isotree 0.2.0 → 0.3.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +16 -1
- data/LICENSE.txt +2 -2
- data/README.md +41 -23
- data/ext/isotree/ext.cpp +144 -31
- data/ext/isotree/extconf.rb +7 -7
- data/lib/isotree/dataset.rb +0 -1
- data/lib/isotree/isolation_forest.rb +114 -30
- data/lib/isotree/version.rb +1 -1
- data/vendor/isotree/LICENSE +1 -1
- data/vendor/isotree/README.md +165 -27
- data/vendor/isotree/include/isotree.hpp +2111 -0
- data/vendor/isotree/include/isotree_oop.hpp +394 -0
- data/vendor/isotree/inst/COPYRIGHTS +62 -0
- data/vendor/isotree/src/RcppExports.cpp +525 -52
- data/vendor/isotree/src/Rwrapper.cpp +1931 -268
- data/vendor/isotree/src/c_interface.cpp +953 -0
- data/vendor/isotree/src/crit.hpp +4232 -0
- data/vendor/isotree/src/dist.hpp +1886 -0
- data/vendor/isotree/src/exp_depth_table.hpp +134 -0
- data/vendor/isotree/src/extended.hpp +1444 -0
- data/vendor/isotree/src/external_facing_generic.hpp +399 -0
- data/vendor/isotree/src/fit_model.hpp +2401 -0
- data/vendor/isotree/src/{dealloc.cpp → headers_joined.hpp} +38 -22
- data/vendor/isotree/src/helpers_iforest.hpp +813 -0
- data/vendor/isotree/src/{impute.cpp → impute.hpp} +353 -122
- data/vendor/isotree/src/indexer.cpp +515 -0
- data/vendor/isotree/src/instantiate_template_headers.cpp +118 -0
- data/vendor/isotree/src/instantiate_template_headers.hpp +240 -0
- data/vendor/isotree/src/isoforest.hpp +1659 -0
- data/vendor/isotree/src/isotree.hpp +1804 -392
- data/vendor/isotree/src/isotree_exportable.hpp +99 -0
- data/vendor/isotree/src/merge_models.cpp +159 -16
- data/vendor/isotree/src/mult.hpp +1321 -0
- data/vendor/isotree/src/oop_interface.cpp +842 -0
- data/vendor/isotree/src/oop_interface.hpp +278 -0
- data/vendor/isotree/src/other_helpers.hpp +219 -0
- data/vendor/isotree/src/predict.hpp +1932 -0
- data/vendor/isotree/src/python_helpers.hpp +134 -0
- data/vendor/isotree/src/ref_indexer.hpp +154 -0
- data/vendor/isotree/src/robinmap/LICENSE +21 -0
- data/vendor/isotree/src/robinmap/README.md +483 -0
- data/vendor/isotree/src/robinmap/include/tsl/robin_growth_policy.h +406 -0
- data/vendor/isotree/src/robinmap/include/tsl/robin_hash.h +1620 -0
- data/vendor/isotree/src/robinmap/include/tsl/robin_map.h +807 -0
- data/vendor/isotree/src/robinmap/include/tsl/robin_set.h +660 -0
- data/vendor/isotree/src/serialize.cpp +4300 -139
- data/vendor/isotree/src/sql.cpp +141 -59
- data/vendor/isotree/src/subset_models.cpp +174 -0
- data/vendor/isotree/src/utils.hpp +3808 -0
- data/vendor/isotree/src/xoshiro.hpp +467 -0
- data/vendor/isotree/src/ziggurat.hpp +405 -0
- metadata +40 -106
- data/vendor/cereal/LICENSE +0 -24
- data/vendor/cereal/README.md +0 -85
- data/vendor/cereal/include/cereal/access.hpp +0 -351
- data/vendor/cereal/include/cereal/archives/adapters.hpp +0 -163
- data/vendor/cereal/include/cereal/archives/binary.hpp +0 -169
- data/vendor/cereal/include/cereal/archives/json.hpp +0 -1019
- data/vendor/cereal/include/cereal/archives/portable_binary.hpp +0 -334
- data/vendor/cereal/include/cereal/archives/xml.hpp +0 -956
- data/vendor/cereal/include/cereal/cereal.hpp +0 -1089
- data/vendor/cereal/include/cereal/details/helpers.hpp +0 -422
- data/vendor/cereal/include/cereal/details/polymorphic_impl.hpp +0 -796
- data/vendor/cereal/include/cereal/details/polymorphic_impl_fwd.hpp +0 -65
- data/vendor/cereal/include/cereal/details/static_object.hpp +0 -127
- data/vendor/cereal/include/cereal/details/traits.hpp +0 -1411
- data/vendor/cereal/include/cereal/details/util.hpp +0 -84
- data/vendor/cereal/include/cereal/external/base64.hpp +0 -134
- data/vendor/cereal/include/cereal/external/rapidjson/allocators.h +0 -284
- data/vendor/cereal/include/cereal/external/rapidjson/cursorstreamwrapper.h +0 -78
- data/vendor/cereal/include/cereal/external/rapidjson/document.h +0 -2652
- data/vendor/cereal/include/cereal/external/rapidjson/encodedstream.h +0 -299
- data/vendor/cereal/include/cereal/external/rapidjson/encodings.h +0 -716
- data/vendor/cereal/include/cereal/external/rapidjson/error/en.h +0 -74
- data/vendor/cereal/include/cereal/external/rapidjson/error/error.h +0 -161
- data/vendor/cereal/include/cereal/external/rapidjson/filereadstream.h +0 -99
- data/vendor/cereal/include/cereal/external/rapidjson/filewritestream.h +0 -104
- data/vendor/cereal/include/cereal/external/rapidjson/fwd.h +0 -151
- data/vendor/cereal/include/cereal/external/rapidjson/internal/biginteger.h +0 -290
- data/vendor/cereal/include/cereal/external/rapidjson/internal/diyfp.h +0 -271
- data/vendor/cereal/include/cereal/external/rapidjson/internal/dtoa.h +0 -245
- data/vendor/cereal/include/cereal/external/rapidjson/internal/ieee754.h +0 -78
- data/vendor/cereal/include/cereal/external/rapidjson/internal/itoa.h +0 -308
- data/vendor/cereal/include/cereal/external/rapidjson/internal/meta.h +0 -186
- data/vendor/cereal/include/cereal/external/rapidjson/internal/pow10.h +0 -55
- data/vendor/cereal/include/cereal/external/rapidjson/internal/regex.h +0 -740
- data/vendor/cereal/include/cereal/external/rapidjson/internal/stack.h +0 -232
- data/vendor/cereal/include/cereal/external/rapidjson/internal/strfunc.h +0 -69
- data/vendor/cereal/include/cereal/external/rapidjson/internal/strtod.h +0 -290
- data/vendor/cereal/include/cereal/external/rapidjson/internal/swap.h +0 -46
- data/vendor/cereal/include/cereal/external/rapidjson/istreamwrapper.h +0 -128
- data/vendor/cereal/include/cereal/external/rapidjson/memorybuffer.h +0 -70
- data/vendor/cereal/include/cereal/external/rapidjson/memorystream.h +0 -71
- data/vendor/cereal/include/cereal/external/rapidjson/msinttypes/inttypes.h +0 -316
- data/vendor/cereal/include/cereal/external/rapidjson/msinttypes/stdint.h +0 -300
- data/vendor/cereal/include/cereal/external/rapidjson/ostreamwrapper.h +0 -81
- data/vendor/cereal/include/cereal/external/rapidjson/pointer.h +0 -1414
- data/vendor/cereal/include/cereal/external/rapidjson/prettywriter.h +0 -277
- data/vendor/cereal/include/cereal/external/rapidjson/rapidjson.h +0 -656
- data/vendor/cereal/include/cereal/external/rapidjson/reader.h +0 -2230
- data/vendor/cereal/include/cereal/external/rapidjson/schema.h +0 -2497
- data/vendor/cereal/include/cereal/external/rapidjson/stream.h +0 -223
- data/vendor/cereal/include/cereal/external/rapidjson/stringbuffer.h +0 -121
- data/vendor/cereal/include/cereal/external/rapidjson/writer.h +0 -709
- data/vendor/cereal/include/cereal/external/rapidxml/license.txt +0 -52
- data/vendor/cereal/include/cereal/external/rapidxml/manual.html +0 -406
- data/vendor/cereal/include/cereal/external/rapidxml/rapidxml.hpp +0 -2624
- data/vendor/cereal/include/cereal/external/rapidxml/rapidxml_iterators.hpp +0 -175
- data/vendor/cereal/include/cereal/external/rapidxml/rapidxml_print.hpp +0 -428
- data/vendor/cereal/include/cereal/external/rapidxml/rapidxml_utils.hpp +0 -123
- data/vendor/cereal/include/cereal/macros.hpp +0 -154
- data/vendor/cereal/include/cereal/specialize.hpp +0 -139
- data/vendor/cereal/include/cereal/types/array.hpp +0 -79
- data/vendor/cereal/include/cereal/types/atomic.hpp +0 -55
- data/vendor/cereal/include/cereal/types/base_class.hpp +0 -203
- data/vendor/cereal/include/cereal/types/bitset.hpp +0 -176
- data/vendor/cereal/include/cereal/types/boost_variant.hpp +0 -164
- data/vendor/cereal/include/cereal/types/chrono.hpp +0 -72
- data/vendor/cereal/include/cereal/types/common.hpp +0 -129
- data/vendor/cereal/include/cereal/types/complex.hpp +0 -56
- data/vendor/cereal/include/cereal/types/concepts/pair_associative_container.hpp +0 -73
- data/vendor/cereal/include/cereal/types/deque.hpp +0 -62
- data/vendor/cereal/include/cereal/types/forward_list.hpp +0 -68
- data/vendor/cereal/include/cereal/types/functional.hpp +0 -43
- data/vendor/cereal/include/cereal/types/list.hpp +0 -62
- data/vendor/cereal/include/cereal/types/map.hpp +0 -36
- data/vendor/cereal/include/cereal/types/memory.hpp +0 -425
- data/vendor/cereal/include/cereal/types/optional.hpp +0 -66
- data/vendor/cereal/include/cereal/types/polymorphic.hpp +0 -483
- data/vendor/cereal/include/cereal/types/queue.hpp +0 -132
- data/vendor/cereal/include/cereal/types/set.hpp +0 -103
- data/vendor/cereal/include/cereal/types/stack.hpp +0 -76
- data/vendor/cereal/include/cereal/types/string.hpp +0 -61
- data/vendor/cereal/include/cereal/types/tuple.hpp +0 -123
- data/vendor/cereal/include/cereal/types/unordered_map.hpp +0 -36
- data/vendor/cereal/include/cereal/types/unordered_set.hpp +0 -99
- data/vendor/cereal/include/cereal/types/utility.hpp +0 -47
- data/vendor/cereal/include/cereal/types/valarray.hpp +0 -89
- data/vendor/cereal/include/cereal/types/variant.hpp +0 -109
- data/vendor/cereal/include/cereal/types/vector.hpp +0 -112
- data/vendor/cereal/include/cereal/version.hpp +0 -52
- data/vendor/isotree/src/Makevars +0 -4
- data/vendor/isotree/src/crit.cpp +0 -912
- data/vendor/isotree/src/dist.cpp +0 -749
- data/vendor/isotree/src/extended.cpp +0 -790
- data/vendor/isotree/src/fit_model.cpp +0 -1090
- data/vendor/isotree/src/helpers_iforest.cpp +0 -324
- data/vendor/isotree/src/isoforest.cpp +0 -771
- data/vendor/isotree/src/mult.cpp +0 -607
- data/vendor/isotree/src/predict.cpp +0 -853
- data/vendor/isotree/src/utils.cpp +0 -1566
@@ -0,0 +1,1321 @@
|
|
1
|
+
/* Isolation forests and variations thereof, with adjustments for incorporation
|
2
|
+
* of categorical variables and missing values.
|
3
|
+
* Writen for C++11 standard and aimed at being used in R and Python.
|
4
|
+
*
|
5
|
+
* This library is based on the following works:
|
6
|
+
* [1] Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou.
|
7
|
+
* "Isolation forest."
|
8
|
+
* 2008 Eighth IEEE International Conference on Data Mining. IEEE, 2008.
|
9
|
+
* [2] Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou.
|
10
|
+
* "Isolation-based anomaly detection."
|
11
|
+
* ACM Transactions on Knowledge Discovery from Data (TKDD) 6.1 (2012): 3.
|
12
|
+
* [3] Hariri, Sahand, Matias Carrasco Kind, and Robert J. Brunner.
|
13
|
+
* "Extended Isolation Forest."
|
14
|
+
* arXiv preprint arXiv:1811.02141 (2018).
|
15
|
+
* [4] Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou.
|
16
|
+
* "On detecting clustered anomalies using SCiForest."
|
17
|
+
* Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, Berlin, Heidelberg, 2010.
|
18
|
+
* [5] https://sourceforge.net/projects/iforest/
|
19
|
+
* [6] https://math.stackexchange.com/questions/3388518/expected-number-of-paths-required-to-separate-elements-in-a-binary-tree
|
20
|
+
* [7] Quinlan, J. Ross. C4. 5: programs for machine learning. Elsevier, 2014.
|
21
|
+
* [8] Cortes, David.
|
22
|
+
* "Distance approximation using Isolation Forests."
|
23
|
+
* arXiv preprint arXiv:1910.12362 (2019).
|
24
|
+
* [9] Cortes, David.
|
25
|
+
* "Imputing missing values with unsupervised random trees."
|
26
|
+
* arXiv preprint arXiv:1911.06646 (2019).
|
27
|
+
* [10] https://math.stackexchange.com/questions/3333220/expected-average-depth-in-random-binary-tree-constructed-top-to-bottom
|
28
|
+
* [11] Cortes, David.
|
29
|
+
* "Revisiting randomized choices in isolation forests."
|
30
|
+
* arXiv preprint arXiv:2110.13402 (2021).
|
31
|
+
* [12] Guha, Sudipto, et al.
|
32
|
+
* "Robust random cut forest based anomaly detection on streams."
|
33
|
+
* International conference on machine learning. PMLR, 2016.
|
34
|
+
* [13] Cortes, David.
|
35
|
+
* "Isolation forests: looking beyond tree depth."
|
36
|
+
* arXiv preprint arXiv:2111.11639 (2021).
|
37
|
+
* [14] Ting, Kai Ming, Yue Zhu, and Zhi-Hua Zhou.
|
38
|
+
* "Isolation kernel and its effect on SVM"
|
39
|
+
* Proceedings of the 24th ACM SIGKDD
|
40
|
+
* International Conference on Knowledge Discovery & Data Mining. 2018.
|
41
|
+
*
|
42
|
+
* BSD 2-Clause License
|
43
|
+
* Copyright (c) 2019-2022, David Cortes
|
44
|
+
* All rights reserved.
|
45
|
+
* Redistribution and use in source and binary forms, with or without
|
46
|
+
* modification, are permitted provided that the following conditions are met:
|
47
|
+
* * Redistributions of source code must retain the above copyright notice, this
|
48
|
+
* list of conditions and the following disclaimer.
|
49
|
+
* * Redistributions in binary form must reproduce the above copyright notice,
|
50
|
+
* this list of conditions and the following disclaimer in the documentation
|
51
|
+
* and/or other materials provided with the distribution.
|
52
|
+
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
53
|
+
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
54
|
+
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
55
|
+
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
|
56
|
+
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
57
|
+
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
|
58
|
+
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
|
59
|
+
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
|
60
|
+
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
61
|
+
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
62
|
+
*/
|
63
|
+
#include "isotree.hpp"
|
64
|
+
|
65
|
+
/* FIXME / TODO: here the calculations of medians do not take weights into account */
|
66
|
+
|
67
|
+
#define SD_MIN 1e-10
|
68
|
+
/* https://www.johndcook.com/blog/standard_deviation/ */
|
69
|
+
|
70
|
+
/* for regular numerical */
|
71
|
+
template <class real_t, class real_t_>
|
72
|
+
void calc_mean_and_sd_t(size_t ix_arr[], size_t st, size_t end, real_t_ *restrict x,
|
73
|
+
MissingAction missing_action, double &restrict x_sd, double &restrict x_mean)
|
74
|
+
{
|
75
|
+
real_t m = 0;
|
76
|
+
real_t s = 0;
|
77
|
+
real_t m_prev = x[ix_arr[st]];
|
78
|
+
real_t xval;
|
79
|
+
|
80
|
+
if (missing_action == Fail)
|
81
|
+
{
|
82
|
+
m_prev = x[ix_arr[st]];
|
83
|
+
for (size_t row = st; row <= end; row++)
|
84
|
+
{
|
85
|
+
xval = x[ix_arr[row]];
|
86
|
+
m += (xval - m) / (real_t)(row - st + 1);
|
87
|
+
s = std::fma(xval - m, xval - m_prev, s);
|
88
|
+
m_prev = m;
|
89
|
+
}
|
90
|
+
|
91
|
+
x_mean = m;
|
92
|
+
x_sd = std::sqrt(s / (real_t)(end - st + 1));
|
93
|
+
}
|
94
|
+
|
95
|
+
else
|
96
|
+
{
|
97
|
+
size_t cnt = 0;
|
98
|
+
while (is_na_or_inf(m_prev) && st <= end)
|
99
|
+
{
|
100
|
+
m_prev = x[ix_arr[++st]];
|
101
|
+
}
|
102
|
+
|
103
|
+
for (size_t row = st; row <= end; row++)
|
104
|
+
{
|
105
|
+
xval = x[ix_arr[row]];
|
106
|
+
if (likely(!is_na_or_inf(xval)))
|
107
|
+
{
|
108
|
+
cnt++;
|
109
|
+
m += (xval - m) / (real_t)cnt;
|
110
|
+
s = std::fma(xval - m, xval - m_prev, s);
|
111
|
+
m_prev = m;
|
112
|
+
}
|
113
|
+
}
|
114
|
+
|
115
|
+
x_mean = m;
|
116
|
+
x_sd = std::sqrt(s / (real_t)cnt);
|
117
|
+
}
|
118
|
+
}
|
119
|
+
|
120
|
+
template <class real_t_>
|
121
|
+
double calc_mean_only(size_t ix_arr[], size_t st, size_t end, real_t_ *restrict x)
|
122
|
+
{
|
123
|
+
size_t cnt = 0;
|
124
|
+
double m = 0;
|
125
|
+
real_t_ xval;
|
126
|
+
for (size_t row = st; row <= end; row++)
|
127
|
+
{
|
128
|
+
xval = x[ix_arr[row]];
|
129
|
+
if (likely(!is_na_or_inf(xval)))
|
130
|
+
{
|
131
|
+
cnt++;
|
132
|
+
m += (xval - m) / (double)cnt;
|
133
|
+
}
|
134
|
+
}
|
135
|
+
|
136
|
+
return m;
|
137
|
+
}
|
138
|
+
|
139
|
+
template <class real_t_, class ldouble_safe>
|
140
|
+
void calc_mean_and_sd(size_t ix_arr[], size_t st, size_t end, real_t_ *restrict x,
|
141
|
+
MissingAction missing_action, double &restrict x_sd, double &restrict x_mean)
|
142
|
+
{
|
143
|
+
if (end - st + 1 < THRESHOLD_LONG_DOUBLE)
|
144
|
+
calc_mean_and_sd_t<double, real_t_>(ix_arr, st, end, x, missing_action, x_sd, x_mean);
|
145
|
+
else
|
146
|
+
calc_mean_and_sd_t<ldouble_safe, real_t_>(ix_arr, st, end, x, missing_action, x_sd, x_mean);
|
147
|
+
x_sd = std::fmax(x_sd, SD_MIN);
|
148
|
+
}
|
149
|
+
|
150
|
+
template <class real_t_, class mapping, class ldouble_safe>
|
151
|
+
void calc_mean_and_sd_weighted(size_t ix_arr[], size_t st, size_t end, real_t_ *restrict x, mapping &restrict w,
|
152
|
+
MissingAction missing_action, double &restrict x_sd, double &restrict x_mean)
|
153
|
+
{
|
154
|
+
ldouble_safe cnt = 0;
|
155
|
+
ldouble_safe w_this;
|
156
|
+
ldouble_safe m = 0;
|
157
|
+
ldouble_safe s = 0;
|
158
|
+
ldouble_safe m_prev = x[ix_arr[st]];
|
159
|
+
ldouble_safe xval;
|
160
|
+
while (is_na_or_inf(m_prev) && st <= end)
|
161
|
+
{
|
162
|
+
m_prev = x[ix_arr[++st]];
|
163
|
+
}
|
164
|
+
|
165
|
+
for (size_t row = st; row <= end; row++)
|
166
|
+
{
|
167
|
+
xval = x[ix_arr[row]];
|
168
|
+
if (likely(!is_na_or_inf(xval)))
|
169
|
+
{
|
170
|
+
w_this = w[ix_arr[row]];
|
171
|
+
cnt += w_this;
|
172
|
+
m = std::fma(w_this, (xval - m) / cnt, m);
|
173
|
+
s = std::fma(w_this, (xval - m) * (xval - m_prev), s);
|
174
|
+
m_prev = m;
|
175
|
+
}
|
176
|
+
}
|
177
|
+
|
178
|
+
x_mean = m;
|
179
|
+
x_sd = std::sqrt((ldouble_safe)s / (ldouble_safe)cnt);
|
180
|
+
}
|
181
|
+
|
182
|
+
template <class real_t_, class mapping>
|
183
|
+
double calc_mean_only_weighted(size_t ix_arr[], size_t st, size_t end, real_t_ *restrict x, mapping &restrict w)
|
184
|
+
{
|
185
|
+
double cnt = 0;
|
186
|
+
double w_this;
|
187
|
+
double m = 0;
|
188
|
+
for (size_t row = st; row <= end; row++)
|
189
|
+
{
|
190
|
+
if (likely(!is_na_or_inf(x[ix_arr[row]])))
|
191
|
+
{
|
192
|
+
w_this = w[ix_arr[row]];
|
193
|
+
cnt += w_this;
|
194
|
+
m = std::fma(w_this, (x[ix_arr[row]] - m) / cnt, m);
|
195
|
+
}
|
196
|
+
}
|
197
|
+
|
198
|
+
return m;
|
199
|
+
}
|
200
|
+
|
201
|
+
/* for sparse numerical */
|
202
|
+
template <class real_t_, class sparse_ix, class real_t>
|
203
|
+
void calc_mean_and_sd_(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num,
|
204
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
205
|
+
double &restrict x_sd, double &restrict x_mean)
|
206
|
+
{
|
207
|
+
/* ix_arr must be already sorted beforehand */
|
208
|
+
if (Xc_indptr[col_num] == Xc_indptr[col_num + 1])
|
209
|
+
{
|
210
|
+
x_sd = 0;
|
211
|
+
x_mean = 0;
|
212
|
+
return;
|
213
|
+
}
|
214
|
+
size_t st_col = Xc_indptr[col_num];
|
215
|
+
size_t end_col = Xc_indptr[col_num + 1] - 1;
|
216
|
+
size_t curr_pos = st_col;
|
217
|
+
size_t ind_end_col = (size_t) Xc_ind[end_col];
|
218
|
+
size_t *ptr_st = std::lower_bound(ix_arr + st, ix_arr + end + 1, (size_t)Xc_ind[st_col]);
|
219
|
+
|
220
|
+
size_t cnt = end - st + 1;
|
221
|
+
size_t added = 0;
|
222
|
+
real_t m = 0;
|
223
|
+
real_t s = 0;
|
224
|
+
real_t m_prev = 0;
|
225
|
+
|
226
|
+
for (size_t *row = ptr_st;
|
227
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
228
|
+
)
|
229
|
+
{
|
230
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
231
|
+
{
|
232
|
+
if (unlikely(is_na_or_inf(Xc[curr_pos])))
|
233
|
+
{
|
234
|
+
cnt--;
|
235
|
+
}
|
236
|
+
|
237
|
+
else
|
238
|
+
{
|
239
|
+
if (added == 0) m_prev = Xc[curr_pos];
|
240
|
+
m += (Xc[curr_pos] - m) / (real_t)(++added);
|
241
|
+
s = std::fma(Xc[curr_pos] - m, Xc[curr_pos] - m_prev, s);
|
242
|
+
m_prev = m;
|
243
|
+
}
|
244
|
+
|
245
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
246
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
247
|
+
}
|
248
|
+
|
249
|
+
else
|
250
|
+
{
|
251
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
252
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
253
|
+
else
|
254
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
255
|
+
}
|
256
|
+
}
|
257
|
+
|
258
|
+
if (added == 0)
|
259
|
+
{
|
260
|
+
x_mean = 0;
|
261
|
+
x_sd = 0;
|
262
|
+
return;
|
263
|
+
}
|
264
|
+
|
265
|
+
/* Note: up to this point:
|
266
|
+
m = sum(x)/nnz
|
267
|
+
s = sum(x^2) - (1/nnz)*(sum(x)^2)
|
268
|
+
Here the standard deviation is given by:
|
269
|
+
sigma = (1/n)*(sum(x^2) - (1/n)*(sum(x)^2))
|
270
|
+
The difference can be put to a closed form. */
|
271
|
+
if (cnt > added)
|
272
|
+
{
|
273
|
+
s += square(m) * ((real_t)added * ((real_t)1 - (real_t)added/(real_t)cnt));
|
274
|
+
m *= (real_t)added / (real_t)cnt;
|
275
|
+
}
|
276
|
+
|
277
|
+
x_mean = m;
|
278
|
+
x_sd = std::sqrt(s / (real_t)cnt);
|
279
|
+
}
|
280
|
+
|
281
|
+
template <class real_t_, class sparse_ix, class ldouble_safe>
|
282
|
+
void calc_mean_and_sd(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num,
|
283
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
284
|
+
double &restrict x_sd, double &restrict x_mean)
|
285
|
+
{
|
286
|
+
if (end - st + 1 < THRESHOLD_LONG_DOUBLE)
|
287
|
+
calc_mean_and_sd_<real_t_, sparse_ix, double>(ix_arr, st, end, col_num, Xc, Xc_ind, Xc_indptr, x_sd, x_mean);
|
288
|
+
else
|
289
|
+
calc_mean_and_sd_<real_t_, sparse_ix, ldouble_safe>(ix_arr, st, end, col_num, Xc, Xc_ind, Xc_indptr, x_sd, x_mean);
|
290
|
+
x_sd = std::fmax(SD_MIN, x_sd);
|
291
|
+
}
|
292
|
+
|
293
|
+
template <class real_t_, class sparse_ix, class ldouble_safe>
|
294
|
+
double calc_mean_only(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num,
|
295
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr)
|
296
|
+
{
|
297
|
+
/* ix_arr must be already sorted beforehand */
|
298
|
+
if (Xc_indptr[col_num] == Xc_indptr[col_num + 1])
|
299
|
+
return 0;
|
300
|
+
size_t st_col = Xc_indptr[col_num];
|
301
|
+
size_t end_col = Xc_indptr[col_num + 1] - 1;
|
302
|
+
size_t curr_pos = st_col;
|
303
|
+
size_t ind_end_col = (size_t) Xc_ind[end_col];
|
304
|
+
size_t *ptr_st = std::lower_bound(ix_arr + st, ix_arr + end + 1, (size_t)Xc_ind[st_col]);
|
305
|
+
|
306
|
+
size_t cnt = end - st + 1;
|
307
|
+
size_t added = 0;
|
308
|
+
double m = 0;
|
309
|
+
|
310
|
+
for (size_t *row = ptr_st;
|
311
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
312
|
+
)
|
313
|
+
{
|
314
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
315
|
+
{
|
316
|
+
if (unlikely(is_na_or_inf(Xc[curr_pos])))
|
317
|
+
cnt--;
|
318
|
+
else
|
319
|
+
m += (Xc[curr_pos] - m) / (double)(++added);
|
320
|
+
|
321
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
322
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
323
|
+
}
|
324
|
+
|
325
|
+
else
|
326
|
+
{
|
327
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
328
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
329
|
+
else
|
330
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
331
|
+
}
|
332
|
+
}
|
333
|
+
|
334
|
+
if (added == 0)
|
335
|
+
return 0;
|
336
|
+
|
337
|
+
if (cnt > added)
|
338
|
+
m *= ((ldouble_safe)added / (ldouble_safe)cnt);
|
339
|
+
|
340
|
+
return m;
|
341
|
+
}
|
342
|
+
|
343
|
+
template <class real_t_, class sparse_ix, class mapping, class ldouble_safe>
|
344
|
+
void calc_mean_and_sd_weighted(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num,
|
345
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
346
|
+
double &restrict x_sd, double &restrict x_mean, mapping &restrict w)
|
347
|
+
{
|
348
|
+
/* ix_arr must be already sorted beforehand */
|
349
|
+
if (Xc_indptr[col_num] == Xc_indptr[col_num + 1])
|
350
|
+
{
|
351
|
+
x_sd = 0;
|
352
|
+
x_mean = 0;
|
353
|
+
return;
|
354
|
+
}
|
355
|
+
size_t st_col = Xc_indptr[col_num];
|
356
|
+
size_t end_col = Xc_indptr[col_num + 1] - 1;
|
357
|
+
size_t curr_pos = st_col;
|
358
|
+
size_t ind_end_col = (size_t) Xc_ind[end_col];
|
359
|
+
size_t *ptr_st = std::lower_bound(ix_arr + st, ix_arr + end + 1, (size_t)Xc_ind[st_col]);
|
360
|
+
|
361
|
+
ldouble_safe cnt = 0.;
|
362
|
+
for (size_t row = st; row <= end; row++)
|
363
|
+
cnt += w[ix_arr[row]];
|
364
|
+
ldouble_safe added = 0;
|
365
|
+
ldouble_safe m = 0;
|
366
|
+
ldouble_safe s = 0;
|
367
|
+
ldouble_safe m_prev = 0;
|
368
|
+
ldouble_safe w_this;
|
369
|
+
|
370
|
+
for (size_t *row = ptr_st;
|
371
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
372
|
+
)
|
373
|
+
{
|
374
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
375
|
+
{
|
376
|
+
if (unlikely(is_na_or_inf(Xc[curr_pos])))
|
377
|
+
{
|
378
|
+
cnt -= w[*row];
|
379
|
+
}
|
380
|
+
|
381
|
+
else
|
382
|
+
{
|
383
|
+
w_this = w[*row];
|
384
|
+
if (added == 0) m_prev = Xc[curr_pos];
|
385
|
+
added += w_this;
|
386
|
+
m = std::fma(w_this, (Xc[curr_pos] - m) / added, m);
|
387
|
+
s = std::fma(w_this, (Xc[curr_pos] - m) * (Xc[curr_pos] - m_prev), s);
|
388
|
+
m_prev = m;
|
389
|
+
}
|
390
|
+
|
391
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
392
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
393
|
+
}
|
394
|
+
|
395
|
+
else
|
396
|
+
{
|
397
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
398
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
399
|
+
else
|
400
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
401
|
+
}
|
402
|
+
}
|
403
|
+
|
404
|
+
if (added == 0)
|
405
|
+
{
|
406
|
+
x_mean = 0;
|
407
|
+
x_sd = 0;
|
408
|
+
return;
|
409
|
+
}
|
410
|
+
|
411
|
+
/* Note: up to this point:
|
412
|
+
m = sum(x)/nnz
|
413
|
+
s = sum(x^2) - (1/nnz)*(sum(x)^2)
|
414
|
+
Here the standard deviation is given by:
|
415
|
+
sigma = (1/n)*(sum(x^2) - (1/n)*(sum(x)^2))
|
416
|
+
The difference can be put to a closed form. */
|
417
|
+
if (cnt > added)
|
418
|
+
{
|
419
|
+
s += square(m) * (added * ((ldouble_safe)1 - (ldouble_safe)added/(ldouble_safe)cnt));
|
420
|
+
m *= added / cnt;
|
421
|
+
}
|
422
|
+
|
423
|
+
x_mean = m;
|
424
|
+
x_sd = std::sqrt(s / (ldouble_safe)cnt);
|
425
|
+
}
|
426
|
+
|
427
|
+
template <class real_t_, class sparse_ix, class mapping, class ldouble_safe>
|
428
|
+
double calc_mean_only_weighted(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num,
|
429
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
430
|
+
mapping &restrict w)
|
431
|
+
{
|
432
|
+
/* ix_arr must be already sorted beforehand */
|
433
|
+
if (Xc_indptr[col_num] == Xc_indptr[col_num + 1])
|
434
|
+
return 0;
|
435
|
+
size_t st_col = Xc_indptr[col_num];
|
436
|
+
size_t end_col = Xc_indptr[col_num + 1] - 1;
|
437
|
+
size_t curr_pos = st_col;
|
438
|
+
size_t ind_end_col = (size_t) Xc_ind[end_col];
|
439
|
+
size_t *ptr_st = std::lower_bound(ix_arr + st, ix_arr + end + 1, (size_t)Xc_ind[st_col]);
|
440
|
+
|
441
|
+
ldouble_safe cnt = 0.;
|
442
|
+
for (size_t row = st; row <= end; row++)
|
443
|
+
cnt += w[ix_arr[row]];
|
444
|
+
ldouble_safe added = 0;
|
445
|
+
ldouble_safe m = 0;
|
446
|
+
ldouble_safe w_this;
|
447
|
+
|
448
|
+
for (size_t *row = ptr_st;
|
449
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
450
|
+
)
|
451
|
+
{
|
452
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
453
|
+
{
|
454
|
+
if (unlikely(is_na_or_inf(Xc[curr_pos]))) {
|
455
|
+
cnt -= w[*row];
|
456
|
+
}
|
457
|
+
|
458
|
+
else {
|
459
|
+
w_this = w[*row];
|
460
|
+
added += w_this;
|
461
|
+
m += w_this * (Xc[curr_pos] - m) / added;
|
462
|
+
}
|
463
|
+
|
464
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
465
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
466
|
+
}
|
467
|
+
|
468
|
+
else
|
469
|
+
{
|
470
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
471
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
472
|
+
else
|
473
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
474
|
+
}
|
475
|
+
}
|
476
|
+
|
477
|
+
if (added == 0)
|
478
|
+
return 0;
|
479
|
+
|
480
|
+
if (cnt > added)
|
481
|
+
m *= (ldouble_safe)added / (ldouble_safe)cnt;
|
482
|
+
|
483
|
+
return m;
|
484
|
+
}
|
485
|
+
|
486
|
+
/* Note about these functions: they write into an array that does not need to match to 'ix_arr',
|
487
|
+
and instead, the index that is stored in ix_arr[n] will have the value in res[n] */
|
488
|
+
|
489
|
+
|
490
|
+
/* for regular numerical */
|
491
|
+
template <class real_t_>
|
492
|
+
void add_linear_comb(size_t ix_arr[], size_t st, size_t end, double *restrict res,
|
493
|
+
real_t_ *restrict x, double &coef, double x_sd, double x_mean, double &restrict fill_val,
|
494
|
+
MissingAction missing_action, double *restrict buffer_arr,
|
495
|
+
size_t *restrict buffer_NAs, bool first_run)
|
496
|
+
{
|
497
|
+
/* TODO: here don't need the buffer for NAs */
|
498
|
+
|
499
|
+
if (first_run)
|
500
|
+
coef /= x_sd;
|
501
|
+
|
502
|
+
size_t cnt = 0;
|
503
|
+
size_t cnt_NA = 0;
|
504
|
+
double *restrict res_write = res - st;
|
505
|
+
|
506
|
+
if (missing_action == Fail)
|
507
|
+
{
|
508
|
+
for (size_t row = st; row <= end; row++)
|
509
|
+
res_write[row] = std::fma(x[ix_arr[row]] - x_mean, coef, res_write[row]);
|
510
|
+
}
|
511
|
+
|
512
|
+
else
|
513
|
+
{
|
514
|
+
if (first_run)
|
515
|
+
{
|
516
|
+
for (size_t row = st; row <= end; row++)
|
517
|
+
{
|
518
|
+
if (likely(!is_na_or_inf(x[ix_arr[row]])))
|
519
|
+
{
|
520
|
+
res_write[row] = std::fma(x[ix_arr[row]] - x_mean, coef, res_write[row]);
|
521
|
+
buffer_arr[cnt++] = x[ix_arr[row]];
|
522
|
+
}
|
523
|
+
|
524
|
+
else
|
525
|
+
{
|
526
|
+
buffer_NAs[cnt_NA++] = row;
|
527
|
+
}
|
528
|
+
|
529
|
+
}
|
530
|
+
}
|
531
|
+
|
532
|
+
else
|
533
|
+
{
|
534
|
+
for (size_t row = st; row <= end; row++)
|
535
|
+
{
|
536
|
+
res_write[row] += (is_na_or_inf(x[ix_arr[row]]))? fill_val : ( (x[ix_arr[row]]-x_mean) * coef );
|
537
|
+
}
|
538
|
+
return;
|
539
|
+
}
|
540
|
+
|
541
|
+
size_t mid_ceil = cnt / 2;
|
542
|
+
std::partial_sort(buffer_arr, buffer_arr + mid_ceil + 1, buffer_arr + cnt);
|
543
|
+
|
544
|
+
if ((cnt % 2) == 0)
|
545
|
+
fill_val = buffer_arr[mid_ceil-1] + (buffer_arr[mid_ceil] - buffer_arr[mid_ceil-1]) / 2.0;
|
546
|
+
else
|
547
|
+
fill_val = buffer_arr[mid_ceil];
|
548
|
+
|
549
|
+
fill_val = (fill_val - x_mean) * coef;
|
550
|
+
if (cnt_NA && fill_val)
|
551
|
+
{
|
552
|
+
for (size_t row = 0; row < cnt_NA; row++)
|
553
|
+
res_write[buffer_NAs[row]] += fill_val;
|
554
|
+
}
|
555
|
+
|
556
|
+
}
|
557
|
+
}
|
558
|
+
|
559
|
+
/* for regular numerical */
|
560
|
+
template <class real_t_, class mapping, class ldouble_safe>
|
561
|
+
void add_linear_comb_weighted(size_t ix_arr[], size_t st, size_t end, double *restrict res,
|
562
|
+
real_t_ *restrict x, double &coef, double x_sd, double x_mean, double &restrict fill_val,
|
563
|
+
MissingAction missing_action, double *restrict buffer_arr,
|
564
|
+
size_t *restrict buffer_NAs, bool first_run, mapping &restrict w)
|
565
|
+
{
|
566
|
+
/* TODO: here don't need the buffer for NAs */
|
567
|
+
|
568
|
+
if (first_run)
|
569
|
+
coef /= x_sd;
|
570
|
+
|
571
|
+
size_t cnt = 0;
|
572
|
+
size_t cnt_NA = 0;
|
573
|
+
double *restrict res_write = res - st;
|
574
|
+
ldouble_safe cumw = 0;
|
575
|
+
double w_this;
|
576
|
+
/* TODO: these buffers should be allocated externally */
|
577
|
+
std::vector<double> obs_weight;
|
578
|
+
|
579
|
+
if (first_run && missing_action != Fail)
|
580
|
+
{
|
581
|
+
obs_weight.resize(end - st + 1, 0.);
|
582
|
+
}
|
583
|
+
|
584
|
+
if (missing_action == Fail)
|
585
|
+
{
|
586
|
+
for (size_t row = st; row <= end; row++)
|
587
|
+
res_write[row] = std::fma(x[ix_arr[row]] - x_mean, coef, res_write[row]);
|
588
|
+
}
|
589
|
+
|
590
|
+
else
|
591
|
+
{
|
592
|
+
if (first_run)
|
593
|
+
{
|
594
|
+
for (size_t row = st; row <= end; row++)
|
595
|
+
{
|
596
|
+
if (likely(!is_na_or_inf(x[ix_arr[row]])))
|
597
|
+
{
|
598
|
+
w_this = w[ix_arr[row]];
|
599
|
+
res_write[row] = std::fma(x[ix_arr[row]] - x_mean, coef, res_write[row]);
|
600
|
+
obs_weight[cnt] = w_this;
|
601
|
+
buffer_arr[cnt++] = x[ix_arr[row]];
|
602
|
+
cumw += w_this;
|
603
|
+
}
|
604
|
+
|
605
|
+
else
|
606
|
+
{
|
607
|
+
buffer_NAs[cnt_NA++] = row;
|
608
|
+
}
|
609
|
+
|
610
|
+
}
|
611
|
+
}
|
612
|
+
|
613
|
+
else
|
614
|
+
{
|
615
|
+
for (size_t row = st; row <= end; row++)
|
616
|
+
{
|
617
|
+
res_write[row] += (is_na_or_inf(x[ix_arr[row]]))? fill_val : ( (x[ix_arr[row]]-x_mean) * coef );
|
618
|
+
}
|
619
|
+
return;
|
620
|
+
}
|
621
|
+
|
622
|
+
|
623
|
+
ldouble_safe mid_point = cumw / (ldouble_safe)2;
|
624
|
+
std::vector<size_t> sorted_ix(cnt);
|
625
|
+
std::iota(sorted_ix.begin(), sorted_ix.end(), (size_t)0);
|
626
|
+
std::sort(sorted_ix.begin(), sorted_ix.end(),
|
627
|
+
[&buffer_arr](const size_t a, const size_t b){return buffer_arr[a] < buffer_arr[b];});
|
628
|
+
ldouble_safe currw = 0;
|
629
|
+
fill_val = buffer_arr[sorted_ix.back()]; /* <- will overwrite later */
|
630
|
+
/* TODO: is this median calculation correct? should it do a weighted interpolation? */
|
631
|
+
for (size_t ix = 0; ix < cnt; ix++)
|
632
|
+
{
|
633
|
+
currw += obs_weight[sorted_ix[ix]];
|
634
|
+
if (currw >= mid_point)
|
635
|
+
{
|
636
|
+
if (currw == mid_point && ix < cnt-1)
|
637
|
+
fill_val = buffer_arr[sorted_ix[ix]] + (buffer_arr[sorted_ix[ix+1]] - buffer_arr[sorted_ix[ix]]) / 2.0;
|
638
|
+
else
|
639
|
+
fill_val = buffer_arr[sorted_ix[ix]];
|
640
|
+
break;
|
641
|
+
}
|
642
|
+
}
|
643
|
+
|
644
|
+
fill_val = (fill_val - x_mean) * coef;
|
645
|
+
if (cnt_NA && fill_val)
|
646
|
+
{
|
647
|
+
for (size_t row = 0; row < cnt_NA; row++)
|
648
|
+
res_write[buffer_NAs[row]] += fill_val;
|
649
|
+
}
|
650
|
+
|
651
|
+
}
|
652
|
+
}
|
653
|
+
|
654
|
+
/* for sparse numerical */
|
655
|
+
template <class real_t_, class sparse_ix>
|
656
|
+
void add_linear_comb(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num, double *restrict res,
|
657
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
658
|
+
double &restrict coef, double x_sd, double x_mean, double &restrict fill_val, MissingAction missing_action,
|
659
|
+
double *restrict buffer_arr, size_t *restrict buffer_NAs, bool first_run)
|
660
|
+
{
|
661
|
+
/* ix_arr must be already sorted beforehand */
|
662
|
+
|
663
|
+
/* if it's all zeros, no need to do anything, but this is not supposed
|
664
|
+
to happen while fitting because the range is determined before calling this */
|
665
|
+
if (
|
666
|
+
Xc_indptr[col_num] == Xc_indptr[col_num + 1] ||
|
667
|
+
Xc_ind[Xc_indptr[col_num]] > (sparse_ix)ix_arr[end] ||
|
668
|
+
Xc_ind[Xc_indptr[col_num + 1] - 1] < (sparse_ix)ix_arr[st]
|
669
|
+
)
|
670
|
+
{
|
671
|
+
if (first_run)
|
672
|
+
{
|
673
|
+
coef /= x_sd;
|
674
|
+
if (missing_action != Fail)
|
675
|
+
fill_val = 0;
|
676
|
+
}
|
677
|
+
|
678
|
+
double *restrict res_write = res - st;
|
679
|
+
double offset = x_mean * coef;
|
680
|
+
if (offset)
|
681
|
+
{
|
682
|
+
for (size_t row = st; row <= end; row++)
|
683
|
+
res_write[row] -= offset;
|
684
|
+
}
|
685
|
+
|
686
|
+
return;
|
687
|
+
}
|
688
|
+
|
689
|
+
size_t st_col = Xc_indptr[col_num];
|
690
|
+
size_t end_col = Xc_indptr[col_num + 1] - 1;
|
691
|
+
size_t curr_pos = st_col;
|
692
|
+
size_t *ptr_st = std::lower_bound(ix_arr + st, ix_arr + end + 1, (size_t)Xc_ind[st_col]);
|
693
|
+
|
694
|
+
size_t cnt_non_NA = 0; /* when NAs need to be imputed */
|
695
|
+
size_t cnt_NA = 0; /* when NAs need to be imputed */
|
696
|
+
size_t n_sample = end - st + 1;
|
697
|
+
size_t *ix_arr_plus_st = ix_arr + st;
|
698
|
+
|
699
|
+
if (first_run)
|
700
|
+
coef /= x_sd;
|
701
|
+
|
702
|
+
double *restrict res_write = res - st;
|
703
|
+
double offset = x_mean * coef;
|
704
|
+
if (offset)
|
705
|
+
{
|
706
|
+
for (size_t row = st; row <= end; row++)
|
707
|
+
res_write[row] -= offset;
|
708
|
+
}
|
709
|
+
|
710
|
+
size_t ind_end_col = Xc_ind[end_col];
|
711
|
+
size_t nmatches = 0;
|
712
|
+
|
713
|
+
if (missing_action != Fail)
|
714
|
+
{
|
715
|
+
if (first_run)
|
716
|
+
{
|
717
|
+
for (size_t *row = ptr_st;
|
718
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
719
|
+
)
|
720
|
+
{
|
721
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
722
|
+
{
|
723
|
+
if (unlikely(is_na_or_inf(Xc[curr_pos])))
|
724
|
+
{
|
725
|
+
buffer_NAs[cnt_NA++] = row - ix_arr_plus_st;
|
726
|
+
}
|
727
|
+
|
728
|
+
else
|
729
|
+
{
|
730
|
+
buffer_arr[cnt_non_NA++] = Xc[curr_pos];
|
731
|
+
res[row - ix_arr_plus_st] = std::fma(Xc[curr_pos], coef, res[row - ix_arr_plus_st]);
|
732
|
+
}
|
733
|
+
|
734
|
+
nmatches++;
|
735
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
736
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
737
|
+
}
|
738
|
+
|
739
|
+
else
|
740
|
+
{
|
741
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
742
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
743
|
+
else
|
744
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
745
|
+
}
|
746
|
+
}
|
747
|
+
}
|
748
|
+
|
749
|
+
else
|
750
|
+
{
|
751
|
+
/* when impute value for missing has already been determined */
|
752
|
+
for (size_t *row = ptr_st;
|
753
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
754
|
+
)
|
755
|
+
{
|
756
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
757
|
+
{
|
758
|
+
res[row - ix_arr_plus_st] += is_na_or_inf(Xc[curr_pos])?
|
759
|
+
(fill_val + offset) : (Xc[curr_pos] * coef);
|
760
|
+
if (row == ix_arr + end) break;
|
761
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
762
|
+
}
|
763
|
+
|
764
|
+
else
|
765
|
+
{
|
766
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
767
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
768
|
+
else
|
769
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
770
|
+
}
|
771
|
+
}
|
772
|
+
|
773
|
+
return;
|
774
|
+
}
|
775
|
+
|
776
|
+
|
777
|
+
/* Determine imputation value */
|
778
|
+
std::sort(buffer_arr, buffer_arr + cnt_non_NA);
|
779
|
+
size_t mid_ceil = (n_sample - cnt_NA) / 2;
|
780
|
+
size_t nzeros = (end - st + 1) - nmatches;
|
781
|
+
if (nzeros > mid_ceil && buffer_arr[0] > 0)
|
782
|
+
{
|
783
|
+
fill_val = 0;
|
784
|
+
return;
|
785
|
+
}
|
786
|
+
|
787
|
+
else
|
788
|
+
{
|
789
|
+
size_t n_neg = (buffer_arr[0] > 0)?
|
790
|
+
0 : ((buffer_arr[cnt_non_NA - 1] < 0)?
|
791
|
+
cnt_non_NA : std::lower_bound(buffer_arr, buffer_arr + cnt_non_NA, (double)0) - buffer_arr);
|
792
|
+
|
793
|
+
|
794
|
+
if (n_neg < (mid_ceil-1) && n_neg + nzeros > mid_ceil)
|
795
|
+
{
|
796
|
+
fill_val = 0;
|
797
|
+
return;
|
798
|
+
}
|
799
|
+
|
800
|
+
else
|
801
|
+
{
|
802
|
+
/* if the sample size is odd, take the middle, otherwise take a simple average */
|
803
|
+
if (((n_sample - cnt_NA) % 2) != 0)
|
804
|
+
{
|
805
|
+
if (mid_ceil < n_neg)
|
806
|
+
fill_val = buffer_arr[mid_ceil];
|
807
|
+
else if (mid_ceil < n_neg + nzeros)
|
808
|
+
fill_val = 0;
|
809
|
+
else
|
810
|
+
fill_val = buffer_arr[mid_ceil - nzeros];
|
811
|
+
}
|
812
|
+
|
813
|
+
else
|
814
|
+
{
|
815
|
+
if (mid_ceil < n_neg)
|
816
|
+
{
|
817
|
+
fill_val = (buffer_arr[mid_ceil - 1] + buffer_arr[mid_ceil]) / 2;
|
818
|
+
}
|
819
|
+
|
820
|
+
else if (mid_ceil < n_neg + nzeros)
|
821
|
+
{
|
822
|
+
if (mid_ceil == n_neg)
|
823
|
+
fill_val = buffer_arr[mid_ceil - 1] / 2;
|
824
|
+
else
|
825
|
+
fill_val = 0;
|
826
|
+
}
|
827
|
+
|
828
|
+
else
|
829
|
+
{
|
830
|
+
if (mid_ceil == n_neg + nzeros && nzeros > 0)
|
831
|
+
fill_val = buffer_arr[n_neg] / 2;
|
832
|
+
else
|
833
|
+
fill_val = (buffer_arr[mid_ceil - nzeros - 1] + buffer_arr[mid_ceil - nzeros]) / 2; /* WRONG!!!! */
|
834
|
+
}
|
835
|
+
}
|
836
|
+
|
837
|
+
/* fill missing if any */
|
838
|
+
fill_val *= coef;
|
839
|
+
if (cnt_NA && fill_val)
|
840
|
+
for (size_t ix = 0; ix < cnt_NA; ix++)
|
841
|
+
res[buffer_NAs[ix]] += fill_val;
|
842
|
+
|
843
|
+
/* next time, it will need to have the offset added */
|
844
|
+
fill_val -= offset;
|
845
|
+
}
|
846
|
+
}
|
847
|
+
}
|
848
|
+
|
849
|
+
else /* no NAs */
|
850
|
+
{
|
851
|
+
for (size_t *row = ptr_st;
|
852
|
+
row != ix_arr + end + 1 && curr_pos != end_col + 1 && ind_end_col >= *row;
|
853
|
+
)
|
854
|
+
{
|
855
|
+
if (Xc_ind[curr_pos] == (sparse_ix)(*row))
|
856
|
+
{
|
857
|
+
res[row - ix_arr_plus_st] += Xc[curr_pos] * coef;
|
858
|
+
if (row == ix_arr + end || curr_pos == end_col) break;
|
859
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *(++row)) - Xc_ind;
|
860
|
+
}
|
861
|
+
|
862
|
+
else
|
863
|
+
{
|
864
|
+
if (Xc_ind[curr_pos] > (sparse_ix)(*row))
|
865
|
+
row = std::lower_bound(row + 1, ix_arr + end + 1, Xc_ind[curr_pos]);
|
866
|
+
else
|
867
|
+
curr_pos = std::lower_bound(Xc_ind + curr_pos + 1, Xc_ind + end_col + 1, *row) - Xc_ind;
|
868
|
+
}
|
869
|
+
}
|
870
|
+
}
|
871
|
+
}
|
872
|
+
|
873
|
+
template <class real_t_, class sparse_ix, class mapping, class ldouble_safe>
|
874
|
+
void add_linear_comb_weighted(size_t *restrict ix_arr, size_t st, size_t end, size_t col_num, double *restrict res,
|
875
|
+
real_t_ *restrict Xc, sparse_ix *restrict Xc_ind, sparse_ix *restrict Xc_indptr,
|
876
|
+
double &restrict coef, double x_sd, double x_mean, double &restrict fill_val, MissingAction missing_action,
|
877
|
+
double *restrict buffer_arr, size_t *restrict buffer_NAs, bool first_run, mapping &restrict w)
|
878
|
+
{
|
879
|
+
/* TODO: there's likely a better way of doing this directly with sparse inputs.
|
880
|
+
Think about some way of doing it efficiently. */
|
881
|
+
if (first_run && missing_action != Fail)
|
882
|
+
{
|
883
|
+
std::vector<double> denseX(end-st+1, 0.);
|
884
|
+
todense(ix_arr, st, end,
|
885
|
+
col_num, Xc, Xc_ind, Xc_indptr,
|
886
|
+
denseX.data());
|
887
|
+
std::vector<double> obs_weight(end-st+1);
|
888
|
+
for (size_t row = st; row <= end; row++)
|
889
|
+
obs_weight[row - st] = w[ix_arr[row]];
|
890
|
+
|
891
|
+
size_t end_new = end - st + 1;
|
892
|
+
for (size_t ix = 0; ix < end-st+1; ix++)
|
893
|
+
{
|
894
|
+
if (unlikely(is_na_or_inf(denseX[ix])))
|
895
|
+
{
|
896
|
+
std::swap(denseX[ix], denseX[--end_new]);
|
897
|
+
std::swap(obs_weight[ix], obs_weight[end_new]);
|
898
|
+
}
|
899
|
+
}
|
900
|
+
|
901
|
+
ldouble_safe cumw = std::accumulate(obs_weight.begin(), obs_weight.begin() + end_new, (ldouble_safe)0);
|
902
|
+
ldouble_safe mid_point = cumw / (ldouble_safe)2;
|
903
|
+
std::vector<size_t> sorted_ix(end_new);
|
904
|
+
std::iota(sorted_ix.begin(), sorted_ix.end(), (size_t)0);
|
905
|
+
std::sort(sorted_ix.begin(), sorted_ix.end(),
|
906
|
+
[&denseX](const size_t a, const size_t b){return denseX[a] < denseX[b];});
|
907
|
+
ldouble_safe currw = 0;
|
908
|
+
fill_val = denseX[sorted_ix.back()]; /* <- will overwrite later */
|
909
|
+
/* TODO: is this median calculation correct? should it do a weighted interpolation? */
|
910
|
+
for (size_t ix = 0; ix < end_new; ix++)
|
911
|
+
{
|
912
|
+
currw += obs_weight[sorted_ix[ix]];
|
913
|
+
if (currw >= mid_point)
|
914
|
+
{
|
915
|
+
if (currw == mid_point && ix < end_new-1)
|
916
|
+
fill_val = denseX[sorted_ix[ix]] + (denseX[sorted_ix[ix+1]] - denseX[sorted_ix[ix]]) / 2.0;
|
917
|
+
else
|
918
|
+
fill_val = denseX[sorted_ix[ix]];
|
919
|
+
break;
|
920
|
+
}
|
921
|
+
}
|
922
|
+
|
923
|
+
fill_val = (fill_val - x_mean) * (coef / x_sd);
|
924
|
+
denseX.clear();
|
925
|
+
obs_weight.clear();
|
926
|
+
sorted_ix.clear();
|
927
|
+
|
928
|
+
add_linear_comb(ix_arr, st, end, col_num, res,
|
929
|
+
Xc, Xc_ind, Xc_indptr,
|
930
|
+
coef, x_sd, x_mean, fill_val, missing_action,
|
931
|
+
buffer_arr, buffer_NAs, false);
|
932
|
+
}
|
933
|
+
|
934
|
+
else
|
935
|
+
{
|
936
|
+
add_linear_comb(ix_arr, st, end, col_num, res,
|
937
|
+
Xc, Xc_ind, Xc_indptr,
|
938
|
+
coef, x_sd, x_mean, fill_val, missing_action,
|
939
|
+
buffer_arr, buffer_NAs, first_run);
|
940
|
+
}
|
941
|
+
}
|
942
|
+
|
943
|
+
/* for categoricals */
|
944
|
+
template <class ldouble_safe>
|
945
|
+
void add_linear_comb(size_t *restrict ix_arr, size_t st, size_t end, double *restrict res,
|
946
|
+
int x[], int ncat, double *restrict cat_coef, double single_cat_coef, int chosen_cat,
|
947
|
+
double &restrict fill_val, double &restrict fill_new, size_t *restrict buffer_cnt, size_t *restrict buffer_pos,
|
948
|
+
NewCategAction new_cat_action, MissingAction missing_action, CategSplit cat_split_type, bool first_run)
|
949
|
+
{
|
950
|
+
double *restrict res_write = res - st;
|
951
|
+
switch(cat_split_type)
|
952
|
+
{
|
953
|
+
case SingleCateg:
|
954
|
+
{
|
955
|
+
/* in this case there's no need to make-up an impute value for new categories, only for NAs */
|
956
|
+
switch(missing_action)
|
957
|
+
{
|
958
|
+
case Fail:
|
959
|
+
{
|
960
|
+
for (size_t row = st; row <= end; row++)
|
961
|
+
res_write[row] += (x[ix_arr[row]] == chosen_cat)? single_cat_coef : 0;
|
962
|
+
return;
|
963
|
+
}
|
964
|
+
|
965
|
+
case Impute:
|
966
|
+
{
|
967
|
+
size_t cnt_NA = 0;
|
968
|
+
size_t cnt_this = 0;
|
969
|
+
size_t cnt = end - st + 1;
|
970
|
+
if (first_run)
|
971
|
+
{
|
972
|
+
for (size_t row = st; row <= end; row++)
|
973
|
+
{
|
974
|
+
if (unlikely(x[ix_arr[row]] < 0))
|
975
|
+
{
|
976
|
+
cnt_NA++;
|
977
|
+
}
|
978
|
+
|
979
|
+
else if (x[ix_arr[row]] == chosen_cat)
|
980
|
+
{
|
981
|
+
cnt_this++;
|
982
|
+
res_write[row] += single_cat_coef;
|
983
|
+
}
|
984
|
+
}
|
985
|
+
}
|
986
|
+
|
987
|
+
else
|
988
|
+
{
|
989
|
+
for (size_t row = st; row <= end; row++)
|
990
|
+
res_write[row] += (x[ix_arr[row]] < 0)? fill_val : ((x[ix_arr[row]] == chosen_cat)? single_cat_coef : 0);
|
991
|
+
return;
|
992
|
+
}
|
993
|
+
|
994
|
+
fill_val = (cnt_this > (cnt - cnt_NA - cnt_this))? single_cat_coef : 0;
|
995
|
+
if (cnt_NA && fill_val)
|
996
|
+
{
|
997
|
+
for (size_t row = st; row <= end; row++)
|
998
|
+
if (x[ix_arr[row]] < 0)
|
999
|
+
res_write[row] += fill_val;
|
1000
|
+
}
|
1001
|
+
return;
|
1002
|
+
}
|
1003
|
+
|
1004
|
+
default:
|
1005
|
+
{
|
1006
|
+
unexpected_error();
|
1007
|
+
break;
|
1008
|
+
}
|
1009
|
+
}
|
1010
|
+
}
|
1011
|
+
|
1012
|
+
case SubSet:
|
1013
|
+
{
|
1014
|
+
/* in this case, since the splits are by more than 1 variable, it's not possible to
|
1015
|
+
divide missing/new categoricals by assigning weights, so they have to be imputed
|
1016
|
+
in both cases, unless using random weights for the new ones, in which case they won't
|
1017
|
+
need to be imputed for new, but sill need it for NA */
|
1018
|
+
|
1019
|
+
if (new_cat_action == Random && missing_action == Fail)
|
1020
|
+
{
|
1021
|
+
for (size_t row = st; row <= end; row++)
|
1022
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1023
|
+
return;
|
1024
|
+
}
|
1025
|
+
|
1026
|
+
if (!first_run)
|
1027
|
+
{
|
1028
|
+
if (missing_action == Fail)
|
1029
|
+
{
|
1030
|
+
for (size_t row = st; row <= end; row++)
|
1031
|
+
res_write[row] += (x[ix_arr[row]] >= ncat)? fill_new : cat_coef[x[ix_arr[row]]];
|
1032
|
+
}
|
1033
|
+
|
1034
|
+
else
|
1035
|
+
{
|
1036
|
+
for (size_t row = st; row <= end; row++)
|
1037
|
+
res_write[row] += (x[ix_arr[row]] < 0)? fill_val : ((x[ix_arr[row]] >= ncat)? fill_new : cat_coef[x[ix_arr[row]]]);
|
1038
|
+
}
|
1039
|
+
return;
|
1040
|
+
}
|
1041
|
+
|
1042
|
+
std::fill(buffer_cnt, buffer_cnt + ncat + 1, 0);
|
1043
|
+
switch(missing_action)
|
1044
|
+
{
|
1045
|
+
case Fail:
|
1046
|
+
{
|
1047
|
+
for (size_t row = st; row <= end; row++)
|
1048
|
+
{
|
1049
|
+
buffer_cnt[x[ix_arr[row]]]++;
|
1050
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1051
|
+
}
|
1052
|
+
break;
|
1053
|
+
}
|
1054
|
+
|
1055
|
+
default:
|
1056
|
+
{
|
1057
|
+
for (size_t row = st; row <= end; row++)
|
1058
|
+
{
|
1059
|
+
if (x[ix_arr[row]] >= 0)
|
1060
|
+
{
|
1061
|
+
buffer_cnt[x[ix_arr[row]]]++;
|
1062
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1063
|
+
}
|
1064
|
+
|
1065
|
+
else
|
1066
|
+
{
|
1067
|
+
buffer_cnt[ncat]++;
|
1068
|
+
}
|
1069
|
+
|
1070
|
+
}
|
1071
|
+
break;
|
1072
|
+
}
|
1073
|
+
}
|
1074
|
+
|
1075
|
+
switch(new_cat_action)
|
1076
|
+
{
|
1077
|
+
case Smallest:
|
1078
|
+
{
|
1079
|
+
size_t smallest = SIZE_MAX;
|
1080
|
+
int cat_smallest = 0;
|
1081
|
+
for (int cat = 0; cat < ncat; cat++)
|
1082
|
+
{
|
1083
|
+
if (buffer_cnt[cat] > 0 && buffer_cnt[cat] < smallest)
|
1084
|
+
{
|
1085
|
+
smallest = buffer_cnt[cat];
|
1086
|
+
cat_smallest = cat;
|
1087
|
+
}
|
1088
|
+
}
|
1089
|
+
fill_new = cat_coef[cat_smallest];
|
1090
|
+
if (missing_action == Fail) break;
|
1091
|
+
}
|
1092
|
+
|
1093
|
+
default:
|
1094
|
+
{
|
1095
|
+
/* Determine imputation value as the category in sorted order that gives 50% + 1 */
|
1096
|
+
ldouble_safe cnt_l = (ldouble_safe)((end - st + 1) - buffer_cnt[ncat]);
|
1097
|
+
std::iota(buffer_pos, buffer_pos + ncat, (size_t)0);
|
1098
|
+
std::sort(buffer_pos, buffer_pos + ncat, [&cat_coef](const size_t a, const size_t b){return cat_coef[a] < cat_coef[b];});
|
1099
|
+
|
1100
|
+
double cumprob = 0;
|
1101
|
+
int cat;
|
1102
|
+
for (cat = 0; cat < ncat; cat++)
|
1103
|
+
{
|
1104
|
+
cumprob += (ldouble_safe)buffer_cnt[buffer_pos[cat]] / cnt_l;
|
1105
|
+
if (cumprob >= .5) break;
|
1106
|
+
}
|
1107
|
+
// cat = std::min(cat, ncat); /* in case it picks the last one */
|
1108
|
+
fill_val = cat_coef[buffer_pos[cat]];
|
1109
|
+
if (new_cat_action != Smallest)
|
1110
|
+
fill_new = fill_val;
|
1111
|
+
|
1112
|
+
if (buffer_cnt[ncat] > 0 && fill_val) /* NAs */
|
1113
|
+
for (size_t row = st; row <= end; row++)
|
1114
|
+
if (unlikely(x[ix_arr[row]] < 0))
|
1115
|
+
res_write[row] += fill_val;
|
1116
|
+
}
|
1117
|
+
}
|
1118
|
+
|
1119
|
+
/* now fill unseen categories */
|
1120
|
+
if (new_cat_action != Random)
|
1121
|
+
for (int cat = 0; cat < ncat; cat++)
|
1122
|
+
if (!buffer_cnt[cat])
|
1123
|
+
cat_coef[cat] = fill_new;
|
1124
|
+
|
1125
|
+
}
|
1126
|
+
}
|
1127
|
+
}
|
1128
|
+
|
1129
|
+
template <class mapping, class ldouble_safe>
|
1130
|
+
void add_linear_comb_weighted(size_t *restrict ix_arr, size_t st, size_t end, double *restrict res,
|
1131
|
+
int x[], int ncat, double *restrict cat_coef, double single_cat_coef, int chosen_cat,
|
1132
|
+
double &restrict fill_val, double &restrict fill_new, size_t *restrict buffer_pos,
|
1133
|
+
NewCategAction new_cat_action, MissingAction missing_action, CategSplit cat_split_type,
|
1134
|
+
bool first_run, mapping &restrict w)
|
1135
|
+
{
|
1136
|
+
double *restrict res_write = res - st;
|
1137
|
+
/* TODO: this buffer should be allocated externally */
|
1138
|
+
|
1139
|
+
switch(cat_split_type)
|
1140
|
+
{
|
1141
|
+
case SingleCateg:
|
1142
|
+
{
|
1143
|
+
/* in this case there's no need to make-up an impute value for new categories, only for NAs */
|
1144
|
+
switch(missing_action)
|
1145
|
+
{
|
1146
|
+
case Fail:
|
1147
|
+
{
|
1148
|
+
for (size_t row = st; row <= end; row++)
|
1149
|
+
res_write[row] += (x[ix_arr[row]] == chosen_cat)? single_cat_coef : 0;
|
1150
|
+
return;
|
1151
|
+
}
|
1152
|
+
|
1153
|
+
case Impute:
|
1154
|
+
{
|
1155
|
+
bool has_NA = false;
|
1156
|
+
ldouble_safe cnt_this = 0;
|
1157
|
+
ldouble_safe cnt_other = 0;
|
1158
|
+
if (first_run)
|
1159
|
+
{
|
1160
|
+
for (size_t row = st; row <= end; row++)
|
1161
|
+
{
|
1162
|
+
if (unlikely(x[ix_arr[row]] < 0))
|
1163
|
+
{
|
1164
|
+
has_NA = true;
|
1165
|
+
}
|
1166
|
+
|
1167
|
+
else if (x[ix_arr[row]] == chosen_cat)
|
1168
|
+
{
|
1169
|
+
cnt_this += w[ix_arr[row]];
|
1170
|
+
res_write[row] += single_cat_coef;
|
1171
|
+
}
|
1172
|
+
|
1173
|
+
else
|
1174
|
+
{
|
1175
|
+
cnt_other += w[ix_arr[row]];
|
1176
|
+
}
|
1177
|
+
}
|
1178
|
+
}
|
1179
|
+
|
1180
|
+
else
|
1181
|
+
{
|
1182
|
+
for (size_t row = st; row <= end; row++)
|
1183
|
+
res_write[row] += (x[ix_arr[row]] < 0)? fill_val : ((x[ix_arr[row]] == chosen_cat)? single_cat_coef : 0);
|
1184
|
+
return;
|
1185
|
+
}
|
1186
|
+
|
1187
|
+
fill_val = (cnt_this > cnt_other)? single_cat_coef : 0;
|
1188
|
+
if (has_NA && fill_val)
|
1189
|
+
{
|
1190
|
+
for (size_t row = st; row <= end; row++)
|
1191
|
+
if (unlikely(x[ix_arr[row]] < 0))
|
1192
|
+
res_write[row] += fill_val;
|
1193
|
+
}
|
1194
|
+
return;
|
1195
|
+
}
|
1196
|
+
|
1197
|
+
default:
|
1198
|
+
{
|
1199
|
+
unexpected_error();
|
1200
|
+
break;
|
1201
|
+
}
|
1202
|
+
}
|
1203
|
+
}
|
1204
|
+
|
1205
|
+
case SubSet:
|
1206
|
+
{
|
1207
|
+
/* in this case, since the splits are by more than 1 variable, it's not possible to
|
1208
|
+
divide missing/new categoricals by assigning weights, so they have to be imputed
|
1209
|
+
in both cases, unless using random weights for the new ones, in which case they won't
|
1210
|
+
need to be imputed for new, but sill need it for NA */
|
1211
|
+
|
1212
|
+
if (new_cat_action == Random && missing_action == Fail)
|
1213
|
+
{
|
1214
|
+
for (size_t row = st; row <= end; row++)
|
1215
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1216
|
+
return;
|
1217
|
+
}
|
1218
|
+
|
1219
|
+
if (!first_run)
|
1220
|
+
{
|
1221
|
+
if (missing_action == Fail)
|
1222
|
+
{
|
1223
|
+
for (size_t row = st; row <= end; row++)
|
1224
|
+
res_write[row] += (x[ix_arr[row]] >= ncat)? fill_new : cat_coef[x[ix_arr[row]]];
|
1225
|
+
}
|
1226
|
+
|
1227
|
+
else
|
1228
|
+
{
|
1229
|
+
for (size_t row = st; row <= end; row++)
|
1230
|
+
res_write[row] += (x[ix_arr[row]] < 0)? fill_val : ((x[ix_arr[row]] >= ncat)? fill_new : cat_coef[x[ix_arr[row]]]);
|
1231
|
+
}
|
1232
|
+
return;
|
1233
|
+
}
|
1234
|
+
|
1235
|
+
/* TODO: this buffer should be allocated externally */
|
1236
|
+
std::vector<ldouble_safe> buffer_cnt(ncat+1, 0.);
|
1237
|
+
switch(missing_action)
|
1238
|
+
{
|
1239
|
+
case Fail:
|
1240
|
+
{
|
1241
|
+
for (size_t row = st; row <= end; row++)
|
1242
|
+
{
|
1243
|
+
buffer_cnt[x[ix_arr[row]]] += w[ix_arr[row]];
|
1244
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1245
|
+
}
|
1246
|
+
break;
|
1247
|
+
}
|
1248
|
+
|
1249
|
+
default:
|
1250
|
+
{
|
1251
|
+
for (size_t row = st; row <= end; row++)
|
1252
|
+
{
|
1253
|
+
if (likely(x[ix_arr[row]] >= 0))
|
1254
|
+
{
|
1255
|
+
buffer_cnt[x[ix_arr[row]]] += w[ix_arr[row]];
|
1256
|
+
res_write[row] += cat_coef[x[ix_arr[row]]];
|
1257
|
+
}
|
1258
|
+
|
1259
|
+
else
|
1260
|
+
{
|
1261
|
+
buffer_cnt[ncat] += w[ix_arr[row]];
|
1262
|
+
}
|
1263
|
+
|
1264
|
+
}
|
1265
|
+
break;
|
1266
|
+
}
|
1267
|
+
}
|
1268
|
+
|
1269
|
+
switch(new_cat_action)
|
1270
|
+
{
|
1271
|
+
case Smallest:
|
1272
|
+
{
|
1273
|
+
ldouble_safe smallest = std::numeric_limits<ldouble_safe>::infinity();
|
1274
|
+
int cat_smallest = 0;
|
1275
|
+
for (int cat = 0; cat < ncat; cat++)
|
1276
|
+
{
|
1277
|
+
if (buffer_cnt[cat] > 0 && buffer_cnt[cat] < smallest)
|
1278
|
+
{
|
1279
|
+
smallest = buffer_cnt[cat];
|
1280
|
+
cat_smallest = cat;
|
1281
|
+
}
|
1282
|
+
}
|
1283
|
+
fill_new = cat_coef[cat_smallest];
|
1284
|
+
if (missing_action == Fail) break;
|
1285
|
+
}
|
1286
|
+
|
1287
|
+
default:
|
1288
|
+
{
|
1289
|
+
/* Determine imputation value as the category in sorted order that gives 50% + 1 */
|
1290
|
+
ldouble_safe cnt_l = std::accumulate(buffer_cnt.begin(), buffer_cnt.begin() + ncat, (ldouble_safe)0);
|
1291
|
+
std::iota(buffer_pos, buffer_pos + ncat, (size_t)0);
|
1292
|
+
std::sort(buffer_pos, buffer_pos + ncat, [&cat_coef](const size_t a, const size_t b){return cat_coef[a] < cat_coef[b];});
|
1293
|
+
|
1294
|
+
double cumprob = 0;
|
1295
|
+
int cat;
|
1296
|
+
for (cat = 0; cat < ncat; cat++)
|
1297
|
+
{
|
1298
|
+
cumprob += buffer_cnt[buffer_pos[cat]] / cnt_l;
|
1299
|
+
if (cumprob >= .5) break;
|
1300
|
+
}
|
1301
|
+
// cat = std::min(cat, ncat); /* in case it picks the last one */
|
1302
|
+
fill_val = cat_coef[buffer_pos[cat]];
|
1303
|
+
if (new_cat_action != Smallest)
|
1304
|
+
fill_new = fill_val;
|
1305
|
+
|
1306
|
+
if (buffer_cnt[ncat] > 0 && fill_val) /* NAs */
|
1307
|
+
for (size_t row = st; row <= end; row++)
|
1308
|
+
if (unlikely(x[ix_arr[row]] < 0))
|
1309
|
+
res_write[row] += fill_val;
|
1310
|
+
}
|
1311
|
+
}
|
1312
|
+
|
1313
|
+
/* now fill unseen categories */
|
1314
|
+
if (new_cat_action != Random)
|
1315
|
+
for (int cat = 0; cat < ncat; cat++)
|
1316
|
+
if (!buffer_cnt[cat])
|
1317
|
+
cat_coef[cat] = fill_new;
|
1318
|
+
|
1319
|
+
}
|
1320
|
+
}
|
1321
|
+
}
|