXP, SEXP nthreadsSEXP) {
99
+ BEGIN_RCPP
100
+ Rcpp::RObject rcpp_result_gen;
101
+ Rcpp::RNGScope rcpp_rngScope_gen;
102
+ Rcpp::traits::input_parameter< Rcpp::NumericVector >::type arr_num(arr_numSEXP);
103
+ Rcpp::traits::input_parameter< size_t >::type nrows(nrowsSEXP);
104
+ Rcpp::traits::input_parameter< size_t >::type ncols(ncolsSEXP);
105
+ Rcpp::traits::input_parameter< int >::type nthreads(nthreadsSEXP);
106
+ rcpp_result_gen = Rcpp::wrap(check_few_values(arr_num, nrows, ncols, nthreads));
107
+ return rcpp_result_gen;
108
+ END_RCPP
109
+ }
110
+
111
+ static const R_CallMethodDef CallEntries[] = {
112
+ {"_outliertree_deserialize_OutlierTree", (DL_FUNC) &_outliertree_deserialize_OutlierTree, 1},
113
+ {"_outliertree_check_null_ptr_model", (DL_FUNC) &_outliertree_check_null_ptr_model, 1},
114
+ {"_outliertree_fit_OutlierTree", (DL_FUNC) &_outliertree_fit_OutlierTree, 33},
115
+ {"_outliertree_predict_OutlierTree", (DL_FUNC) &_outliertree_predict_OutlierTree, 13},
116
+ {"_outliertree_check_few_values", (DL_FUNC) &_outliertree_check_few_values, 4},
117
+ {NULL, NULL, 0}
118
+ };
119
+
120
+ RcppExport void R_init_outliertree(DllInfo *dll) {
121
+ R_registerRoutines(dll, NULL, CallEntries, NULL, NULL);
122
+ R_useDynamicSymbols(dll, FALSE);
123
+ }
data/vendor/outliertree/src/Rwrapper.cpp ADDED
@@ -0,0 +1,1225 @@
1
+ #include <Rcpp.h>
2
+ // [[Rcpp::plugins(cpp11)]]
3
+
4
+ /* This is to serialize the model objects */
5
+ // [[Rcpp::depends(Rcereal)]]
6
+ #include <cereal/archives/binary.hpp>
7
+ #include <cereal/types/vector.hpp>
8
+ #include <sstream>
9
+ #include <string>
10
+
11
+ /* This is the package's header */
12
+ #include "outlier_tree.hpp"
13
+
14
+ /* for model serialization and re-usage in R */
15
+ /* https://stackoverflow.com/questions/18474292/how-to-handle-c-internal-data-structure-in-r-in-order-to-allow-save-load */
16
+ /* this extra comment below the link is a workaround for Rcpp issue 675 in GitHub, do not remove it */
17
+ #include <Rinternals.h>
18
+ Rcpp::RawVector serialize_OutlierTree(ModelOutputs *model_outputs)
19
+ {
20
+ std::stringstream ss;
21
+ {
22
+ cereal::BinaryOutputArchive oarchive(ss); // Create an output archive
23
+ oarchive(*model_outputs);
24
+ }
25
+ ss.seekg(0, ss.end);
26
+ Rcpp::RawVector retval(ss.tellg());
27
+ ss.seekg(0, ss.beg);
28
+ ss.read(reinterpret_cast<char*>(&retval[0]), retval.size());
29
+ return retval;
30
+ }
31
+
32
+ // [[Rcpp::export]]
33
+ SEXP deserialize_OutlierTree(Rcpp::RawVector src)
34
+ {
35
+ std::stringstream ss;
36
+ ss.write(reinterpret_cast<char*>(&src[0]), src.size());
37
+ ss.seekg(0, ss.beg);
38
+ std::unique_ptr<ModelOutputs> model_outputs = std::unique_ptr<ModelOutputs>(new ModelOutputs());
39
+ {
40
+ cereal::BinaryInputArchive iarchive(ss);
41
+ iarchive(*model_outputs);
42
+ }
43
+ return Rcpp::XPtr<ModelOutputs>(model_outputs.release(), true);
44
+ }
45
+
46
+ // [[Rcpp::export]]
47
+ Rcpp::LogicalVector check_null_ptr_model(SEXP ptr_model)
48
+ {
49
+ return Rcpp::LogicalVector(R_ExternalPtrAddr(ptr_model) == NULL);
50
+ }
51
+
52
+ double* set_R_nan_as_C_nan(double *restrict x_R, std::vector<double> &x_C, size_t n, int nthreads)
53
+ {
54
+ x_C.assign(x_R, x_R + n);
55
+ #pragma omp parallel for schedule(static) num_threads(nthreads) shared(x_R, x_C, n)
56
+ for (size_t_for i = 0; i < n; i++)
57
+ if (isnan(x_R[i]) || Rcpp::NumericVector::is_na(x_R[i]) || Rcpp::traits::is_nan<REALSXP>(x_R[i]))
58
+ x_C[i] = NAN;
59
+ return x_C.data();
60
+ }
61
+
62
+
63
+ /* for predicting outliers */
64
+ Rcpp::List describe_outliers(ModelOutputs &model_outputs,
65
+ double *arr_num,
66
+ int *arr_cat,
67
+ int *arr_ord,
68
+ Rcpp::ListOf<Rcpp::StringVector> cat_levels,
69
+ Rcpp::ListOf<Rcpp::StringVector> ord_levels,
70
+ Rcpp::StringVector colnames_num,
71
+ Rcpp::StringVector colnames_cat,
72
+ Rcpp::StringVector colnames_ord,
73
+ Rcpp::NumericVector min_date,
74
+ Rcpp::NumericVector min_ts)
75
+ {
76
+ size_t nrows = model_outputs.outlier_scores_final.size();
77
+ size_t ncols_num = model_outputs.ncols_numeric;
78
+ size_t ncols_cat = model_outputs.ncols_categ;
79
+ size_t ncols_num_num = model_outputs.ncols_numeric - min_date.size() - min_ts.size();
80
+ size_t ncols_date = min_date.size();
81
+ size_t ncols_cat_cat = cat_levels.size();
82
+ Rcpp::List outp;
83
+
84
+ Rcpp::LogicalVector has_na_col = Rcpp::LogicalVector(nrows, NA_LOGICAL);
85
+ Rcpp::IntegerVector tree_depth = Rcpp::IntegerVector(nrows, NA_INTEGER);
86
+ Rcpp::NumericVector outlier_score = Rcpp::NumericVector(nrows, NA_REAL);
87
+ Rcpp::ListOf<Rcpp::List> outlier_val = Rcpp::ListOf<Rcpp::List>(nrows);
88
+ Rcpp::ListOf<Rcpp::List> lst_stats = Rcpp::ListOf<Rcpp::List>(nrows);
89
+ Rcpp::ListOf<Rcpp::List> lst_cond = Rcpp::ListOf<Rcpp::List>(nrows);
90
+
91
+
92
+ size_t outl_col;
93
+ size_t outl_clust;
94
+ size_t curr_tree;
95
+ size_t parent_tree;
96
+ Rcpp::LogicalVector tmp_bool;
97
+
98
+ for (size_t row = 0; row < nrows; row++) {
99
+ if (model_outputs.outlier_scores_final[row] < 1) {
100
+
101
+ outl_col = model_outputs.outlier_columns_final[row];
102
+ outl_clust = model_outputs.outlier_clusters_final[row];
103
+
104
+ /* metrics of outlierness - used to rank when choosing which to print */
105
+ outlier_score[row] = model_outputs.outlier_scores_final[row];
106
+ tree_depth[row] = (int)model_outputs.outlier_depth_final[row];
107
+ has_na_col[row] = model_outputs.all_clusters[outl_col][outl_clust].has_NA_branch;
108
+
109
+ /* first determine outlier column and suspected value */
110
+ if (outl_col < ncols_num) {
111
+ if (outl_col < ncols_num_num) {
112
+ outlier_val[row] = Rcpp::List::create(
113
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_num[outl_col]),
114
+ Rcpp::_["value"] = Rcpp::wrap(arr_num[row + outl_col * nrows]),
115
+ Rcpp::_["decimals"] = Rcpp::wrap(model_outputs.outlier_decimals_distr[row])
116
+ );
117
+ } else if (outl_col < (ncols_num_num + ncols_date)) {
118
+ outlier_val[row] = Rcpp::List::create(
119
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_num[outl_col]),
120
+ Rcpp::_["value"] = Rcpp::Date(arr_num[row + outl_col * nrows] - 1 + min_date[outl_col - ncols_num_num])
121
+ );
122
+ } else {
123
+ outlier_val[row] = Rcpp::List::create(
124
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_num[outl_col]),
125
+ Rcpp::_["value"] = Rcpp::Datetime(arr_num[row + outl_col * nrows] - 1 + min_ts[outl_col - ncols_num_num - ncols_date])
126
+ );
127
+ }
128
+ } else if (outl_col < (ncols_num + ncols_cat)) {
129
+ if (outl_col < (ncols_num + ncols_cat_cat)) {
130
+ outlier_val[row] = Rcpp::List::create(
131
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_cat[outl_col - ncols_num]),
132
+ Rcpp::_["value"] = Rcpp::CharacterVector(1, cat_levels[outl_col - ncols_num]
133
+ [arr_cat[row + (outl_col - ncols_num) * nrows]])
134
+ );
135
+ } else {
136
+ outlier_val[row] = Rcpp::List::create(
137
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_cat[outl_col - ncols_num]),
138
+ Rcpp::_["value"] = Rcpp::wrap((bool)arr_cat[row + (outl_col - ncols_num) * nrows])
139
+ );
140
+ }
141
+ } else {
142
+ outlier_val[row] = Rcpp::List::create(
143
+ Rcpp::_["column"] = Rcpp::CharacterVector(1, colnames_ord[outl_col - ncols_num - ncols_cat]),
144
+ Rcpp::_["value"] = Rcpp::CharacterVector(1, ord_levels[outl_col - ncols_num - ncols_cat]
145
+ [arr_ord[row + (outl_col - ncols_num - ncols_cat) * nrows]])
146
+ );
147
+ }
148
+
149
+
150
+ /* info about the normal observations in the cluster */
151
+ if (outl_col < ncols_num) {
152
+ if (outl_col < ncols_num_num) {
153
+ if (arr_num[row + outl_col * nrows] >= model_outputs.all_clusters[outl_col][outl_clust].upper_lim) {
154
+ lst_stats[row] = Rcpp::List::create(
155
+ Rcpp::_["upper_thr"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_lim_high),
156
+ Rcpp::_["pct_below"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_below),
157
+ Rcpp::_["mean"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_mean),
158
+ Rcpp::_["sd"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_sd),
159
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
160
+ );
161
+ } else {
162
+ lst_stats[row] = Rcpp::List::create(
163
+ Rcpp::_["lower_thr"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_lim_low),
164
+ Rcpp::_["pct_above"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_above),
165
+ Rcpp::_["mean"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_mean),
166
+ Rcpp::_["sd"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].display_sd),
167
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
168
+ );
169
+ }
170
+ } else if (outl_col < (ncols_num_num + ncols_date)) {
171
+ if (arr_num[row + outl_col * nrows] >= model_outputs.all_clusters[outl_col][outl_clust].upper_lim) {
172
+ lst_stats[row] = Rcpp::List::create(
173
+ Rcpp::_["upper_thr"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].display_lim_high
174
+ - 1 + min_date[outl_col - ncols_num_num]),
175
+ Rcpp::_["pct_below"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_below),
176
+ Rcpp::_["mean"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].display_mean - 1 + min_date[outl_col - ncols_num_num]),
177
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
178
+ );
179
+ } else {
180
+ lst_stats[row] = Rcpp::List::create(
181
+ Rcpp::_["lower_thr"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].display_lim_low
182
+ - 1 + min_date[outl_col - ncols_num_num]),
183
+ Rcpp::_["pct_above"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_above),
184
+ Rcpp::_["mean"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].display_mean - 1 + min_date[outl_col - ncols_num_num]),
185
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
186
+ );
187
+ }
188
+ } else {
189
+ if (arr_num[row + outl_col * nrows] >= model_outputs.all_clusters[outl_col][outl_clust].upper_lim) {
190
+ lst_stats[row] = Rcpp::List::create(
191
+ Rcpp::_["upper_thr"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].display_lim_high
192
+ - 1 + min_ts[outl_col - ncols_num_num - ncols_date]),
193
+ Rcpp::_["pct_below"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_below),
194
+ Rcpp::_["mean"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].display_mean
195
+ - 1 + min_ts[outl_col - ncols_num_num - ncols_date]),
196
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
197
+ );
198
+ } else {
199
+ lst_stats[row] = Rcpp::List::create(
200
+ Rcpp::_["lower_thr"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].display_lim_low
201
+ - 1 + min_ts[outl_col - ncols_num_num - ncols_date]),
202
+ Rcpp::_["pct_above"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_above),
203
+ Rcpp::_["mean"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].display_mean
204
+ - 1 + min_ts[outl_col - ncols_num_num - ncols_date]),
205
+ Rcpp::_["n_obs"] = Rcpp::wrap((int)model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
206
+ );
207
+ }
208
+ }
209
+ } else if (outl_col < (ncols_num + ncols_cat)) {
210
+ if (outl_col < (ncols_num + ncols_cat_cat)) {
211
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_clusters[outl_col][outl_clust].subset_common.size(), false);
212
+ for (size_t cat = 0; cat < tmp_bool.size(); cat++) {
213
+ if (model_outputs.all_clusters[outl_col][outl_clust].subset_common[cat] == 0) {
214
+ tmp_bool[cat] = true;
215
+ }
216
+ }
217
+ if (model_outputs.all_clusters[outl_col][outl_clust].split_type != Root) {
218
+ if (model_outputs.all_clusters[outl_col][outl_clust].categ_maj < 0) {
219
+ lst_stats[row] = Rcpp::List::create(
220
+ Rcpp::_["categs_common"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[outl_col - ncols_num][tmp_bool]),
221
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
222
+ Rcpp::_["pct_next_most_comm"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_next_most_comm),
223
+ Rcpp::_["prior_prob"] = Rcpp::wrap(model_outputs.prop_categ[model_outputs.start_ix_cat_counts[outl_col - ncols_num] +
224
+ arr_cat[row + (outl_col - ncols_num) * nrows]]),
225
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
226
+ );
227
+ } else {
228
+ lst_stats[row] = Rcpp::List::create(
229
+ Rcpp::_["categ_maj"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[outl_col - ncols_num][
230
+ model_outputs.all_clusters[outl_col][outl_clust].categ_maj
231
+ ]),
232
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
233
+ Rcpp::_["prior_prob"] = Rcpp::wrap(model_outputs.prop_categ[model_outputs.start_ix_cat_counts[outl_col - ncols_num] +
234
+ arr_cat[row + (outl_col - ncols_num) * nrows]]),
235
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
236
+ );
237
+ }
238
+ } else {
239
+ lst_stats[row] = Rcpp::List::create(
240
+ Rcpp::_["categs_common"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[outl_col - ncols_num][tmp_bool]),
241
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
242
+ Rcpp::_["pct_next_most_comm"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_next_most_comm),
243
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
244
+ );
245
+ }
246
+ } else {
247
+ lst_stats[row] = Rcpp::List::create(
248
+ Rcpp::_["pct_other"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
249
+ Rcpp::_["prior_prob"] = Rcpp::wrap(model_outputs.prop_categ[model_outputs.start_ix_cat_counts[outl_col - ncols_num] +
250
+ arr_cat[row + (outl_col - ncols_num) * nrows]]),
251
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
252
+ );
253
+ }
254
+ } else {
255
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_clusters[outl_col][outl_clust].subset_common.size(), false);
256
+ for (size_t cat = 0; cat < tmp_bool.size(); cat++) {
257
+ if (model_outputs.all_clusters[outl_col][outl_clust].subset_common[cat] == 0) {
258
+ tmp_bool[cat] = true;
259
+ }
260
+ }
261
+ if (model_outputs.all_clusters[outl_col][outl_clust].split_type != Root) {
262
+ if (model_outputs.all_clusters[outl_col][outl_clust].categ_maj < 0) {
263
+ lst_stats[row] = Rcpp::List::create(
264
+ Rcpp::_["categs_common"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[outl_col - ncols_num - ncols_cat][tmp_bool]),
265
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
266
+ Rcpp::_["pct_next_most_comm"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_next_most_comm),
267
+ Rcpp::_["prior_prob"] = Rcpp::wrap(model_outputs.prop_categ[model_outputs.start_ix_cat_counts[outl_col - ncols_num] +
268
+ arr_ord[row + (outl_col - ncols_num - ncols_cat) * nrows]]),
269
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
270
+ );
271
+ } else {
272
+ lst_stats[row] = Rcpp::List::create(
273
+ Rcpp::_["categ_maj"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[outl_col - ncols_num - ncols_cat][
274
+ model_outputs.all_clusters[outl_col][outl_clust].categ_maj
275
+ ]),
276
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
277
+ Rcpp::_["prior_prob"] = Rcpp::wrap(model_outputs.prop_categ[model_outputs.start_ix_cat_counts[outl_col - ncols_num] +
278
+ arr_ord[row + (outl_col - ncols_num - ncols_cat) * nrows]]),
279
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
280
+ );
281
+ }
282
+ } else {
283
+ lst_stats[row] = Rcpp::List::create(
284
+ Rcpp::_["categs_common"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[outl_col - ncols_num - ncols_cat][tmp_bool]),
285
+ Rcpp::_["pct_common"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_in_subset),
286
+ Rcpp::_["pct_next_most_comm"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].perc_next_most_comm),
287
+ Rcpp::_["n_obs"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].cluster_size)
288
+ );
289
+ }
290
+ }
291
+
292
+
293
+ /* then determine conditions from the cluster */
294
+ Rcpp::List cond_clust;
295
+ if (model_outputs.all_clusters[outl_col][outl_clust].column_type != NoType) {
296
+
297
+ /* add the column name and actual value for the row */
298
+ switch(model_outputs.all_clusters[outl_col][outl_clust].column_type) {
299
+ case Numeric:
300
+ {
301
+ cond_clust["column"] = Rcpp::CharacterVector(1, colnames_num[model_outputs.all_clusters[outl_col][outl_clust].col_num]);
302
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_num_num) {
303
+ cond_clust["value_this"] = Rcpp::wrap(arr_num[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]);
304
+ if (model_outputs.all_clusters[outl_col][outl_clust].split_type != IsNa)
305
+ cond_clust["decimals"] = Rcpp::wrap(model_outputs.min_decimals_col[model_outputs.all_clusters[outl_col][outl_clust].col_num]);
306
+ } else if (model_outputs.all_clusters[outl_col][outl_clust].col_num < (ncols_num_num + ncols_date)) {
307
+ cond_clust["value_this"] = Rcpp::Date(arr_num[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]
308
+ - 1 + min_date[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num]);
309
+ } else {
310
+ cond_clust["value_this"] = Rcpp::Datetime(arr_num[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]
311
+ - 1 + min_ts[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num - ncols_date]);
312
+ }
313
+ break;
314
+ }
315
+
316
+ case Categorical:
317
+ {
318
+ cond_clust["column"] = Rcpp::CharacterVector(1, colnames_cat[model_outputs.all_clusters[outl_col][outl_clust].col_num]);
319
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_cat_cat) {
320
+ if (arr_cat[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows] >= 0) {
321
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
322
+ [arr_cat[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]]);
323
+ } else {
324
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
325
+ }
326
+ } else {
327
+
328
+ if (arr_cat[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows] >= 0) {
329
+ cond_clust["value_this"] = Rcpp::wrap((bool)arr_cat[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]);
330
+ } else {
331
+ cond_clust["value_this"] = Rcpp::LogicalVector(1, NA_LOGICAL);
332
+ }
333
+ }
334
+ break;
335
+ }
336
+
337
+ case Ordinal:
338
+ {
339
+ cond_clust["column"] = Rcpp::CharacterVector(1, colnames_ord[model_outputs.all_clusters[outl_col][outl_clust].col_num]);
340
+ if (arr_ord[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows] >= 0) {
341
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
342
+ [arr_ord[row + model_outputs.all_clusters[outl_col][outl_clust].col_num * nrows]]);
343
+ } else {
344
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
345
+ }
346
+ break;
347
+ }
348
+ }
349
+
350
+ /* add the comparison point */
351
+ switch(model_outputs.all_clusters[outl_col][outl_clust].split_type) {
352
+
353
+ case IsNa:
354
+ {
355
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
356
+ switch(model_outputs.all_clusters[outl_col][outl_clust].column_type) {
357
+ case Numeric:
358
+ {
359
+ /* http://lists.r-forge.r-project.org/pipermail/rcpp-devel/2012-October/004379.html */
360
+ /* this comment below will prevent bug with Rcpp comments having forward slashes */
361
+ cond_clust["value_comp"] = Rcpp::wrap(NA_REAL);
362
+ break;
363
+ }
364
+
365
+ case Categorical:
366
+ {
367
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_cat_cat) {
368
+ cond_clust["value_comp"] = Rcpp::wrap(NA_STRING);
369
+ } else {
370
+ cond_clust["value_comp"] = Rcpp::LogicalVector(1, NA_LOGICAL);
371
+ }
372
+ break;
373
+ }
374
+
375
+ case Ordinal:
376
+ {
377
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
378
+ break;
379
+ }
380
+ }
381
+ break;
382
+ }
383
+
384
+ case LessOrEqual:
385
+ {
386
+ if (model_outputs.all_clusters[outl_col][outl_clust].column_type == Numeric) {
387
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_num_num) {
388
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
389
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].split_point);
390
+ } else if (model_outputs.all_clusters[outl_col][outl_clust].col_num < (ncols_num_num + ncols_date)) {
391
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
392
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].split_point
393
+ - 1 + min_date[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num]);
394
+ } else {
395
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
396
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].split_point
397
+ - 1 + min_ts[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num - ncols_date]);
398
+ }
399
+ } else {
400
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num].size(), false);
401
+ for (int cat = 0; cat <= model_outputs.all_clusters[outl_col][outl_clust].split_lev; cat++) tmp_bool[cat] = true;
402
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
403
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num][tmp_bool]);
404
+ }
405
+ break;
406
+ }
407
+
408
+ case Greater:
409
+ {
410
+ if (model_outputs.all_clusters[outl_col][outl_clust].column_type == Numeric) {
411
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_num_num) {
412
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
413
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_clusters[outl_col][outl_clust].split_point);
414
+ } else if (model_outputs.all_clusters[outl_col][outl_clust].col_num < (ncols_num_num + ncols_date)) {
415
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
416
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_clusters[outl_col][outl_clust].split_point
417
+ - 1 + min_date[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num]);
418
+ } else {
419
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
420
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_clusters[outl_col][outl_clust].split_point
421
+ - 1 + min_ts[model_outputs.all_clusters[outl_col][outl_clust].col_num - ncols_num_num - ncols_date]);
422
+ }
423
+ } else {
424
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num].size(), true);
425
+ for (int cat = 0; cat <= model_outputs.all_clusters[outl_col][outl_clust].split_lev; cat++) tmp_bool[cat] = false;
426
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
427
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num][tmp_bool]);
428
+ }
429
+ break;
430
+ }
431
+
432
+ case InSubset:
433
+ {
434
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_clusters[outl_col][outl_clust].split_subset.size(), false);
435
+ for (size_t cat = 0; cat < model_outputs.all_clusters[outl_col][outl_clust].split_subset.size(); cat++) {
436
+ if (model_outputs.all_clusters[outl_col][outl_clust].split_subset[cat] > 0) {
437
+ tmp_bool[cat] = true;
438
+ }
439
+ }
440
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
441
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num][tmp_bool]);
442
+ break;
443
+ }
444
+
445
+ case NotInSubset:
446
+ {
447
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_clusters[outl_col][outl_clust].split_subset.size(), false);
448
+ for (size_t cat = 0; cat < model_outputs.all_clusters[outl_col][outl_clust].split_subset.size(); cat++) {
449
+ if (model_outputs.all_clusters[outl_col][outl_clust].split_subset[cat] == 0) {
450
+ tmp_bool[cat] = true;
451
+ }
452
+ }
453
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
454
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num][tmp_bool]);
455
+ break;
456
+ }
457
+
458
+ case Equal:
459
+ {
460
+ if (model_outputs.all_clusters[outl_col][outl_clust].column_type == Categorical) {
461
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_cat_cat) {
462
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
463
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
464
+ [model_outputs.all_clusters[outl_col][outl_clust].split_lev]);
465
+ } else {
466
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
467
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_clusters[outl_col][outl_clust].split_lev);
468
+ }
469
+ } else {
470
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
471
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
472
+ [model_outputs.all_clusters[outl_col][outl_clust].split_lev]);
473
+ }
474
+ break;
475
+ }
476
+
477
+ case NotEqual:
478
+ {
479
+ if (model_outputs.all_clusters[outl_col][outl_clust].column_type == Categorical) {
480
+ if (model_outputs.all_clusters[outl_col][outl_clust].col_num < ncols_cat_cat) {
481
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
482
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
483
+ [model_outputs.all_clusters[outl_col][outl_clust].split_lev]);
484
+ } else {
485
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
486
+ cond_clust["value_comp"] = Rcpp::wrap(!((bool)model_outputs.all_clusters[outl_col][outl_clust].split_lev));
487
+ }
488
+ } else {
489
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
490
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_clusters[outl_col][outl_clust].col_num]
491
+ [model_outputs.all_clusters[outl_col][outl_clust].split_lev]);
492
+ }
493
+ break;
494
+ }
495
+
496
+ }
497
+ lst_cond[row] = Rcpp::List::create(Rcpp::clone(cond_clust));
498
+
499
+ /* finally, add conditions from branches that lead to the cluster */
500
+ curr_tree = model_outputs.outlier_trees_final[row];
501
+ Rcpp::List temp_list;
502
+ while (true) {
503
+ if (curr_tree == 0 || model_outputs.all_trees[outl_col][curr_tree].parent_branch == SubTrees) {
504
+ break;
505
+ }
506
+ parent_tree = model_outputs.all_trees[outl_col][curr_tree].parent;
507
+ cond_clust = Rcpp::List();
508
+
509
+ /* when using 'follow_all' */
510
+ if (model_outputs.all_trees[outl_col][parent_tree].all_branches.size() > 0) {
511
+
512
+ /* add column name and value */
513
+ switch(model_outputs.all_trees[outl_col][curr_tree].column_type) {
514
+ case Numeric:
515
+ {
516
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_num[model_outputs.all_trees[outl_col][curr_tree].col_num]);
517
+ break;
518
+ }
519
+
520
+ case Categorical:
521
+ {
522
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_cat[model_outputs.all_trees[outl_col][curr_tree].col_num]);
523
+ break;
524
+ }
525
+
526
+ case Ordinal:
527
+ {
528
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_ord[model_outputs.all_trees[outl_col][curr_tree].col_num]);
529
+ break;
530
+ }
531
+ }
532
+
533
+ /* add conditions from tree */
534
+ switch(model_outputs.all_trees[outl_col][curr_tree].column_type) {
535
+
536
+ case Numeric:
537
+ {
538
+ /* add decimals if appropriate */
539
+ if (
540
+ model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_num_num &&
541
+ model_outputs.all_trees[outl_col][curr_tree].split_this_branch != IsNa
542
+ )
543
+ {
544
+ cond_clust["decimals"] = Rcpp::wrap(model_outputs.min_decimals_col[model_outputs.all_trees[outl_col][curr_tree].col_num]);
545
+ }
546
+
547
+ /* then conditions */
548
+ switch(model_outputs.all_trees[outl_col][curr_tree].split_this_branch) {
549
+
550
+ case IsNa:
551
+ {
552
+ cond_clust["value_this"] = Rcpp::wrap(NA_REAL);
553
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
554
+ cond_clust["value_comp"] = Rcpp::wrap(NA_REAL);
555
+ break;
556
+ }
557
+
558
+ case LessOrEqual:
559
+ {
560
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_num_num) {
561
+ cond_clust["value_this"] = Rcpp::wrap(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
562
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
563
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_trees[outl_col][curr_tree].split_point);
564
+ } else if (model_outputs.all_trees[outl_col][curr_tree].col_num < (ncols_num_num + ncols_date)) {
565
+ cond_clust["value_this"] = Rcpp::Date(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]
566
+ - 1 + min_date[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num]);
567
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
568
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_trees[outl_col][curr_tree].split_point
569
+ - 1 + min_date[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num]);
570
+ } else {
571
+ cond_clust["value_this"] = Rcpp::Datetime(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]
572
+ - 1 + min_ts[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num - ncols_date]);
573
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
574
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_trees[outl_col][curr_tree].split_point
575
+ - 1 + min_ts[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num - ncols_date]);
576
+ }
577
+ break;
578
+ }
579
+
580
+ case Greater:
581
+ {
582
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_num_num) {
583
+ cond_clust["value_this"] = Rcpp::wrap(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
584
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
585
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_trees[outl_col][curr_tree].split_point);
586
+ } else if (model_outputs.all_trees[outl_col][curr_tree].col_num < (ncols_num_num + ncols_date)) {
587
+ cond_clust["value_this"] = Rcpp::Date(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]
588
+ - 1 + min_date[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num]);
589
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
590
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_trees[outl_col][curr_tree].split_point
591
+ - 1 + min_date[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num]);
592
+ } else {
593
+ cond_clust["value_this"] = Rcpp::Datetime(arr_num[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]
594
+ - 1 + min_ts[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num - ncols_date]);
595
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
596
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_trees[outl_col][curr_tree].split_point
597
+ - 1 + min_ts[model_outputs.all_trees[outl_col][curr_tree].col_num - ncols_num_num - ncols_date]);
598
+ }
599
+ break;
600
+ }
601
+
602
+ }
603
+ break;
604
+ }
605
+
606
+ case Categorical:
607
+ {
608
+ switch(model_outputs.all_trees[outl_col][curr_tree].split_this_branch) {
609
+
610
+ case IsNa:
611
+ {
612
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_cat_cat) {
613
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
614
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
615
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
616
+ } else {
617
+ cond_clust["value_this"] = Rcpp::LogicalVector(1, NA_LOGICAL);
618
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
619
+ cond_clust["value_comp"] = Rcpp::LogicalVector(1, NA_LOGICAL);
620
+ }
621
+ break;
622
+ }
623
+
624
+ case InSubset:
625
+ {
626
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_cat_cat) {
627
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_trees[outl_col][curr_tree].split_subset.size(), false);
628
+ for (size_t cat = 0; cat < model_outputs.all_trees[outl_col][curr_tree].split_subset.size(); cat++) {
629
+ if (model_outputs.all_trees[outl_col][curr_tree].split_subset[cat] > 0) {
630
+ tmp_bool[cat] = true;
631
+ }
632
+ }
633
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
634
+ [arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
635
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
636
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num][tmp_bool]);
637
+ } else {
638
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
639
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
640
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][curr_tree].split_subset[1]);
641
+ }
642
+ break;
643
+ }
644
+
645
+ case NotInSubset:
646
+ {
647
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_cat_cat) {
648
+ tmp_bool = Rcpp::LogicalVector(model_outputs.all_trees[outl_col][curr_tree].split_subset.size(), true);
649
+ for (size_t cat = 0; cat < model_outputs.all_trees[outl_col][curr_tree].split_subset.size(); cat++) {
650
+ if (model_outputs.all_trees[outl_col][curr_tree].split_subset[cat] > 0) {
651
+ tmp_bool[cat] = false;
652
+ }
653
+ }
654
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
655
+ [arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
656
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
657
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num][tmp_bool]);
658
+ } else {
659
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
660
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
661
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][curr_tree].split_subset[0]);
662
+ }
663
+ break;
664
+ }
665
+
666
+ case Equal:
667
+ {
668
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_cat_cat) {
669
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
670
+ [arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
671
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
672
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
673
+ [model_outputs.all_trees[outl_col][curr_tree].split_lev]);
674
+ } else {
675
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
676
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
677
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][curr_tree].split_lev);
678
+ }
679
+ break;
680
+ }
681
+
682
+ case NotEqual:
683
+ {
684
+ if (model_outputs.all_trees[outl_col][curr_tree].col_num < ncols_cat_cat) {
685
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
686
+ [arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
687
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
688
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
689
+ [model_outputs.all_trees[outl_col][curr_tree].split_lev]);
690
+ } else {
691
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]);
692
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
693
+ cond_clust["value_comp"] = Rcpp::wrap((bool) !model_outputs.all_trees[outl_col][curr_tree].split_lev);
694
+ /* note: booleans should always get converted to Equals, this code is redundant */
695
+ }
696
+ break;
697
+ }
698
+
699
+ }
700
+ break;
701
+ }
702
+
703
+ case Ordinal:
704
+ {
705
+ switch(model_outputs.all_trees[outl_col][curr_tree].split_this_branch) {
706
+
707
+ case IsNa:
708
+ {
709
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
710
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
711
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
712
+ break;
713
+ }
714
+
715
+ case LessOrEqual:
716
+ {
717
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num].size(), false);
718
+ for (int cat = 0; cat <= model_outputs.all_trees[outl_col][curr_tree].split_lev; cat++) {
719
+ tmp_bool[cat] = true;
720
+ }
721
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
722
+ [arr_ord[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
723
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
724
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num][tmp_bool]);
725
+ break;
726
+ }
727
+
728
+ case Greater:
729
+ {
730
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num].size(), true);
731
+ for (int cat = 0; cat <= model_outputs.all_trees[outl_col][curr_tree].split_lev; cat++) {
732
+ tmp_bool[cat] = false;
733
+ }
734
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
735
+ [arr_ord[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
736
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
737
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num][tmp_bool]);
738
+ break;
739
+ }
740
+
741
+ case Equal:
742
+ {
743
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
744
+ [arr_ord[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
745
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
746
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
747
+ [model_outputs.all_trees[outl_col][curr_tree].split_lev]);
748
+ break;
749
+ }
750
+
751
+ case NotEqual:
752
+ {
753
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
754
+ [arr_ord[row + model_outputs.all_trees[outl_col][curr_tree].col_num * nrows]]);
755
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
756
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][curr_tree].col_num]
757
+ [model_outputs.all_trees[outl_col][curr_tree].split_lev]);
758
+ break;
759
+ }
760
+
761
+ }
762
+ break;
763
+ }
764
+
765
+ }
766
+ }
767
+
768
+ /* regular case (no 'follow_all') */
769
+ else
770
+ {
771
+
772
+ /* add column name and value */
773
+ switch(model_outputs.all_trees[outl_col][parent_tree].column_type) {
774
+ case Numeric:
775
+ {
776
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_num[model_outputs.all_trees[outl_col][parent_tree].col_num]);
777
+ /* add decimals if appropriate */
778
+ if (
779
+ model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_num_num &&
780
+ model_outputs.all_trees[outl_col][curr_tree].parent_branch != IsNa
781
+ )
782
+ {
783
+ cond_clust["decimals"] = Rcpp::wrap(model_outputs.min_decimals_col[model_outputs.all_trees[outl_col][parent_tree].col_num]);
784
+ }
785
+ break;
786
+ }
787
+
788
+ case Categorical:
789
+ {
790
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_cat[model_outputs.all_trees[outl_col][parent_tree].col_num]);
791
+ break;
792
+ }
793
+
794
+ case Ordinal:
795
+ {
796
+ cond_clust["column"] = Rcpp::as<Rcpp::CharacterVector>(colnames_ord[model_outputs.all_trees[outl_col][parent_tree].col_num]);
797
+ break;
798
+ }
799
+ }
800
+
801
+
802
+ /* add conditions from tree */
803
+ switch(model_outputs.all_trees[outl_col][curr_tree].parent_branch) {
804
+
805
+
806
+ case IsNa:
807
+ {
808
+ switch(model_outputs.all_trees[outl_col][parent_tree].column_type) {
809
+ case Numeric:
810
+ {
811
+ cond_clust["value_this"] = Rcpp::wrap(NA_REAL);
812
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
813
+ cond_clust["value_comp"] = Rcpp::wrap(NA_REAL);
814
+ break;
815
+ }
816
+
817
+ case Categorical:
818
+ {
819
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
820
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
821
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
822
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
823
+ } else {
824
+ cond_clust["value_this"] = Rcpp::LogicalVector(1, NA_LOGICAL);
825
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
826
+ cond_clust["value_comp"] = Rcpp::LogicalVector(1, NA_LOGICAL);
827
+ }
828
+ break;
829
+ }
830
+
831
+ case Ordinal:
832
+ {
833
+ cond_clust["value_this"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
834
+ cond_clust["comparison"] = Rcpp::CharacterVector("is NA");
835
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(NA_STRING);
836
+ break;
837
+ }
838
+ }
839
+ break;
840
+ }
841
+
842
+ case LessOrEqual:
843
+ {
844
+ if (model_outputs.all_trees[outl_col][parent_tree].column_type == Numeric) {
845
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_num_num) {
846
+ cond_clust["value_this"] = Rcpp::wrap(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
847
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
848
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_trees[outl_col][parent_tree].split_point);
849
+ } else if (model_outputs.all_trees[outl_col][parent_tree].col_num < (ncols_num_num + ncols_date)) {
850
+ cond_clust["value_this"] = Rcpp::Date(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]
851
+ - 1 + min_date[model_outputs.all_trees[outl_col][parent_tree].col_num - ncols_num_num]);
852
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
853
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_trees[outl_col][parent_tree].split_point
854
+ - 1 + min_date[model_outputs.all_trees[outl_col][parent_tree].col_num - ncols_num_num]);
855
+ } else {
856
+ cond_clust["value_this"] = Rcpp::Datetime(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]
857
+ - 1 + min_ts[model_outputs.all_trees[outl_col][parent_tree].col_num
858
+ - ncols_num_num - ncols_date]);
859
+ cond_clust["comparison"] = Rcpp::CharacterVector("<=");
860
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_trees[outl_col][parent_tree].split_point
861
+ - 1 + min_ts[model_outputs.all_trees[outl_col][parent_tree].col_num
862
+ - ncols_num_num - ncols_date]);
863
+ }
864
+ } else {
865
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num].size(), false);
866
+ for (int cat = 0; cat <= model_outputs.all_trees[outl_col][parent_tree].split_lev; cat++) tmp_bool[cat] = true;
867
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
868
+ [arr_ord[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
869
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
870
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num][tmp_bool]);
871
+ }
872
+ break;
873
+ }
874
+
875
+ case Greater:
876
+ {
877
+ if (model_outputs.all_trees[outl_col][parent_tree].column_type == Numeric) {
878
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_num_num) {
879
+ cond_clust["value_this"] = Rcpp::wrap(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
880
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
881
+ cond_clust["value_comp"] = Rcpp::wrap(model_outputs.all_trees[outl_col][parent_tree].split_point);
882
+ } else if (model_outputs.all_trees[outl_col][parent_tree].col_num < (ncols_num_num + ncols_date)) {
883
+ cond_clust["value_this"] = Rcpp::Date(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]
884
+ - 1 + min_date[model_outputs.all_trees[outl_col][parent_tree].col_num - ncols_num_num]);
885
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
886
+ cond_clust["value_comp"] = Rcpp::Date(model_outputs.all_trees[outl_col][parent_tree].split_point
887
+ - 1 + min_date[model_outputs.all_trees[outl_col][parent_tree].col_num - ncols_num_num]);
888
+ } else {
889
+ cond_clust["value_this"] = Rcpp::Datetime(arr_num[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]
890
+ - 1 + min_ts[model_outputs.all_trees[outl_col][parent_tree].col_num
891
+ - ncols_num_num - ncols_date]);
892
+ cond_clust["comparison"] = Rcpp::CharacterVector(">");
893
+ cond_clust["value_comp"] = Rcpp::Datetime(model_outputs.all_trees[outl_col][parent_tree].split_point
894
+ - 1 + min_ts[model_outputs.all_trees[outl_col][parent_tree].col_num
895
+ - ncols_num_num - ncols_date]);
896
+ }
897
+ } else {
898
+ tmp_bool = Rcpp::LogicalVector(ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num].size(), true);
899
+ for (int cat = 0; cat <= model_outputs.all_trees[outl_col][parent_tree].split_lev; cat++) tmp_bool[cat] = false;
900
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
901
+ [arr_ord[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
902
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
903
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num][tmp_bool]);
904
+ }
905
+ break;
906
+ }
907
+
908
+ case InSubset:
909
+ {
910
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
911
+ tmp_bool = Rcpp::LogicalVector(cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num].size(), false);
912
+ for (size_t cat = 0; cat < model_outputs.all_trees[outl_col][parent_tree].split_subset.size(); cat++) {
913
+ if (model_outputs.all_trees[outl_col][parent_tree].split_subset[cat] > 0) {
914
+ tmp_bool[cat] = true;
915
+ }
916
+ }
917
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
918
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
919
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
920
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num][tmp_bool]);
921
+ } else {
922
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
923
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
924
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][parent_tree].split_subset[1]);
925
+ }
926
+ break;
927
+ }
928
+
929
+ case NotInSubset:
930
+ {
931
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
932
+ tmp_bool = Rcpp::LogicalVector(cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num].size(), false);
933
+ for (size_t cat = 0; cat < model_outputs.all_trees[outl_col][parent_tree].split_subset.size(); cat++) {
934
+ if (model_outputs.all_trees[outl_col][parent_tree].split_subset[cat] == 0) {
935
+ tmp_bool[cat] = true;
936
+ }
937
+ }
938
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
939
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
940
+ cond_clust["comparison"] = Rcpp::CharacterVector("in");
941
+ cond_clust["value_comp"] = Rcpp::as<Rcpp::CharacterVector>(cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num][tmp_bool]);
942
+ } else {
943
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
944
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
945
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][parent_tree].split_subset[0]);
946
+ }
947
+ break;
948
+ }
949
+
950
+ case Equal:
951
+ {
952
+ if (model_outputs.all_trees[outl_col][parent_tree].column_type == Categorical) {
953
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
954
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
955
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
956
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
957
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
958
+ [model_outputs.all_trees[outl_col][parent_tree].split_lev]);
959
+ } else {
960
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
961
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
962
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][parent_tree].split_subset[1]);
963
+ }
964
+ } else {
965
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
966
+ [arr_ord[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
967
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
968
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
969
+ [model_outputs.all_trees[outl_col][parent_tree].split_lev]);
970
+ }
971
+ break;
972
+ }
973
+
974
+ case NotEqual:
975
+ {
976
+ if (model_outputs.all_trees[outl_col][parent_tree].column_type == Categorical) {
977
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
978
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
979
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
980
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
981
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
982
+ [model_outputs.all_trees[outl_col][parent_tree].split_lev]);
983
+ } else {
984
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
985
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
986
+ cond_clust["value_comp"] = Rcpp::wrap((bool) model_outputs.all_trees[outl_col][parent_tree].split_subset[0]);
987
+ }
988
+ } else {
989
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
990
+ [arr_ord[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
991
+ cond_clust["comparison"] = Rcpp::CharacterVector("!=");
992
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, ord_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
993
+ [model_outputs.all_trees[outl_col][parent_tree].split_lev]);
994
+ }
995
+ break;
996
+ }
997
+
998
+ case SingleCateg:
999
+ {
1000
+ if (model_outputs.all_trees[outl_col][parent_tree].col_num < ncols_cat_cat) {
1001
+ cond_clust["value_this"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
1002
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
1003
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
1004
+ cond_clust["value_comp"] = Rcpp::CharacterVector(1, cat_levels[model_outputs.all_trees[outl_col][parent_tree].col_num]
1005
+ [arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]]);
1006
+ } else {
1007
+ cond_clust["value_this"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
1008
+ cond_clust["comparison"] = Rcpp::CharacterVector("=");
1009
+ cond_clust["value_comp"] = Rcpp::wrap((bool) arr_cat[row + model_outputs.all_trees[outl_col][parent_tree].col_num * nrows]);
1010
+ }
1011
+ break;
1012
+ }
1013
+
1014
+ }
1015
+
1016
+
1017
+ }
1018
+
1019
+ /* https://github.com/RcppCore/Rcpp/issues/979 */
1020
+ /* this comment below will fix Rcpp issue with having slashes in the comment above */
1021
+ temp_list = lst_cond[row];
1022
+ temp_list.push_back(Rcpp::clone(cond_clust));
1023
+ lst_cond[row] = temp_list;
1024
+ curr_tree = parent_tree;
1025
+ }
1026
+
1027
+ }
1028
+
1029
+ }
1030
+ }
1031
+
1032
+ outp["suspicous_value"] = outlier_val;
1033
+ outp["group_statistics"] = lst_stats;
1034
+ outp["conditions"] = lst_cond;
1035
+ outp["tree_depth"] = tree_depth;
1036
+ outp["uses_NA_branch"] = has_na_col;
1037
+ outp["outlier_score"] = outlier_score;
1038
+ return outp;
1039
+ }
1040
+
1041
+ /* for extracting info about flaggable outliers */
1042
+ Rcpp::List extract_outl_bounds(ModelOutputs &model_outputs,
1043
+ Rcpp::ListOf<Rcpp::StringVector> cat_levels,
1044
+ Rcpp::ListOf<Rcpp::StringVector> ord_levels,
1045
+ Rcpp::NumericVector min_date,
1046
+ Rcpp::NumericVector min_ts)
1047
+ {
1048
+ size_t ncols_num = model_outputs.ncols_numeric;
1049
+ size_t ncols_cat = model_outputs.ncols_categ;
1050
+ size_t ncols_ord = model_outputs.ncols_ord;
1051
+ size_t col_lim_num = model_outputs.ncols_numeric - min_date.size() - min_ts.size();
1052
+ size_t col_lim_date = model_outputs.ncols_numeric - min_ts.size();
1053
+ size_t ncols_cat_cat = cat_levels.size();
1054
+ size_t tot_cols = ncols_num + ncols_cat + ncols_ord;
1055
+ Rcpp::LogicalVector temp_bool;
1056
+ Rcpp::LogicalVector bool_choice(2, false); bool_choice[1] = true;
1057
+ Rcpp::List outp(tot_cols);
1058
+
1059
+ for (size_t cl = 0; cl < tot_cols; cl++) {
1060
+ if (cl < col_lim_num) {
1061
+ /* numeric */
1062
+ outp[cl] = Rcpp::List::create(Rcpp::_["lb"] = Rcpp::wrap(model_outputs.min_outlier_any_cl[cl]),
1063
+ Rcpp::_["ub"] = Rcpp::wrap(model_outputs.max_outlier_any_cl[cl]));
1064
+ } else if (cl < col_lim_date) {
1065
+ /* date */
1066
+ outp[cl] = Rcpp::List::create(
1067
+ Rcpp::_["lb"] = Rcpp::Date(model_outputs.min_outlier_any_cl[cl] - 1 + min_date[cl - col_lim_num]),
1068
+ Rcpp::_["ub"] = Rcpp::Date(model_outputs.max_outlier_any_cl[cl] - 1 + min_date[cl - col_lim_num])
1069
+ );
1070
+ } else if (cl < ncols_num) {
1071
+ /* timestamp */
1072
+ outp[cl] = Rcpp::List::create(
1073
+ Rcpp::_["lb"] = Rcpp::Datetime(model_outputs.min_outlier_any_cl[cl] - 1 + min_ts[cl - col_lim_date]),
1074
+ Rcpp::_["ub"] = Rcpp::Datetime(model_outputs.max_outlier_any_cl[cl] - 1 + min_ts[cl - col_lim_date])
1075
+ );
1076
+ } else if (cl < (ncols_num + ncols_cat_cat)) {
1077
+ /* categorical */
1078
+ if (model_outputs.cat_outlier_any_cl[cl - ncols_num].size()) {
1079
+ temp_bool = Rcpp::wrap(model_outputs.cat_outlier_any_cl[cl - ncols_num]);
1080
+ outp[cl] = cat_levels[cl - ncols_num][temp_bool];
1081
+ } else {
1082
+ outp[cl] = Rcpp::StringVector();
1083
+ }
1084
+ } else if (cl < (ncols_num + ncols_cat)) {
1085
+ /* boolean */
1086
+ if (model_outputs.cat_outlier_any_cl[cl - ncols_num].size()) {
1087
+ temp_bool = Rcpp::wrap(model_outputs.cat_outlier_any_cl[cl - ncols_num]);
1088
+ outp[cl] = bool_choice[temp_bool];
1089
+ } else {
1090
+ outp[cl] = Rcpp::LogicalVector();
1091
+ }
1092
+ } else {
1093
+ /* ordinal */
1094
+ if (model_outputs.cat_outlier_any_cl[cl - ncols_num].size()) {
1095
+ temp_bool = Rcpp::wrap(model_outputs.cat_outlier_any_cl[cl - ncols_num]);
1096
+ outp[cl] = ord_levels[cl - ncols_num - ncols_cat][temp_bool];
1097
+ } else {
1098
+ outp[cl] = Rcpp::StringVector();
1099
+ }
1100
+ }
1101
+ }
1102
+ return outp;
1103
+ }
1104
+
1105
+
1106
+ /* external functions for fitting the model and predicting outliers */
1107
+ // [[Rcpp::export]]
1108
+ Rcpp::List fit_OutlierTree(Rcpp::NumericVector arr_num, size_t ncols_numeric,
1109
+ Rcpp::IntegerVector arr_cat, size_t ncols_categ, Rcpp::IntegerVector ncat,
1110
+ Rcpp::IntegerVector arr_ord, size_t ncols_ord, Rcpp::IntegerVector ncat_ord,
1111
+ size_t nrows, Rcpp::LogicalVector cols_ignore_r, int nthreads,
1112
+ bool categ_as_bin, bool ord_as_bin, bool cat_bruteforce_subset, bool categ_from_maj, bool take_mid,
1113
+ size_t max_depth, double max_perc_outliers, size_t min_size_numeric, size_t min_size_categ,
1114
+ double min_gain, bool follow_all, bool gain_as_pct, double z_norm, double z_outlier,
1115
+ bool return_outliers,
1116
+ Rcpp::ListOf<Rcpp::StringVector> cat_levels,
1117
+ Rcpp::ListOf<Rcpp::StringVector> ord_levels,
1118
+ Rcpp::StringVector colnames_num,
1119
+ Rcpp::StringVector colnames_cat,
1120
+ Rcpp::StringVector colnames_ord,
1121
+ Rcpp::NumericVector min_date,
1122
+ Rcpp::NumericVector min_ts)
1123
+ {
1124
+ bool found_outliers;
1125
+ Rcpp::List outp;
1126
+ size_t tot_cols = ncols_numeric + ncols_categ + ncols_ord;
1127
+ std::vector<char> cols_ignore;
1128
+ char *cols_ignore_ptr = NULL;
1129
+ if (cols_ignore_r.size() > 0) {
1130
+ cols_ignore.resize(tot_cols, false);
1131
+ for (size_t cl = 0; cl < tot_cols; cl++) cols_ignore[cl] = (bool) cols_ignore_r[cl];
1132
+ cols_ignore_ptr = &cols_ignore[0];
1133
+ }
1134
+ std::vector<double> Xcpp;
1135
+ double *arr_num_C = set_R_nan_as_C_nan(&arr_num[0], Xcpp, arr_num.size(), nthreads);
1136
+
1137
+ std::unique_ptr<ModelOutputs> model_outputs = std::unique_ptr<ModelOutputs>(new ModelOutputs());
1138
+ found_outliers = fit_outliers_models(*model_outputs,
1139
+ arr_num_C, ncols_numeric,
1140
+ &arr_cat[0], ncols_categ, &ncat[0],
1141
+ &arr_ord[0], ncols_ord, &ncat_ord[0],
1142
+ nrows, cols_ignore_ptr, nthreads,
1143
+ categ_as_bin, ord_as_bin, cat_bruteforce_subset, categ_from_maj, take_mid,
1144
+ max_depth, max_perc_outliers, min_size_numeric, min_size_categ,
1145
+ min_gain, gain_as_pct, follow_all, z_norm, z_outlier);
1146
+
1147
+ outp["bounds"] = extract_outl_bounds(*model_outputs,
1148
+ cat_levels,
1149
+ ord_levels,
1150
+ min_date,
1151
+ min_ts);
1152
+
1153
+ outp["serialized_obj"] = serialize_OutlierTree(model_outputs.get());
1154
+ if (return_outliers) {
1155
+ outp["outliers_info"] = describe_outliers(*model_outputs,
1156
+ arr_num_C,
1157
+ &arr_cat[0],
1158
+ &arr_ord[0],
1159
+ cat_levels,
1160
+ ord_levels,
1161
+ colnames_num,
1162
+ colnames_cat,
1163
+ colnames_ord,
1164
+ min_date,
1165
+ min_ts);
1166
+ }
1167
+ /* add number of trees and clusters */
1168
+ size_t ntrees = 0, nclust = 0;
1169
+ for (size_t col = 0; col < model_outputs->all_trees.size(); col++) {
1170
+ ntrees += model_outputs->all_trees[col].size();
1171
+ nclust += model_outputs->all_clusters[col].size();
1172
+ }
1173
+ outp["ntrees"] = Rcpp::wrap((int) ntrees);
1174
+ outp["nclust"] = Rcpp::wrap((int) nclust);
1175
+ outp["found_outliers"] = Rcpp::wrap(found_outliers);
1176
+
1177
+ forget_row_outputs(*model_outputs);
1178
+ outp["ptr_model"] = Rcpp::XPtr<ModelOutputs>(model_outputs.release(), true);
1179
+ return outp;
1180
+ }
1181
+
1182
+ // [[Rcpp::export]]
1183
+ Rcpp::List predict_OutlierTree(SEXP ptr_model, size_t nrows, int nthreads,
1184
+ Rcpp::NumericVector arr_num, Rcpp::IntegerVector arr_cat, Rcpp::IntegerVector arr_ord,
1185
+ Rcpp::ListOf<Rcpp::StringVector> cat_levels,
1186
+ Rcpp::ListOf<Rcpp::StringVector> ord_levels,
1187
+ Rcpp::StringVector colnames_num,
1188
+ Rcpp::StringVector colnames_cat,
1189
+ Rcpp::StringVector colnames_ord,
1190
+ Rcpp::NumericVector min_date,
1191
+ Rcpp::NumericVector min_ts)
1192
+ {
1193
+ std::vector<double> Xcpp;
1194
+ double *arr_num_C = set_R_nan_as_C_nan(&arr_num[0], Xcpp, arr_num.size(), nthreads);
1195
+
1196
+ ModelOutputs *model_outputs = static_cast<ModelOutputs*>(R_ExternalPtrAddr(ptr_model));
1197
+ bool found_outliers = find_new_outliers(&arr_num[0], &arr_cat[0], &arr_ord[0],
1198
+ nrows, nthreads, *model_outputs);
1199
+ Rcpp::List outp = describe_outliers(*model_outputs,
1200
+ arr_num_C,
1201
+ &arr_cat[0],
1202
+ &arr_ord[0],
1203
+ cat_levels,
1204
+ ord_levels,
1205
+ colnames_num,
1206
+ colnames_cat,
1207
+ colnames_ord,
1208
+ min_date,
1209
+ min_ts);
1210
+ outp["found_outliers"] = Rcpp::LogicalVector(found_outliers);
1211
+ forget_row_outputs(*model_outputs);
1212
+ return outp;
1213
+ }
1214
+
1215
+ // [[Rcpp::export]]
1216
+ Rcpp::LogicalVector check_few_values(Rcpp::NumericVector arr_num, size_t nrows, size_t ncols, int nthreads)
1217
+ {
1218
+ std::vector<char> too_few_vals(ncols, 0);
1219
+ check_more_two_values(&arr_num[0], nrows, ncols, nthreads, too_few_vals.data());
1220
+ Rcpp::LogicalVector outp(ncols);
1221
+ for (size_t col = 0; col < ncols; col++) {
1222
+ outp[col] = (bool) too_few_vals[col];
1223
+ }
1224
+ return outp;
1225
+ }