outliertree 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/CHANGELOG.md +3 -0
- data/LICENSE.txt +674 -0
- data/NOTICE.txt +14 -0
- data/README.md +107 -0
- data/ext/outliertree/ext.cpp +260 -0
- data/ext/outliertree/extconf.rb +21 -0
- data/lib/outliertree.rb +17 -0
- data/lib/outliertree/dataset.rb +35 -0
- data/lib/outliertree/model.rb +128 -0
- data/lib/outliertree/result.rb +190 -0
- data/lib/outliertree/version.rb +3 -0
- data/vendor/outliertree/LICENSE +674 -0
- data/vendor/outliertree/README.md +155 -0
- data/vendor/outliertree/src/Makevars +3 -0
- data/vendor/outliertree/src/RcppExports.cpp +123 -0
- data/vendor/outliertree/src/Rwrapper.cpp +1225 -0
- data/vendor/outliertree/src/cat_outlier.cpp +328 -0
- data/vendor/outliertree/src/clusters.cpp +972 -0
- data/vendor/outliertree/src/fit_model.cpp +1932 -0
- data/vendor/outliertree/src/misc.cpp +685 -0
- data/vendor/outliertree/src/outlier_tree.hpp +758 -0
- data/vendor/outliertree/src/predict.cpp +706 -0
- data/vendor/outliertree/src/split.cpp +1098 -0
- metadata +150 -0
@@ -0,0 +1,328 @@
|
|
1
|
+
/********************************************************************************************************************
|
2
|
+
* Explainable outlier detection
|
3
|
+
*
|
4
|
+
* Tries to detect outliers by generating decision trees that attempt to predict the values of each column based on
|
5
|
+
* each other column, testing in each branch of every tried split (if it meets some minimum criteria) whether there
|
6
|
+
* are observations that seem too distant from the others in a 1-D distribution for the column that the split tries
|
7
|
+
* to "predict" (will not generate a score for each observation).
|
8
|
+
* Splits are based on gain, while outlierness is based on confidence intervals.
|
9
|
+
* Similar in spirit to the GritBot software developed by RuleQuest research. Reference article is:
|
10
|
+
* Cortes, David. "Explainable outlier detection through decision tree conditioning."
|
11
|
+
* arXiv preprint arXiv:2001.00636 (2020).
|
12
|
+
*
|
13
|
+
*
|
14
|
+
* Copyright 2020 David Cortes.
|
15
|
+
*
|
16
|
+
* Written for C++11 standard and OpenMP 2.0 or later. Code is meant to be wrapped into scripting languages
|
17
|
+
* such as R or Python.
|
18
|
+
*
|
19
|
+
* This file is part of OutlierTree.
|
20
|
+
*
|
21
|
+
* OutlierTree is free software: you can redistribute it and/or modify
|
22
|
+
* it under the terms of the GNU General Public License as published by
|
23
|
+
* the Free Software Foundation, either version 3 of the License, or
|
24
|
+
* (at your option) any later version.
|
25
|
+
*
|
26
|
+
* OutlierTree is distributed in the hope that it will be useful,
|
27
|
+
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
28
|
+
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
29
|
+
* GNU General Public License for more details.
|
30
|
+
*
|
31
|
+
* You should have received a copy of the GNU General Public License
|
32
|
+
* along with OutlierTree. If not, see <https://www.gnu.org/licenses/>.
|
33
|
+
********************************************************************************************************************/
|
34
|
+
#include "outlier_tree.hpp"
|
35
|
+
|
36
|
+
|
37
|
+
/* Check whether to consider any category as outlier, based on current counts and prior probabilities
|
38
|
+
*
|
39
|
+
* Function is to be applied to some subset of the data obtained by splitting by one or more columns.
|
40
|
+
* For outliers before any split there is a separate function. Note that since it required current
|
41
|
+
* probability to be lower than prior probability in order to consider as outlier, it cannot be
|
42
|
+
* used with the full data (only with subsets).
|
43
|
+
*
|
44
|
+
* Parameters:
|
45
|
+
* - categ_counts[ncateg] (in)
|
46
|
+
* Counts of each category in the subset (including non-present categories).
|
47
|
+
* - ncateg (in)
|
48
|
+
* Number of categories for this column (including non-present categories).
|
49
|
+
* - tot (in)
|
50
|
+
* Number of rows in the subset.
|
51
|
+
* - max_perc_outliers (in)
|
52
|
+
* Model parameter. Default value is 0.01.
|
53
|
+
* - perc_threshold[ncateg] (in)
|
54
|
+
* Threshold for the proportion/probability of each category below which it can be considered
|
55
|
+
* to be an outlier in a subset of the data. Note that in addition it will build a confidence
|
56
|
+
* interval here which might make it even smaller.
|
57
|
+
* - buffer_ix[ncateg] (temp)
|
58
|
+
* Buffer where to store indices of categories sorted by proportion.
|
59
|
+
* - buffer_perc[ncateg] (temp)
|
60
|
+
* Buffer where to store proportions of counts.
|
61
|
+
* - z_norm (in)
|
62
|
+
* Model parameter. Default value is 2.67.
|
63
|
+
* - is_outlier[ncateg] (out)
|
64
|
+
* Array where to define whether any category is an outlier. Values will be as follows:
|
65
|
+
* (-1) -> Category had zero count, but would be an outlier if it appeared among this group
|
66
|
+
* 0 -> Category is not an outlier
|
67
|
+
* (+1) -> Category is an outlier
|
68
|
+
* - found_outliers (out)
|
69
|
+
* Whether there were any outliers identified among the counts.
|
70
|
+
* - new_is_outlier (out)
|
71
|
+
* Whether any of the categories with zero count would be flagged as outlier if they appeared in this group.
|
72
|
+
* - next_most_comm (out)
|
73
|
+
* Proportion of the least common category that is not flagged as outlier.
|
74
|
+
*/
|
75
|
+
void find_outlier_categories(size_t categ_counts[], size_t ncateg, size_t tot, double max_perc_outliers,
|
76
|
+
long double perc_threshold[], size_t buffer_ix[], long double buffer_perc[],
|
77
|
+
double z_norm, char is_outlier[], bool *found_outliers, bool *new_is_outlier,
|
78
|
+
double *next_most_comm)
|
79
|
+
{
|
80
|
+
//TODO: must also establish bounds for new, unseen categories
|
81
|
+
|
82
|
+
/* initialize parameters as needed */
|
83
|
+
*found_outliers = false;
|
84
|
+
*new_is_outlier = false;
|
85
|
+
size_t st_non_zero = 0;
|
86
|
+
size_t end_tail = 0;
|
87
|
+
size_t max_outliers = (size_t) calculate_max_cat_outliers((long double)tot, max_perc_outliers, z_norm);
|
88
|
+
long double tot_dbl = (long double) tot;
|
89
|
+
long double pct_unseen = (long double)1 / (long double)(tot + 1);
|
90
|
+
size_t size_tail = 0;
|
91
|
+
|
92
|
+
/* reset the temporary arrays and fill them */
|
93
|
+
memset(is_outlier, 0, ncateg * sizeof(char));
|
94
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
95
|
+
buffer_ix[cat] = cat;
|
96
|
+
buffer_perc[cat] = (categ_counts[cat] > 0)? ((long double)categ_counts[cat] / tot_dbl) : 0;
|
97
|
+
}
|
98
|
+
|
99
|
+
/* sort the categories by counts */
|
100
|
+
std::sort(buffer_ix, buffer_ix + ncateg,
|
101
|
+
[&categ_counts](const size_t a, const size_t b){return categ_counts[a] < categ_counts[b];});
|
102
|
+
|
103
|
+
/* find the first non-zero */
|
104
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
105
|
+
if (categ_counts[ buffer_ix[cat] ] > 0) {
|
106
|
+
st_non_zero = cat;
|
107
|
+
break;
|
108
|
+
}
|
109
|
+
}
|
110
|
+
|
111
|
+
/* check that least common is not common enough to be normal */
|
112
|
+
if (categ_counts[ buffer_ix[st_non_zero] ] > max_outliers) return;
|
113
|
+
|
114
|
+
/* find tail among non-zero proportions
|
115
|
+
* a tail is considered to be so if:
|
116
|
+
* - the difference is above z_norm sd's of either proportion
|
117
|
+
* - the difference is greater than some fraction of the larger
|
118
|
+
* - the actual proportion here is lower than a CI of the prior proportion
|
119
|
+
* - the actual proportion here is half or less of the prior proportion
|
120
|
+
*/
|
121
|
+
for (size_t cat = st_non_zero; cat < ncateg - 1; cat++) {
|
122
|
+
if (
|
123
|
+
(
|
124
|
+
(buffer_perc[buffer_ix[cat + 1]] - buffer_perc[buffer_ix[cat]])
|
125
|
+
>
|
126
|
+
z_norm * sqrtl(
|
127
|
+
fmaxl(
|
128
|
+
buffer_perc[buffer_ix[cat + 1]] * ((long double)1 - buffer_perc[buffer_ix[cat + 1]]),
|
129
|
+
buffer_perc[buffer_ix[cat]] * ((long double)1 - buffer_perc[buffer_ix[cat]])
|
130
|
+
)
|
131
|
+
/ tot_dbl
|
132
|
+
)
|
133
|
+
)
|
134
|
+
&&
|
135
|
+
(
|
136
|
+
buffer_perc[buffer_ix[cat + 1]] * 0.5 > buffer_perc[buffer_ix[cat]]
|
137
|
+
)
|
138
|
+
)
|
139
|
+
{
|
140
|
+
end_tail = cat;
|
141
|
+
*next_most_comm = buffer_perc[buffer_ix[cat + 1]];
|
142
|
+
break;
|
143
|
+
}
|
144
|
+
}
|
145
|
+
|
146
|
+
/* if the tail is too long, don't identify any as outlier, but see if unseen categories (with prior > 0) would create a new tail */
|
147
|
+
for (size_t cat = st_non_zero; cat <= end_tail; cat++) size_tail += categ_counts[ buffer_ix[cat] ];
|
148
|
+
|
149
|
+
if (size_tail >= max_outliers) {
|
150
|
+
|
151
|
+
if (
|
152
|
+
st_non_zero == 0 ||
|
153
|
+
// ((long double)buffer_ix[buffer_ix[st_non_zero]] / (tot_dbl + 1)) * 0.5 <= pct_unseen ||
|
154
|
+
( ((long double)buffer_ix[buffer_ix[st_non_zero]] * 0.5) / (tot_dbl + 1)) <= pct_unseen ||
|
155
|
+
((long double)(buffer_ix[buffer_ix[st_non_zero]] - 1) / (tot_dbl + 1))
|
156
|
+
- (long double)z_norm * sqrtl(buffer_perc[buffer_ix[st_non_zero]] * ((long double)1 - buffer_perc[buffer_ix[st_non_zero]]) / tot_dbl)
|
157
|
+
>= pct_unseen
|
158
|
+
) return;
|
159
|
+
|
160
|
+
for (size_t cat = 0; cat < st_non_zero; cat++) {
|
161
|
+
if (perc_threshold[buffer_ix[cat]] > pct_unseen) {
|
162
|
+
*new_is_outlier = true;
|
163
|
+
is_outlier[buffer_ix[cat]] = -1;
|
164
|
+
}
|
165
|
+
}
|
166
|
+
*next_most_comm = buffer_perc[buffer_ix[st_non_zero]];
|
167
|
+
return;
|
168
|
+
|
169
|
+
}
|
170
|
+
|
171
|
+
/* now determine if any category in the tail is an outlier */
|
172
|
+
for (size_t cat = st_non_zero; cat <= end_tail; cat++) {
|
173
|
+
|
174
|
+
/* must have a proportion below CI and below half of prior */
|
175
|
+
if (buffer_perc[buffer_ix[cat]] < perc_threshold[buffer_ix[cat]]) {
|
176
|
+
is_outlier[buffer_ix[cat]] = 1;
|
177
|
+
*found_outliers = true;
|
178
|
+
}
|
179
|
+
}
|
180
|
+
|
181
|
+
/* check if any new categories would be outliers */
|
182
|
+
if (st_non_zero > 0) {
|
183
|
+
for (size_t cat = 0; cat < st_non_zero; cat++) {
|
184
|
+
if (perc_threshold[buffer_ix[cat]] > pct_unseen) {
|
185
|
+
*new_is_outlier = true;
|
186
|
+
is_outlier[buffer_ix[cat]] = -1;
|
187
|
+
}
|
188
|
+
}
|
189
|
+
}
|
190
|
+
if (*new_is_outlier && !(*found_outliers)) {
|
191
|
+
*next_most_comm = buffer_perc[buffer_ix[st_non_zero]];
|
192
|
+
}
|
193
|
+
|
194
|
+
}
|
195
|
+
|
196
|
+
/* Check whether to consider any category as outlier, based on majority category and prior probabilties
|
197
|
+
*
|
198
|
+
* Function is to be applied to some subset of the data obtained by splitting by one or more columns.
|
199
|
+
* For outliers before any split there is a separate function. This is an alternative to the "tail"
|
200
|
+
* approach above which is more in line with GritBot.
|
201
|
+
*
|
202
|
+
* Parameters:
|
203
|
+
* - categ_counts[ncateg] (in)
|
204
|
+
* Counts of each category in the subset (including non-present categories).
|
205
|
+
* - ncateg (in)
|
206
|
+
* Number of categories for this column (including non-present categories).
|
207
|
+
* - tot (in)
|
208
|
+
* Number of rows in the subset.
|
209
|
+
* - max_perc_outliers (in)
|
210
|
+
* Model parameter. Default value is 0.01.
|
211
|
+
* - prior_prob[ncateg] (in)
|
212
|
+
* Proportions that each category had in the full data.
|
213
|
+
* - z_outlier (in)
|
214
|
+
* Model parameter. Default value is 8.0
|
215
|
+
* - is_outlier[ncateg] (out)
|
216
|
+
* Array where to define whether any category is an outlier. Values will be as follows:
|
217
|
+
* (-1) -> Category had zero count, but would be an outlier if it appeared among this group
|
218
|
+
* 0 -> Category is not an outlier
|
219
|
+
* (+1) -> Category is an outlier
|
220
|
+
* - found_outliers (out)
|
221
|
+
* Whether there were any outliers identified among the counts.
|
222
|
+
* - new_is_outlier (out)
|
223
|
+
* Whether any of the categories with zero count would be flagged as outlier if they appeared in this group.
|
224
|
+
* - categ_maj (out)
|
225
|
+
* Category to which the majority of the observations belong.
|
226
|
+
*/
|
227
|
+
void find_outlier_categories_by_maj(size_t categ_counts[], size_t ncateg, size_t tot, double max_perc_outliers,
|
228
|
+
long double prior_prob[], double z_outlier, char is_outlier[],
|
229
|
+
bool *found_outliers, bool *new_is_outlier, int *categ_maj)
|
230
|
+
{
|
231
|
+
/* initialize parameters as needed */
|
232
|
+
*found_outliers = false;
|
233
|
+
*new_is_outlier = false;
|
234
|
+
memset(is_outlier, 0, ncateg * sizeof(char));
|
235
|
+
size_t max_outliers = (size_t) calculate_max_outliers((long double)tot, max_perc_outliers);
|
236
|
+
long double tot_dbl = (long double) (tot + 1);
|
237
|
+
size_t n_non_maj;
|
238
|
+
long double thr_prop = (double)1 / square(z_outlier);
|
239
|
+
|
240
|
+
/* check if any can be considered as outlier */
|
241
|
+
size_t *ptr_maj = std::max_element(categ_counts, categ_counts + ncateg);
|
242
|
+
*categ_maj = (int)(ptr_maj - categ_counts);
|
243
|
+
n_non_maj = tot - *ptr_maj;
|
244
|
+
if (n_non_maj > max_outliers)
|
245
|
+
return;
|
246
|
+
|
247
|
+
/* determine proportions and check for outlierness */
|
248
|
+
long double n_non_maj_dbl = (long double) n_non_maj;
|
249
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
250
|
+
|
251
|
+
if ((int)cat == *categ_maj) continue;
|
252
|
+
|
253
|
+
if ( (n_non_maj_dbl / (tot_dbl * prior_prob[cat])) < thr_prop ) {
|
254
|
+
if (categ_counts[cat]) {
|
255
|
+
is_outlier[cat] = 1;
|
256
|
+
*found_outliers = true;
|
257
|
+
} else {
|
258
|
+
is_outlier[cat] = -1;
|
259
|
+
*new_is_outlier = true;
|
260
|
+
}
|
261
|
+
}
|
262
|
+
}
|
263
|
+
|
264
|
+
/* TODO: implement formula for flagging unsen categories (not in the sample, nor the full data) as outliers */
|
265
|
+
}
|
266
|
+
|
267
|
+
|
268
|
+
/* Check whether to consider any category as outlier before splitting, based on prior counts
|
269
|
+
*
|
270
|
+
* Follows very rough criteria: there can be at most 1-3 outliers depending on size of dataset,
|
271
|
+
* and the next most common category must have a count of at least 250.
|
272
|
+
*
|
273
|
+
* Parameters:
|
274
|
+
* - categ_counts[ncateg] (in)
|
275
|
+
* Frequencies of each category in the full data.
|
276
|
+
* - ncateg (in)
|
277
|
+
* Number of categories with non-zero count.
|
278
|
+
* - tot (in)
|
279
|
+
* Number of rows.
|
280
|
+
* - is_outlier[ncateg] (out)
|
281
|
+
* Array indicating whether any category is outlier (0 = non-outlier, 1 = outlier).
|
282
|
+
* - next_most_comm (out)
|
283
|
+
* Proportion of the least common non-outlier category.
|
284
|
+
*/
|
285
|
+
bool find_outlier_categories_no_cond(size_t categ_counts[], size_t ncateg, size_t tot,
|
286
|
+
char is_outlier[], double *next_most_comm)
|
287
|
+
{
|
288
|
+
/* if sample is too small, don't flag any as outliers */
|
289
|
+
if (tot < 1000) return false;
|
290
|
+
|
291
|
+
/* set a very low outlier threshold with a hard limit of 3 */
|
292
|
+
size_t max_outliers = (tot < 10000)? 1 : ((tot < 100000)? 2 : 3);
|
293
|
+
|
294
|
+
/* will only consider a category as outlier if the next most common is very common */
|
295
|
+
size_t max_next_most_comm = 250;
|
296
|
+
|
297
|
+
/* look if there's any category meeting the first condition and none meeting the second one */
|
298
|
+
bool has_outlier_cat = false;
|
299
|
+
memset(is_outlier, 0, sizeof(char) * ncateg);
|
300
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
301
|
+
if (categ_counts[cat] > max_outliers && categ_counts[cat] < max_next_most_comm) {
|
302
|
+
has_outlier_cat = false;
|
303
|
+
break;
|
304
|
+
}
|
305
|
+
|
306
|
+
if (categ_counts[cat] > 0 && categ_counts[cat] <= max_outliers) {
|
307
|
+
/* can only have 1 outlier category in the whole column */
|
308
|
+
if (has_outlier_cat) { has_outlier_cat = false; break; }
|
309
|
+
|
310
|
+
has_outlier_cat = true;
|
311
|
+
is_outlier[cat] = 1;
|
312
|
+
}
|
313
|
+
|
314
|
+
}
|
315
|
+
|
316
|
+
/* if outlier is found, find next most common frequency for printed statistics */
|
317
|
+
if (has_outlier_cat) {
|
318
|
+
size_t next_most_comm_cat = INT_MAX;
|
319
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
320
|
+
if (categ_counts[cat] > 0 && !is_outlier[cat]) {
|
321
|
+
next_most_comm_cat = std::min(next_most_comm_cat, categ_counts[cat]);
|
322
|
+
}
|
323
|
+
}
|
324
|
+
*next_most_comm = (long double)next_most_comm_cat / (long double)tot;
|
325
|
+
}
|
326
|
+
|
327
|
+
return has_outlier_cat;
|
328
|
+
}
|
@@ -0,0 +1,972 @@
|
|
1
|
+
/********************************************************************************************************************
|
2
|
+
* Explainable outlier detection
|
3
|
+
*
|
4
|
+
* Tries to detect outliers by generating decision trees that attempt to predict the values of each column based on
|
5
|
+
* each other column, testing in each branch of every tried split (if it meets some minimum criteria) whether there
|
6
|
+
* are observations that seem too distant from the others in a 1-D distribution for the column that the split tries
|
7
|
+
* to "predict" (will not generate a score for each observation).
|
8
|
+
* Splits are based on gain, while outlierness is based on confidence intervals.
|
9
|
+
* Similar in spirit to the GritBot software developed by RuleQuest research. Reference article is:
|
10
|
+
* Cortes, David. "Explainable outlier detection through decision tree conditioning."
|
11
|
+
* arXiv preprint arXiv:2001.00636 (2020).
|
12
|
+
*
|
13
|
+
*
|
14
|
+
* Copyright 2020 David Cortes.
|
15
|
+
*
|
16
|
+
* Written for C++11 standard and OpenMP 2.0 or later. Code is meant to be wrapped into scripting languages
|
17
|
+
* such as R or Python.
|
18
|
+
*
|
19
|
+
* This file is part of OutlierTree.
|
20
|
+
*
|
21
|
+
* OutlierTree is free software: you can redistribute it and/or modify
|
22
|
+
* it under the terms of the GNU General Public License as published by
|
23
|
+
* the Free Software Foundation, either version 3 of the License, or
|
24
|
+
* (at your option) any later version.
|
25
|
+
*
|
26
|
+
* OutlierTree is distributed in the hope that it will be useful,
|
27
|
+
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
28
|
+
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
29
|
+
* GNU General Public License for more details.
|
30
|
+
*
|
31
|
+
* You should have received a copy of the GNU General Public License
|
32
|
+
* along with OutlierTree. If not, see <https://www.gnu.org/licenses/>.
|
33
|
+
********************************************************************************************************************/
|
34
|
+
#include "outlier_tree.hpp"
|
35
|
+
|
36
|
+
|
37
|
+
|
38
|
+
/* Characterize a homogenous 1-dimensional cluster
|
39
|
+
*
|
40
|
+
* Calcualtes limits and display statistics on the distribution of one numerical variable,
|
41
|
+
* flagging potential outliers if found. Can be run on the full data or on subsets obtained from splitting
|
42
|
+
* by other variables.
|
43
|
+
*
|
44
|
+
* In order to flag an observation as outlier, it must:
|
45
|
+
* * Be in a very small/large percentile of the subset passed here.
|
46
|
+
* * Have a large absolute Z value (standardized and centered).
|
47
|
+
* * Have a large gap in the Z value with respect to the next largest/smallest ovservation.
|
48
|
+
* * Not be in a long tail (unless the variable was transformed by exponentiating or taking logarithm).
|
49
|
+
*
|
50
|
+
* Parameters:
|
51
|
+
* - x[n] (in)
|
52
|
+
* Variable for which to define the cluster.
|
53
|
+
* - ix_arr[n] (in)
|
54
|
+
* Indices to take from the array above.
|
55
|
+
* - st (in)
|
56
|
+
* Position at which ix_arr starts (inclusive).
|
57
|
+
* - end (in)
|
58
|
+
* Position at which ix_arr ends (inclusive).
|
59
|
+
* - outlier_scores[n] (in, out)
|
60
|
+
* Outlier scores (based on chebyshyov's inequality) that are already assigned to the observations from this column
|
61
|
+
* from previous runs of this function in larger subsets (should be started to 1).
|
62
|
+
* - outlier_clusters[n] (in, out)
|
63
|
+
* Cluster number under which an observation is the most anomalous.
|
64
|
+
* - outlier_trees[n] (in, out)
|
65
|
+
* Tree under which the outlier cluster assigned lies.
|
66
|
+
* - outlier_depth[n] (in, out)
|
67
|
+
* Tree depth at which the outlier cluster assigned is found.
|
68
|
+
* - cluster (in, out)
|
69
|
+
* Outlier cluster object with statistics and limits.
|
70
|
+
* - clusters (in)
|
71
|
+
* Vector containing all cluster already generated.
|
72
|
+
* - cluster_num (in)
|
73
|
+
* Number to give to this cluster.
|
74
|
+
* - tree_num (in)
|
75
|
+
* Number of the tree under which this cluster is to be found.
|
76
|
+
* - tree_depth (in)
|
77
|
+
* Distance form the tree root at which this tree is to be found.
|
78
|
+
* - is_log_transf (in)
|
79
|
+
* Whether the column 'x' has undergone a logarithmic transformation.
|
80
|
+
* - log_minval (in)
|
81
|
+
* Value that was added to 'x' before taking its logarithm (if it was log-transformed).
|
82
|
+
* - is_exp_transf (in)
|
83
|
+
* Whether the column 'x' has undergone an exponential transformation on its standardized values.
|
84
|
+
* - orig_mean (in)
|
85
|
+
* Mean of the variable 'x' before being standardized (if it was exponentiated).
|
86
|
+
* - orig_sd (in)
|
87
|
+
* Standard deviation of the variable 'x' before being standardized (if it was exponentiated).
|
88
|
+
* - left_tail (in)
|
89
|
+
* Value of 'x' after which it is considered a long tail, in which outliers will not be searched for.
|
90
|
+
* - rught_tail (in)
|
91
|
+
* Value of 'x' before which it is considered a long tail, in which outliers will not be searched for.
|
92
|
+
* - orig_x (in)
|
93
|
+
* Original values of 'x' if it was transformed (log or exp).
|
94
|
+
* - max_perc_outliers (in)
|
95
|
+
* Model parameter. Default is 0.01.
|
96
|
+
* - z_norm (in)
|
97
|
+
* Model parameter. Default is 2.67.
|
98
|
+
* - z_outlier (in)
|
99
|
+
* Model parameter. Default is 8.0. Must be greater than z_norm.
|
100
|
+
*
|
101
|
+
* Returns:
|
102
|
+
* - Whether there were any outliers detected.
|
103
|
+
*/
|
104
|
+
bool define_numerical_cluster(double *restrict x, size_t *restrict ix_arr, size_t st, size_t end,
|
105
|
+
double *restrict outlier_scores, size_t *restrict outlier_clusters, size_t *restrict outlier_trees,
|
106
|
+
size_t *restrict outlier_depth, Cluster &cluster, std::vector<Cluster> &clusters,
|
107
|
+
size_t cluster_num, size_t tree_num, size_t tree_depth,
|
108
|
+
bool is_log_transf, double log_minval, bool is_exp_transf, double orig_mean, double orig_sd,
|
109
|
+
double left_tail, double right_tail, double *restrict orig_x,
|
110
|
+
double max_perc_outliers, double z_norm, double z_outlier)
|
111
|
+
{
|
112
|
+
|
113
|
+
/* TODO: this function could try to determine if the distribution is multimodal, and if so,
|
114
|
+
take only the most extreme means/sd for outlier comparisons */
|
115
|
+
|
116
|
+
/* TODO: statistics like SD, mean; are already available from the splitting function which
|
117
|
+
is called right before this, so these should *only* need to be recalculated them if the column
|
118
|
+
has undergone log or exp transform */
|
119
|
+
|
120
|
+
/* NAs and Inf should have already been removed, and outliers with fewer conditionals already discarded */
|
121
|
+
bool has_low_values = false;
|
122
|
+
bool has_high_values = false;
|
123
|
+
long double running_mean = 0;
|
124
|
+
long double mean_prev = 0;
|
125
|
+
long double running_ssq = 0;
|
126
|
+
double xval;
|
127
|
+
double mean;
|
128
|
+
double sd;
|
129
|
+
size_t cnt;
|
130
|
+
size_t tail_size = (size_t) calculate_max_outliers((long double)(end - st + 1), max_perc_outliers);
|
131
|
+
size_t st_non_tail = st + tail_size;
|
132
|
+
size_t end_non_tail = end - tail_size;
|
133
|
+
size_t st_normals = 0;
|
134
|
+
size_t end_normals = 0;
|
135
|
+
double min_gap = z_outlier - z_norm;
|
136
|
+
|
137
|
+
/* TODO: here it's not necessary to sort the whole data, only top/bottom N */
|
138
|
+
|
139
|
+
/* sort the data */
|
140
|
+
std::sort(ix_arr + st, ix_arr + end + 1, [&x](const size_t a, const size_t b){return x[a] < x[b];});
|
141
|
+
|
142
|
+
/* calculate statistics with tails and previous outliers excluded */
|
143
|
+
cnt = end_non_tail - st_non_tail + 1;
|
144
|
+
for (size_t row = st_non_tail; row <= end_non_tail; row++) {
|
145
|
+
xval = x[ ix_arr[row] ];
|
146
|
+
running_mean += (xval - running_mean) / (long double)(row - st_non_tail + 1);
|
147
|
+
running_ssq += (xval - running_mean) * (xval - mean_prev);
|
148
|
+
mean_prev = running_mean;
|
149
|
+
|
150
|
+
}
|
151
|
+
mean = (double) running_mean;
|
152
|
+
sd = (double) sqrtl(running_ssq / (long double)(cnt - 1));
|
153
|
+
|
154
|
+
/* adjust SD heuristically to account for reduced size, by (N + tail)/(N-tail) --- note that cnt = N-2*tail */
|
155
|
+
sd *= (long double)(cnt + 3 * tail_size) / (long double)(cnt + tail_size);
|
156
|
+
/* re-adjust if there's a one-sided tail and no transformation was applies */
|
157
|
+
if ((!isinf(left_tail) || !isinf(right_tail)) && !is_log_transf && !is_exp_transf) {
|
158
|
+
sd *= 0.5;
|
159
|
+
}
|
160
|
+
cluster.cluster_mean = mean;
|
161
|
+
cluster.cluster_sd = sd;
|
162
|
+
cnt = end - st + 1;
|
163
|
+
|
164
|
+
/* see if the minimum and/or maximum values qualify for outliers */
|
165
|
+
if (-z_score(x[ix_arr[st]], mean, sd) >= z_outlier && x[ix_arr[st]] > left_tail) has_low_values = true;
|
166
|
+
if ( z_score(x[ix_arr[end]], mean, sd) >= z_outlier && x[ix_arr[end]] < right_tail) has_high_values = true;
|
167
|
+
|
168
|
+
/* look for a large gap in the z-scores */
|
169
|
+
if (has_low_values) {
|
170
|
+
for (size_t row = st; row < st + tail_size; row++) {
|
171
|
+
|
172
|
+
if (( z_score(x[ix_arr[row + 1]], mean, sd) - z_score(x[ix_arr[row]], mean, sd) ) >= min_gap) {
|
173
|
+
st_normals = row + 1;
|
174
|
+
if (is_exp_transf) {
|
175
|
+
cluster.lower_lim = log(x[ix_arr[row + 1]] - min_gap * sd) * orig_sd + orig_mean;
|
176
|
+
} else if (is_log_transf) {
|
177
|
+
cluster.lower_lim = exp(x[ix_arr[row + 1]] - min_gap * sd) + log_minval;
|
178
|
+
} else {
|
179
|
+
cluster.lower_lim = x[ix_arr[row + 1]] - min_gap * sd;
|
180
|
+
}
|
181
|
+
cluster.display_lim_low = orig_x[ix_arr[row + 1]];
|
182
|
+
cluster.perc_above = (long double)(end - st_normals + 1) / (long double)(end - st + 1);
|
183
|
+
break;
|
184
|
+
}
|
185
|
+
if (z_score(x[ix_arr[row]], mean, sd) > -z_outlier) break;
|
186
|
+
|
187
|
+
}
|
188
|
+
if (st_normals == 0) {
|
189
|
+
has_low_values = false;
|
190
|
+
} else {
|
191
|
+
for (size_t row = st; row < st_normals; row++) {
|
192
|
+
|
193
|
+
/* assign outlier if it's a better cluster than previously assigned */
|
194
|
+
if (
|
195
|
+
outlier_scores[ix_arr[row]] >= 1.0 ||
|
196
|
+
(clusters[outlier_clusters[ix_arr[row]]].has_NA_branch && !cluster.has_NA_branch) ||
|
197
|
+
(
|
198
|
+
cluster.has_NA_branch == clusters[outlier_clusters[ix_arr[row]]].has_NA_branch
|
199
|
+
&&
|
200
|
+
(
|
201
|
+
tree_depth < outlier_depth[ix_arr[row]] ||
|
202
|
+
(
|
203
|
+
tree_depth == outlier_depth[ix_arr[row]] &&
|
204
|
+
clusters[outlier_clusters[ix_arr[row]]].cluster_size < (cnt - 2 * tail_size)
|
205
|
+
)
|
206
|
+
)
|
207
|
+
)
|
208
|
+
)
|
209
|
+
{
|
210
|
+
outlier_scores[ix_arr[row]] = chebyshyov_bound(z_score(x[ix_arr[row]], mean, sd));
|
211
|
+
if (is_na_or_inf(outlier_scores[ix_arr[row]])) outlier_scores[ix_arr[row]] = 0;
|
212
|
+
outlier_clusters[ix_arr[row]] = cluster_num;
|
213
|
+
outlier_trees[ix_arr[row]] = tree_num;
|
214
|
+
outlier_depth[ix_arr[row]] = tree_depth;
|
215
|
+
}
|
216
|
+
|
217
|
+
}
|
218
|
+
}
|
219
|
+
}
|
220
|
+
if (!has_low_values) {
|
221
|
+
cluster.perc_above = 1.0;
|
222
|
+
if (!is_log_transf && !is_exp_transf) {
|
223
|
+
|
224
|
+
if (isinf(left_tail)) {
|
225
|
+
cluster.lower_lim = x[ix_arr[st]] - min_gap * sd;
|
226
|
+
} else {
|
227
|
+
cluster.lower_lim = -HUGE_VAL;
|
228
|
+
}
|
229
|
+
|
230
|
+
} else if (is_exp_transf) {
|
231
|
+
cluster.lower_lim = log(x[ix_arr[st]] - min_gap * sd) * orig_sd + orig_mean;
|
232
|
+
} else {
|
233
|
+
cluster.lower_lim = exp(x[ix_arr[st]] - min_gap * sd) + log_minval;
|
234
|
+
}
|
235
|
+
|
236
|
+
cluster.display_lim_low = orig_x[ix_arr[st]];
|
237
|
+
|
238
|
+
}
|
239
|
+
|
240
|
+
if (has_high_values) {
|
241
|
+
for (size_t row = end; row > (end - tail_size); row--) {
|
242
|
+
|
243
|
+
if (( z_score(x[ix_arr[row]], mean, sd) - z_score(x[ix_arr[row - 1]], mean, sd) ) >= min_gap) {
|
244
|
+
end_normals = row - 1;
|
245
|
+
if (is_exp_transf) {
|
246
|
+
cluster.upper_lim = log(x[ix_arr[row - 1]] + min_gap * sd) * orig_sd + orig_mean;
|
247
|
+
} else if (is_log_transf) {
|
248
|
+
cluster.upper_lim = exp(x[ix_arr[row - 1]] + min_gap * sd) + log_minval;
|
249
|
+
} else {
|
250
|
+
cluster.upper_lim = x[ix_arr[row - 1]] + min_gap * sd;
|
251
|
+
}
|
252
|
+
cluster.display_lim_high = orig_x[ix_arr[row - 1]];
|
253
|
+
cluster.perc_below = (long double)(end_normals - st + 1) / (long double)(end - st + 1);
|
254
|
+
break;
|
255
|
+
}
|
256
|
+
if (z_score(x[ix_arr[row]], mean, sd) < z_outlier) break;
|
257
|
+
|
258
|
+
}
|
259
|
+
if (end_normals == 0) {
|
260
|
+
has_high_values = false;
|
261
|
+
} else {
|
262
|
+
for (size_t row = end; row > end_normals; row--) {
|
263
|
+
|
264
|
+
/* assign outlier if it's a better cluster than previously assigned - Note that it might produce slight mismatches
|
265
|
+
against the predict function (the latter is more trustable) due to the size of the cluster not yet being known
|
266
|
+
at the moment of determinining whether to overwrite previous in here */
|
267
|
+
if (
|
268
|
+
outlier_scores[ix_arr[row]] >= 1.0 ||
|
269
|
+
(clusters[outlier_clusters[ix_arr[row]]].has_NA_branch && !cluster.has_NA_branch) ||
|
270
|
+
(
|
271
|
+
cluster.has_NA_branch == clusters[outlier_clusters[ix_arr[row]]].has_NA_branch
|
272
|
+
&&
|
273
|
+
(
|
274
|
+
tree_depth < outlier_depth[ix_arr[row]] ||
|
275
|
+
(
|
276
|
+
tree_depth == outlier_depth[ix_arr[row]] &&
|
277
|
+
clusters[outlier_clusters[ix_arr[row]]].cluster_size < (cnt - 2 * tail_size)
|
278
|
+
)
|
279
|
+
)
|
280
|
+
)
|
281
|
+
)
|
282
|
+
{
|
283
|
+
outlier_scores[ix_arr[row]] = chebyshyov_bound(z_score(x[ix_arr[row]], mean, sd));
|
284
|
+
if (is_na_or_inf(outlier_scores[ix_arr[row]])) outlier_scores[ix_arr[row]] = 0;
|
285
|
+
outlier_clusters[ix_arr[row]] = cluster_num;
|
286
|
+
outlier_trees[ix_arr[row]] = tree_num;
|
287
|
+
outlier_depth[ix_arr[row]] = tree_depth;
|
288
|
+
}
|
289
|
+
|
290
|
+
}
|
291
|
+
}
|
292
|
+
}
|
293
|
+
if (!has_high_values) {
|
294
|
+
cluster.perc_below = 1.0;
|
295
|
+
if (!is_log_transf && !is_exp_transf) {
|
296
|
+
|
297
|
+
if (isinf(right_tail)) {
|
298
|
+
cluster.upper_lim = x[ix_arr[end]] + min_gap * sd;
|
299
|
+
} else {
|
300
|
+
cluster.upper_lim = HUGE_VAL;
|
301
|
+
}
|
302
|
+
} else if (is_exp_transf) {
|
303
|
+
cluster.upper_lim = log(x[ix_arr[end]] + min_gap * sd) * orig_sd + orig_mean;
|
304
|
+
} else {
|
305
|
+
cluster.upper_lim = exp(x[ix_arr[end]] + min_gap * sd) + log_minval;
|
306
|
+
}
|
307
|
+
|
308
|
+
cluster.display_lim_high = orig_x[ix_arr[end]];
|
309
|
+
}
|
310
|
+
|
311
|
+
/* save displayed statistics for cluster */
|
312
|
+
if (has_high_values || has_low_values || is_log_transf || is_exp_transf) {
|
313
|
+
size_t st_disp = has_low_values? st_normals : st;
|
314
|
+
size_t end_disp = has_high_values? end_normals : end;
|
315
|
+
running_mean = 0;
|
316
|
+
mean_prev = 0;
|
317
|
+
running_ssq = 0;
|
318
|
+
for (size_t row = st_disp; row <= end_disp; row++) {
|
319
|
+
xval = orig_x[ix_arr[row]];
|
320
|
+
running_mean += (xval - running_mean) / (long double)(row - st_disp + 1);
|
321
|
+
running_ssq += (xval - running_mean) * (xval - mean_prev);
|
322
|
+
mean_prev = running_mean;
|
323
|
+
}
|
324
|
+
cluster.cluster_size = end_disp - st_disp + 1;
|
325
|
+
cluster.display_mean = (double) running_mean;
|
326
|
+
cluster.display_sd = (double) sqrtl(running_ssq / (long double)(cluster.cluster_size - 1));
|
327
|
+
} else {
|
328
|
+
cluster.display_mean = cluster.cluster_mean;
|
329
|
+
cluster.display_sd = cluster.cluster_sd;
|
330
|
+
cluster.cluster_size = end - st + 1;
|
331
|
+
}
|
332
|
+
|
333
|
+
/* report whether outliers were found or not */
|
334
|
+
return has_low_values || has_high_values;
|
335
|
+
}
|
336
|
+
|
337
|
+
|
338
|
+
/* Characterize a homogeneous categorical cluster from the *full* data
|
339
|
+
*
|
340
|
+
* Function is meant for the data as it comes, before splitting it, as once split, it will
|
341
|
+
* not be able to detect these outliers. As such, it takes fewer parameters, since it can only
|
342
|
+
* be the first tree and cluster in a column. It assumes the outliers have already been identified.
|
343
|
+
*
|
344
|
+
* Parameters:
|
345
|
+
* - x[n]
|
346
|
+
* Array indicating the category to which each observation belongs.
|
347
|
+
* - ix_arr[n] (in)
|
348
|
+
* Indices to take from the array above.
|
349
|
+
* - st (in)
|
350
|
+
* Position at which ix_arr starts (inclusive).
|
351
|
+
* - end (in)
|
352
|
+
* Position at which ix_arr ends (inclusive).
|
353
|
+
* - ncateg (in)
|
354
|
+
* Number of categories in this column.
|
355
|
+
* - outlier_scores[n] (in, out)
|
356
|
+
* Array where to assign outlier scores (based on proportion) to each observation belonging to an outlier category.
|
357
|
+
* - outlier_clusters[n] (in, out)
|
358
|
+
* Array where to assign cluster number to each observation belonging to an outlier category.
|
359
|
+
* - outlier_trees[n] (in, out)
|
360
|
+
* Array where to assign tree number to each observation belonging to an outlier category.
|
361
|
+
* - outlier_depth[n] (in, out)
|
362
|
+
* Array where to assign tree depth to each observation belonging to an outlier category.
|
363
|
+
* - cluster (in, out)
|
364
|
+
* Outlier cluster object with statistics and classifications.
|
365
|
+
* - categ_counts[ncateg] (in)
|
366
|
+
* Array with the frequencies of each category in the data.
|
367
|
+
* - is_outlier[ncateg] (in)
|
368
|
+
* Array indicating which categories are to be considered as outliers (must be already calculated).
|
369
|
+
* - perc_next_most_comm (in)
|
370
|
+
* Proportion of the least common non-outlier category (must be already calculated).
|
371
|
+
*/
|
372
|
+
void define_categ_cluster_no_cond(int *restrict x, size_t *restrict ix_arr, size_t st, size_t end, size_t ncateg,
|
373
|
+
double *restrict outlier_scores, size_t *restrict outlier_clusters, size_t *restrict outlier_trees,
|
374
|
+
size_t *restrict outlier_depth, Cluster &cluster,
|
375
|
+
size_t *restrict categ_counts, char *restrict is_outlier, double perc_next_most_comm)
|
376
|
+
{
|
377
|
+
size_t cnt_common = end - st + 1;
|
378
|
+
cluster.cluster_size = cnt_common;
|
379
|
+
double pct_outl;
|
380
|
+
cluster.subset_common.assign(is_outlier, is_outlier + ncateg);
|
381
|
+
cluster.score_categ.resize(ncateg, 0);
|
382
|
+
|
383
|
+
|
384
|
+
for (size_t row = st; row <= end; row++) {
|
385
|
+
if (is_outlier[x[ix_arr[row]]]) {
|
386
|
+
cnt_common--;
|
387
|
+
pct_outl = (long double)categ_counts[ x[ix_arr[row]] ] / (long double)cluster.cluster_size;
|
388
|
+
pct_outl = pct_outl + sqrt(pct_outl * (1 - pct_outl) / (long double)cluster.cluster_size);
|
389
|
+
cluster.score_categ[ x[ix_arr[row]] ] = pct_outl;
|
390
|
+
outlier_scores[ix_arr[row]] = pct_outl;
|
391
|
+
outlier_clusters[ix_arr[row]] = 0;
|
392
|
+
outlier_trees[ix_arr[row]] = 0;
|
393
|
+
outlier_depth[ix_arr[row]] = 0;
|
394
|
+
}
|
395
|
+
}
|
396
|
+
cluster.perc_in_subset = (long double)cnt_common / (long double)cluster.cluster_size;
|
397
|
+
cluster.perc_next_most_comm = perc_next_most_comm;
|
398
|
+
}
|
399
|
+
|
400
|
+
|
401
|
+
/* Characterize a homogeneous categorical cluster form a subset of the data, or report if it's not homogeneous
|
402
|
+
*
|
403
|
+
* Function is meant to be called with subsets of the data only. Will calculate the counts inside it.
|
404
|
+
* In order to consider a category as outlier, it must:
|
405
|
+
* * Have a proportion smaller than its prior probability and than a condifence interval of its prior.
|
406
|
+
* * Have a large gap with respect to the next most-common category.
|
407
|
+
* * Be in a cluster in which few or no observations belong to a category meeting such conditions.
|
408
|
+
* It's oftentimes not possible to create a cluster with category frequencies that would produce outliers,
|
409
|
+
* in which case it will report whether the cluster should be dropped.
|
410
|
+
*
|
411
|
+
* Parameters:
|
412
|
+
* - x[n]
|
413
|
+
* Array indicating the category to which each observation belongs.
|
414
|
+
* - ix_arr[n] (in)
|
415
|
+
* Indices to take from the array above.
|
416
|
+
* - st (in)
|
417
|
+
* Position at which ix_arr starts (inclusive).
|
418
|
+
* - end (in)
|
419
|
+
* Position at which ix_arr ends (inclusive).
|
420
|
+
* - ncateg (in)
|
421
|
+
* Number of categories in this column.
|
422
|
+
* - by_maj (in)
|
423
|
+
* Model parameter. Default is 'false'. Indicates whether to detect outliers according to the number of non-majority
|
424
|
+
* obsevations compared to the expected number for each category.
|
425
|
+
* - outlier_scores[n] (in, out)
|
426
|
+
* Outlier scores (based on observed category proportion) that are already assigned to the observations from this column
|
427
|
+
* from previous runs of this function in larger subsets (should be started to 1).
|
428
|
+
* - outlier_clusters[n] (in, out)
|
429
|
+
* Cluster number under which an observation is the most anomalous.
|
430
|
+
* - outlier_trees[n] (in, out)
|
431
|
+
* Tree under which the outlier cluster assigned lies.
|
432
|
+
* - outlier_depth[n] (in, out)
|
433
|
+
* Tree depth at which the outlier cluster assigned is found.
|
434
|
+
* - cluster (in, out)
|
435
|
+
* Outlier cluster object with statistics and limits.
|
436
|
+
* - clusters (in)
|
437
|
+
* Vector containing all cluster already generated.
|
438
|
+
* - cluster_num (in)
|
439
|
+
* Number to give to this cluster.
|
440
|
+
* - tree_num (in)
|
441
|
+
* Number of the tree under which this cluster is to be found.
|
442
|
+
* - tree_depth (in)
|
443
|
+
* Distance form the tree root at which this tree is to be found.
|
444
|
+
* - max_perc_outliers (in)
|
445
|
+
* Model parameter. Default is 0.01.
|
446
|
+
* - z_norm (in)
|
447
|
+
* Model parameter. Default is 2.67.
|
448
|
+
* - z_outlier (in)
|
449
|
+
* Model parameter. Default is 8.0.
|
450
|
+
* - perc_threshold[ncateg] (in)
|
451
|
+
* Observed proportion below which a category can be considered as outlier.
|
452
|
+
* - prop_prior[ncateg] (in)
|
453
|
+
* Prior probability of each category in the full data (only used when passing 'by_maj' = 'true').
|
454
|
+
* - buffer_categ_counts[ncateg] (temp)
|
455
|
+
* Buffer where to save the observed frequencies of each category.
|
456
|
+
* - buffer_categ_pct[ncateg] (temp)
|
457
|
+
* Buffer where to save the observed proportion of each category.
|
458
|
+
* - buffer_categ_ix[ncateg] (temp)
|
459
|
+
* Buffer where to save the category numbers sorted by proportion.
|
460
|
+
* - buffer_outliers[ncateg] (temp)
|
461
|
+
* Buffer where to save the results of which categories are flagged as outliers
|
462
|
+
* before copying it to the cluster (will not copy if none is flagged).
|
463
|
+
* - drop_cluster (out)
|
464
|
+
* Whethet the cluster should be dropped (i.e. it was not possible to flag any present
|
465
|
+
* or non-present category as outlier).
|
466
|
+
*
|
467
|
+
* Returns:
|
468
|
+
* - Whether it identified any outliers or not.
|
469
|
+
*/
|
470
|
+
bool define_categ_cluster(int *restrict x, size_t *restrict ix_arr, size_t st, size_t end, size_t ncateg, bool by_maj,
|
471
|
+
double *restrict outlier_scores, size_t *restrict outlier_clusters, size_t *restrict outlier_trees,
|
472
|
+
size_t *restrict outlier_depth, Cluster &cluster, std::vector<Cluster> &clusters,
|
473
|
+
size_t cluster_num, size_t tree_num, size_t tree_depth,
|
474
|
+
double max_perc_outliers, double z_norm, double z_outlier,
|
475
|
+
long double *restrict perc_threshold, long double *restrict prop_prior,
|
476
|
+
size_t *restrict buffer_categ_counts, long double *restrict buffer_categ_pct,
|
477
|
+
size_t *restrict buffer_categ_ix, char *restrict buffer_outliers,
|
478
|
+
bool *restrict drop_cluster)
|
479
|
+
{
|
480
|
+
bool found_outliers, new_is_outlier;
|
481
|
+
size_t tot = end - st + 1;
|
482
|
+
size_t sz_maj = tot;
|
483
|
+
long double tot_dbl = (long double) tot;
|
484
|
+
size_t tail_size = (size_t) calculate_max_outliers(tot_dbl, max_perc_outliers);
|
485
|
+
cluster.perc_in_subset = 1;
|
486
|
+
double pct_outl;
|
487
|
+
|
488
|
+
/* calculate category counts */
|
489
|
+
memset(buffer_categ_counts, 0, ncateg * sizeof(size_t));
|
490
|
+
for (size_t row = st; row <= end; row++) {
|
491
|
+
buffer_categ_counts[ x[ix_arr[row]] ]++;
|
492
|
+
}
|
493
|
+
|
494
|
+
/* flag categories as outliers if appropriate */
|
495
|
+
if (!by_maj)
|
496
|
+
find_outlier_categories(buffer_categ_counts, ncateg, tot, max_perc_outliers,
|
497
|
+
perc_threshold, buffer_categ_ix, buffer_categ_pct,
|
498
|
+
z_norm, buffer_outliers, &found_outliers,
|
499
|
+
&new_is_outlier, &cluster.perc_next_most_comm);
|
500
|
+
else
|
501
|
+
find_outlier_categories_by_maj(buffer_categ_counts, ncateg, tot, max_perc_outliers,
|
502
|
+
prop_prior, z_outlier, buffer_outliers,
|
503
|
+
&found_outliers, &new_is_outlier, &cluster.categ_maj);
|
504
|
+
|
505
|
+
if (found_outliers) {
|
506
|
+
for (size_t row = st; row <= end; row++) {
|
507
|
+
if (buffer_outliers[ x[ix_arr[row]] ]) {
|
508
|
+
|
509
|
+
/* follow usual rules for preferring this cluster over others */
|
510
|
+
if (
|
511
|
+
outlier_scores[ix_arr[row]] >= 1.0 ||
|
512
|
+
(clusters[outlier_clusters[ix_arr[row]]].has_NA_branch && !cluster.has_NA_branch) ||
|
513
|
+
(
|
514
|
+
cluster.has_NA_branch == clusters[outlier_clusters[ix_arr[row]]].has_NA_branch
|
515
|
+
&&
|
516
|
+
(
|
517
|
+
tree_depth < outlier_depth[ix_arr[row]] ||
|
518
|
+
(
|
519
|
+
tree_depth == outlier_depth[ix_arr[row]] &&
|
520
|
+
clusters[outlier_clusters[ix_arr[row]]].cluster_size < (tot - tail_size)
|
521
|
+
)
|
522
|
+
)
|
523
|
+
)
|
524
|
+
)
|
525
|
+
{
|
526
|
+
if (!by_maj) {
|
527
|
+
pct_outl = (long double)buffer_categ_counts[ x[ix_arr[row]] ] / tot_dbl;
|
528
|
+
pct_outl = pct_outl + sqrt(pct_outl * (1 - pct_outl) / tot_dbl);
|
529
|
+
outlier_scores[ix_arr[row]] = pct_outl;
|
530
|
+
} else {
|
531
|
+
pct_outl = (long double)(tot - buffer_categ_counts[cluster.categ_maj]) / (tot_dbl * prop_prior[ x[ix_arr[row]] ]);
|
532
|
+
outlier_scores[ix_arr[row]] = square(pct_outl);
|
533
|
+
}
|
534
|
+
outlier_clusters[ix_arr[row]] = cluster_num;
|
535
|
+
outlier_trees[ix_arr[row]] = tree_num;
|
536
|
+
outlier_depth[ix_arr[row]] = tree_depth;
|
537
|
+
}
|
538
|
+
sz_maj--;
|
539
|
+
|
540
|
+
}
|
541
|
+
}
|
542
|
+
cluster.perc_in_subset = (long double)sz_maj / tot_dbl;
|
543
|
+
}
|
544
|
+
|
545
|
+
if (new_is_outlier && !found_outliers) {
|
546
|
+
cluster.perc_in_subset = 1.0;
|
547
|
+
}
|
548
|
+
|
549
|
+
if (new_is_outlier || found_outliers) {
|
550
|
+
*drop_cluster = false;
|
551
|
+
cluster.cluster_size = sz_maj;
|
552
|
+
cluster.subset_common.assign(buffer_outliers, buffer_outliers + ncateg);
|
553
|
+
cluster.score_categ.resize(ncateg, 0);
|
554
|
+
if (!by_maj) {
|
555
|
+
|
556
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
557
|
+
if (cluster.subset_common[cat] > 0) {
|
558
|
+
pct_outl = (long double)buffer_categ_counts[cat] / tot_dbl;
|
559
|
+
cluster.score_categ[cat] = pct_outl + sqrt(pct_outl * (1 - pct_outl) / tot_dbl);
|
560
|
+
} else if (cluster.subset_common[cat] < 0) {
|
561
|
+
pct_outl = (long double)1 / (long double)(tot + 2);
|
562
|
+
cluster.score_categ[cat] = pct_outl + sqrt(pct_outl * (1 - pct_outl) / (long double)(tot + 2));
|
563
|
+
}
|
564
|
+
}
|
565
|
+
|
566
|
+
} else {
|
567
|
+
|
568
|
+
cluster.perc_in_subset = (long double) buffer_categ_counts[cluster.categ_maj] / tot_dbl;
|
569
|
+
for (size_t cat = 0; cat < ncateg; cat++) {
|
570
|
+
if (cat == cluster.categ_maj)
|
571
|
+
continue;
|
572
|
+
if (cluster.subset_common[cat] != 0) {
|
573
|
+
cluster.score_categ[cat] = (long double)(tot - buffer_categ_counts[cluster.categ_maj] + 1)
|
574
|
+
/ ((long double)(tot + 2) * prop_prior[cat]);
|
575
|
+
cluster.score_categ[cat] = square(cluster.score_categ[cat]);
|
576
|
+
}
|
577
|
+
}
|
578
|
+
|
579
|
+
}
|
580
|
+
} else {
|
581
|
+
*drop_cluster = true;
|
582
|
+
}
|
583
|
+
|
584
|
+
return found_outliers;
|
585
|
+
}
|
586
|
+
|
587
|
+
/* Convert in/not-in conditions to 'equals' or 'not equals' when they look for only 1 category */
|
588
|
+
void simplify_when_equal_cond(std::vector<Cluster> &clusters, int ncat_ord[])
|
589
|
+
{
|
590
|
+
|
591
|
+
int col_equal;
|
592
|
+
size_t size_subset;
|
593
|
+
size_t size_subset_excl;
|
594
|
+
for (size_t clust = 0; clust < clusters.size(); clust++) {
|
595
|
+
if (clusters[clust].split_type == IsNa) continue;
|
596
|
+
|
597
|
+
switch(clusters[clust].column_type) {
|
598
|
+
|
599
|
+
case Categorical:
|
600
|
+
{
|
601
|
+
|
602
|
+
col_equal = -1;
|
603
|
+
if (clusters[clust].split_subset.size() == 2) {
|
604
|
+
|
605
|
+
switch(col_equal = clusters[clust].split_type) {
|
606
|
+
case InSubset:
|
607
|
+
{
|
608
|
+
col_equal = clusters[clust].split_subset[0]? 0 : 1;
|
609
|
+
break;
|
610
|
+
}
|
611
|
+
|
612
|
+
case NotInSubset:
|
613
|
+
{
|
614
|
+
col_equal = clusters[clust].split_subset[0]? 1 : 0;
|
615
|
+
break;
|
616
|
+
}
|
617
|
+
|
618
|
+
case SingleCateg:
|
619
|
+
{
|
620
|
+
col_equal = clusters[clust].split_subset[0]? 0 : 1;
|
621
|
+
break;
|
622
|
+
}
|
623
|
+
}
|
624
|
+
clusters[clust].split_type = Equal;
|
625
|
+
|
626
|
+
} else {
|
627
|
+
|
628
|
+
size_subset_excl = std::accumulate(clusters[clust].split_subset.begin(), clusters[clust].split_subset.end(), (size_t)0,
|
629
|
+
[](const size_t a, const char b){return a + ((b < 0)? 1 : 0);});
|
630
|
+
if (size_subset_excl > 0) continue;
|
631
|
+
size_subset = std::accumulate(clusters[clust].split_subset.begin(), clusters[clust].split_subset.end(), (size_t)0,
|
632
|
+
[](const size_t a, const char b){return a + ((b > 0)? 1 : 0);});
|
633
|
+
if (size_subset == 1) {
|
634
|
+
|
635
|
+
do {col_equal++;} while (clusters[clust].split_subset[col_equal] <= 0);
|
636
|
+
if (clusters[clust].split_type == InSubset || clusters[clust].split_type == SingleCateg)
|
637
|
+
clusters[clust].split_type = Equal;
|
638
|
+
else
|
639
|
+
clusters[clust].split_type = NotEqual;
|
640
|
+
|
641
|
+
} else if (size_subset == (clusters[clust].split_subset.size() - 1)) {
|
642
|
+
|
643
|
+
do {col_equal++;} while (clusters[clust].split_subset[col_equal] != 0);
|
644
|
+
if (clusters[clust].split_type == NotInSubset)
|
645
|
+
clusters[clust].split_type = Equal;
|
646
|
+
else
|
647
|
+
clusters[clust].split_type = NotEqual;
|
648
|
+
|
649
|
+
}
|
650
|
+
|
651
|
+
}
|
652
|
+
if (col_equal >= 0) {
|
653
|
+
clusters[clust].split_subset.resize(0);
|
654
|
+
clusters[clust].split_lev = col_equal;
|
655
|
+
}
|
656
|
+
break;
|
657
|
+
}
|
658
|
+
|
659
|
+
|
660
|
+
case Ordinal:
|
661
|
+
{
|
662
|
+
|
663
|
+
if (clusters[clust].split_lev == 0) {
|
664
|
+
|
665
|
+
if (clusters[clust].split_type == LessOrEqual)
|
666
|
+
clusters[clust].split_type = Equal;
|
667
|
+
else
|
668
|
+
clusters[clust].split_type = NotEqual;
|
669
|
+
|
670
|
+
}
|
671
|
+
|
672
|
+
else if (clusters[clust].split_lev == (ncat_ord[clusters[clust].col_num] - 2)) {
|
673
|
+
|
674
|
+
clusters[clust].split_lev++;
|
675
|
+
if (clusters[clust].split_type == Greater)
|
676
|
+
clusters[clust].split_type = Equal;
|
677
|
+
else
|
678
|
+
clusters[clust].split_type = NotEqual;
|
679
|
+
|
680
|
+
}
|
681
|
+
break;
|
682
|
+
}
|
683
|
+
|
684
|
+
}
|
685
|
+
|
686
|
+
}
|
687
|
+
|
688
|
+
}
|
689
|
+
|
690
|
+
/*
|
691
|
+
* Convert in/not-in conditions to 'equals' when they look for only 1 category
|
692
|
+
* Note: unlike in the case of clusters, trees do not store the split type, but rather
|
693
|
+
* always assume left is in/l.e. and right the opposite, so it's not possible to
|
694
|
+
* simplify ordinal splits to equals (as the tree will not distinguish between
|
695
|
+
* an ordinal split with equals and another with l.e./g.e.). Thus, this part needs
|
696
|
+
* to be done in the function that prints the outlier conditions.
|
697
|
+
*/
|
698
|
+
void simplify_when_equal_cond(std::vector<ClusterTree> &trees, int ncat_ord[])
|
699
|
+
{
|
700
|
+
|
701
|
+
int col_equal;
|
702
|
+
size_t size_subset;
|
703
|
+
size_t size_subset_excl;
|
704
|
+
size_t temp_swap;
|
705
|
+
for (size_t tree = 0; tree < trees.size(); tree++) {
|
706
|
+
|
707
|
+
if (trees[tree].all_branches.size() == 0 && trees[tree].tree_left == 0 && trees[tree].tree_right == 0) continue;
|
708
|
+
if (trees[trees[tree].parent].all_branches.size() > 0 && trees[tree].split_this_branch == IsNa) continue;
|
709
|
+
switch(trees[tree].column_type) {
|
710
|
+
|
711
|
+
case Categorical:
|
712
|
+
{
|
713
|
+
size_subset_excl = std::accumulate(trees[tree].split_subset.begin(), trees[tree].split_subset.end(), (size_t)0,
|
714
|
+
[](const size_t a, const char b){return a + ((b < 0)? 1 : 0);});
|
715
|
+
if (size_subset_excl > 0) continue;
|
716
|
+
|
717
|
+
col_equal = -1;
|
718
|
+
if (trees[tree].split_subset.size() == 2) {
|
719
|
+
|
720
|
+
col_equal = 0;
|
721
|
+
if (trees[tree].split_subset[0] == 0) {
|
722
|
+
temp_swap = trees[tree].tree_left;
|
723
|
+
trees[tree].tree_left = trees[tree].tree_right;
|
724
|
+
trees[tree].tree_right = temp_swap;
|
725
|
+
}
|
726
|
+
if (trees[tree].tree_left > 0)
|
727
|
+
trees[trees[tree].tree_left].parent_branch = Equal;
|
728
|
+
if (trees[tree].tree_right > 0)
|
729
|
+
trees[trees[tree].tree_right].parent_branch = NotEqual;
|
730
|
+
|
731
|
+
if (trees[trees[tree].parent].all_branches.size() > 0) {
|
732
|
+
switch(trees[tree].split_this_branch) {
|
733
|
+
case InSubset:
|
734
|
+
{
|
735
|
+
trees[tree].split_this_branch = Equal;
|
736
|
+
break;
|
737
|
+
}
|
738
|
+
|
739
|
+
case NotInSubset:
|
740
|
+
{
|
741
|
+
trees[tree].split_this_branch = NotEqual;
|
742
|
+
break;
|
743
|
+
}
|
744
|
+
|
745
|
+
case SingleCateg:
|
746
|
+
{
|
747
|
+
trees[tree].split_this_branch = Equal;
|
748
|
+
break;
|
749
|
+
}
|
750
|
+
}
|
751
|
+
}
|
752
|
+
|
753
|
+
}
|
754
|
+
|
755
|
+
else {
|
756
|
+
|
757
|
+
size_subset = std::accumulate(trees[tree].split_subset.begin(), trees[tree].split_subset.end(), (size_t)0,
|
758
|
+
[](const size_t a, const char b){return a + ((b > 0)? 1 : 0);});
|
759
|
+
if (size_subset == 1) {
|
760
|
+
|
761
|
+
do {col_equal++;} while (trees[tree].split_subset[col_equal] <= 0);
|
762
|
+
if (trees[trees[tree].parent].all_branches.size() > 0) {
|
763
|
+
switch(trees[tree].split_this_branch) {
|
764
|
+
case InSubset:
|
765
|
+
{
|
766
|
+
trees[tree].split_this_branch = Equal;
|
767
|
+
break;
|
768
|
+
}
|
769
|
+
|
770
|
+
case NotInSubset:
|
771
|
+
{
|
772
|
+
trees[tree].split_this_branch = NotEqual;
|
773
|
+
break;
|
774
|
+
}
|
775
|
+
|
776
|
+
case SingleCateg:
|
777
|
+
{
|
778
|
+
trees[tree].split_this_branch = Equal;
|
779
|
+
break;
|
780
|
+
}
|
781
|
+
}
|
782
|
+
}
|
783
|
+
|
784
|
+
|
785
|
+
} else if (size_subset == (trees[tree].split_subset.size() - 1)) {
|
786
|
+
|
787
|
+
do {col_equal++;} while (trees[tree].split_subset[col_equal] != 0);
|
788
|
+
temp_swap = trees[tree].tree_left;
|
789
|
+
trees[tree].tree_left = trees[tree].tree_right;
|
790
|
+
trees[tree].tree_right = temp_swap;
|
791
|
+
if (trees[trees[tree].parent].all_branches.size() > 0) {
|
792
|
+
switch(trees[tree].split_this_branch) {
|
793
|
+
case InSubset:
|
794
|
+
{
|
795
|
+
trees[tree].split_this_branch = NotEqual;
|
796
|
+
break;
|
797
|
+
}
|
798
|
+
|
799
|
+
case NotInSubset:
|
800
|
+
{
|
801
|
+
trees[tree].split_this_branch = Equal;
|
802
|
+
break;
|
803
|
+
}
|
804
|
+
}
|
805
|
+
}
|
806
|
+
|
807
|
+
}
|
808
|
+
|
809
|
+
}
|
810
|
+
|
811
|
+
if (col_equal >= 0) {
|
812
|
+
trees[tree].split_subset.resize(0);
|
813
|
+
trees[tree].split_lev = col_equal;
|
814
|
+
if (trees[tree].tree_left > 0)
|
815
|
+
trees[trees[tree].tree_left].parent_branch = Equal;
|
816
|
+
if (trees[tree].tree_right > 0)
|
817
|
+
trees[trees[tree].tree_right].parent_branch = NotEqual;
|
818
|
+
|
819
|
+
}
|
820
|
+
break;
|
821
|
+
}
|
822
|
+
|
823
|
+
|
824
|
+
case Ordinal:
|
825
|
+
{
|
826
|
+
if (trees[trees[tree].parent].all_branches.size() == 0) continue;
|
827
|
+
|
828
|
+
if (trees[tree].split_lev == 0) {
|
829
|
+
|
830
|
+
if (trees[tree].split_this_branch == LessOrEqual)
|
831
|
+
trees[tree].split_this_branch = Equal;
|
832
|
+
else
|
833
|
+
trees[tree].split_this_branch = NotEqual;
|
834
|
+
|
835
|
+
}
|
836
|
+
|
837
|
+
else if (trees[tree].split_lev == (ncat_ord[trees[tree].col_num] - 2)) {
|
838
|
+
|
839
|
+
trees[tree].split_lev++;
|
840
|
+
if (trees[tree].split_this_branch == Greater)
|
841
|
+
trees[tree].split_this_branch = Equal;
|
842
|
+
else
|
843
|
+
trees[tree].split_this_branch = NotEqual;
|
844
|
+
|
845
|
+
}
|
846
|
+
break;
|
847
|
+
}
|
848
|
+
|
849
|
+
}
|
850
|
+
|
851
|
+
}
|
852
|
+
|
853
|
+
}
|
854
|
+
|
855
|
+
#ifdef TEST_MODE_DEFINE
|
856
|
+
/*
|
857
|
+
* Goodie to help with testing and debugging (not used in the final code)
|
858
|
+
*
|
859
|
+
* This function tries to unconnect unnecessary trees so that, if a tree has no clusters and its children
|
860
|
+
* don't have any clusters either, such tree would not be reached at prediction time. It will drop trees from the vector
|
861
|
+
* if they happen to lie at the end of it, but otherwise will just leave them there so as not to have to recalculate
|
862
|
+
* the tree indexes and avoid having to update them everywhere they are referenced (such as in identified outliers).
|
863
|
+
*
|
864
|
+
* This is only for categorical and ordinal columns, as numerical columns will always produce produce clusters when
|
865
|
+
* they have children.
|
866
|
+
*
|
867
|
+
* This is supposed to be done with the conditions at the end of each recursive function, but this piece of
|
868
|
+
* code can provide help in identifying errors when the code is modified.
|
869
|
+
*/
|
870
|
+
void prune_unused_trees(std::vector<ClusterTree> &trees)
|
871
|
+
{
|
872
|
+
/* TODO: when using 'follow_all', function should delete instead of disconnect by setting to zero */
|
873
|
+
if (trees.size() == 0) return;
|
874
|
+
for (size_t t = trees.size() - 1; t >= 0; t--) {
|
875
|
+
|
876
|
+
if (trees[t].binary_branches.size() > 0) {
|
877
|
+
for (size_t br = 0; br < trees[t].binary_branches.size(); br++) {
|
878
|
+
if (trees[t].binary_branches[br] == 0) continue;
|
879
|
+
if (trees[t].binary_branches[br] >= trees.size()) trees[t].binary_branches[br] = 0;
|
880
|
+
if (check_tree_is_not_needed(trees[trees[t].binary_branches[br]])) trees[t].binary_branches[br] = 0;
|
881
|
+
}
|
882
|
+
}
|
883
|
+
|
884
|
+
if (trees[t].all_branches.size() > 0) {
|
885
|
+
for (size_t br = 0; br < trees[t].all_branches.size(); br++) {
|
886
|
+
if (trees[t].all_branches[br] == 0) continue;
|
887
|
+
if (trees[t].all_branches[br] >= trees.size()) trees[t].all_branches[br] = 0;
|
888
|
+
if (check_tree_is_not_needed(trees[trees[t].all_branches[br]])) trees[t].all_branches[br] = 0;
|
889
|
+
}
|
890
|
+
}
|
891
|
+
|
892
|
+
|
893
|
+
if (check_tree_is_not_needed(trees[t])) {
|
894
|
+
|
895
|
+
/* disconnect tree from parent */
|
896
|
+
switch(trees[t].parent_branch) {
|
897
|
+
case IsNa:
|
898
|
+
{
|
899
|
+
trees[trees[t].parent].tree_NA = 0;
|
900
|
+
break;
|
901
|
+
}
|
902
|
+
|
903
|
+
case LessOrEqual:
|
904
|
+
{
|
905
|
+
trees[trees[t].parent].tree_left = 0;
|
906
|
+
break;
|
907
|
+
}
|
908
|
+
|
909
|
+
case Greater:
|
910
|
+
{
|
911
|
+
trees[trees[t].parent].tree_right = 0;
|
912
|
+
break;
|
913
|
+
}
|
914
|
+
|
915
|
+
case InSubset:
|
916
|
+
{
|
917
|
+
trees[trees[t].parent].tree_left = 0;
|
918
|
+
break;
|
919
|
+
}
|
920
|
+
|
921
|
+
case NotInSubset:
|
922
|
+
{
|
923
|
+
trees[trees[t].parent].tree_right = 0;
|
924
|
+
break;
|
925
|
+
}
|
926
|
+
|
927
|
+
}
|
928
|
+
|
929
|
+
if (t == (trees.size() - 1)) trees.pop_back();
|
930
|
+
}
|
931
|
+
if (t == 0) break;
|
932
|
+
}
|
933
|
+
}
|
934
|
+
#endif
|
935
|
+
|
936
|
+
/* Check whether a tree has no clusters and no children with clusters either */
|
937
|
+
bool check_tree_is_not_needed(ClusterTree &tree)
|
938
|
+
{
|
939
|
+
return
|
940
|
+
tree.tree_NA == 0 && tree.tree_left == 0 && tree.tree_right == 0 &&
|
941
|
+
tree.clusters.size() == 0 &&
|
942
|
+
(tree.binary_branches.size() == 0 || *std::max_element(tree.binary_branches.begin(), tree.binary_branches.end()) == 0) &&
|
943
|
+
(tree.all_branches.size() == 0 || *std::max_element(tree.all_branches.begin(), tree.all_branches.end()) == 0)
|
944
|
+
;
|
945
|
+
}
|
946
|
+
|
947
|
+
/*
|
948
|
+
* These functions simply check what's the minimum/maximum value that could identify an observation
|
949
|
+
* as outlier in any cluster, or which categories could be possibly flagged as outliers in any cluster.
|
950
|
+
* This info is redundant, as outliers can be identified by following splits, but it can help speed up
|
951
|
+
* things at prediction time by not having to even bother checking a column if the value is within
|
952
|
+
* non-flaggable limits.
|
953
|
+
*/
|
954
|
+
void calculate_cluster_minimums(ModelOutputs &model_outputs, size_t col)
|
955
|
+
{
|
956
|
+
for (size_t cl = 0; cl < model_outputs.all_clusters[col].size(); cl++) {
|
957
|
+
model_outputs.min_outlier_any_cl[col] = fmax(model_outputs.min_outlier_any_cl[col], model_outputs.all_clusters[col][cl].lower_lim);
|
958
|
+
model_outputs.max_outlier_any_cl[col] = fmin(model_outputs.max_outlier_any_cl[col], model_outputs.all_clusters[col][cl].upper_lim);
|
959
|
+
}
|
960
|
+
|
961
|
+
}
|
962
|
+
|
963
|
+
void calculate_cluster_poss_categs(ModelOutputs &model_outputs, size_t col, size_t col_rel)
|
964
|
+
{
|
965
|
+
if (model_outputs.all_clusters[col].size() == 0) return;
|
966
|
+
model_outputs.cat_outlier_any_cl[col_rel].resize(model_outputs.all_clusters[col][0].subset_common.size(), 0);
|
967
|
+
for (size_t cl = 0; cl < model_outputs.all_clusters[col].size(); cl++) {
|
968
|
+
for (size_t cat = 0; cat < model_outputs.all_clusters[col][cl].subset_common.size(); cat++) {
|
969
|
+
if (model_outputs.all_clusters[col][cl].subset_common[cat] != 0) model_outputs.cat_outlier_any_cl[col_rel][cat] = true;
|
970
|
+
}
|
971
|
+
}
|
972
|
+
}
|