rumale 0.22.5 → 0.23.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +21 -0
- data/LICENSE.txt +1 -1
- data/README.md +34 -2
- data/ext/rumale/extconf.rb +1 -1
- data/ext/rumale/{tree.c → rumaleext.c} +51 -85
- data/ext/rumale/{tree.h → rumaleext.h} +5 -5
- data/lib/rumale/clustering/hdbscan.rb +28 -8
- data/lib/rumale/clustering/single_linkage.rb +23 -5
- data/lib/rumale/decomposition/fast_ica.rb +1 -1
- data/lib/rumale/ensemble/gradient_boosting_classifier.rb +2 -2
- data/lib/rumale/ensemble/gradient_boosting_regressor.rb +1 -1
- data/lib/rumale/ensemble/random_forest_classifier.rb +1 -1
- data/lib/rumale/ensemble/random_forest_regressor.rb +1 -1
- data/lib/rumale/evaluation_measure/roc_auc.rb +1 -2
- data/lib/rumale/kernel_approximation/nystroem.rb +1 -1
- data/lib/rumale/kernel_machine/kernel_svc.rb +1 -1
- data/lib/rumale/linear_model/linear_regression.rb +5 -3
- data/lib/rumale/linear_model/ridge.rb +3 -3
- data/lib/rumale/model_selection/grid_search_cv.rb +3 -3
- data/lib/rumale/naive_bayes/bernoulli_nb.rb +1 -1
- data/lib/rumale/naive_bayes/gaussian_nb.rb +1 -1
- data/lib/rumale/naive_bayes/multinomial_nb.rb +1 -1
- data/lib/rumale/nearest_neighbors/vp_tree.rb +2 -0
- data/lib/rumale/tree/base_decision_tree.rb +15 -10
- data/lib/rumale/tree/decision_tree_classifier.rb +14 -11
- data/lib/rumale/tree/decision_tree_regressor.rb +0 -1
- data/lib/rumale/tree/gradient_tree_regressor.rb +15 -11
- data/lib/rumale/version.rb +1 -1
- data/lib/rumale.rb +132 -133
- metadata +6 -17
- data/.coveralls.yml +0 -1
- data/.github/workflows/build.yml +0 -26
- data/.github/workflows/coverage.yml +0 -28
- data/.gitignore +0 -23
- data/.rspec +0 -3
- data/.rubocop.yml +0 -93
- data/.travis.yml +0 -17
- data/Gemfile +0 -17
- data/Rakefile +0 -14
- data/ext/rumale/rumale.c +0 -10
- data/ext/rumale/rumale.h +0 -8
- data/rumale.gemspec +0 -49
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 4564c37af7744bc4fe14dec5c5fc1e236687c3a241d2e17ef2d89f1c57056af9
|
4
|
+
data.tar.gz: 6f70d79a10b890bbd127f60f1c7f26934fcd88f71458af8839ac049b7a07efc8
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 5671a08ac8e9881f51896c4478ce5f4b54457c83d9b7194623febfd1859123cda5947c0d344aa551686c2c964359e9bdbd5ad13e9c921d2a3393a76717c00093
|
7
|
+
data.tar.gz: bb022827e8ca9d939addb9cfdd9b5fa5b643cd56150a84f41a224dde0c75992badbf792f77d06194943f015572beb5bafdd3c84e43efd98be5cc53beb9347ab0
|
data/CHANGELOG.md
CHANGED
@@ -1,3 +1,24 @@
|
|
1
|
+
# 0.23.2
|
2
|
+
Rumale project will be rebooted on version 0.24.0.
|
3
|
+
This version is probably the last release of the series starting with version 0.8.0.
|
4
|
+
|
5
|
+
- Refactor some codes and configs.
|
6
|
+
- Deprecate VPTree class.
|
7
|
+
|
8
|
+
# 0.23.1
|
9
|
+
- Fix all estimators to return inference results in a contiguous narray.
|
10
|
+
- Fix to use until statement instead of recursive call on apply methods of tree estimators.
|
11
|
+
- Rename native extension files.
|
12
|
+
- Introduce clang-format for native extension codes.
|
13
|
+
|
14
|
+
# 0.23.0
|
15
|
+
## Breaking change
|
16
|
+
- Change automalically selected solver from sgd to lbfgs in
|
17
|
+
[LinearRegression](https://yoshoku.github.io/rumale/doc/Rumale/LinearModel/LinearRegression.html) and
|
18
|
+
[Ridge](https://yoshoku.github.io/rumale/doc/Rumale/LinearModel/Ridge.html).
|
19
|
+
- When given 'auto' to solver parameter, these estimator select the 'svd' solver if Numo::Linalg is loaded.
|
20
|
+
Otherwise, they select the 'lbfgs' solver.
|
21
|
+
|
1
22
|
# 0.22.5
|
2
23
|
- Add transformer class for calculating kernel matrix.
|
3
24
|
- [KernelCalculator](https://yoshoku.github.io/rumale/doc/Rumale/Preprocessing/KernelCalculator.html)
|
data/LICENSE.txt
CHANGED
data/README.md
CHANGED
@@ -1,9 +1,10 @@
|
|
1
1
|
# Rumale
|
2
2
|
|
3
|
+
**This project is suspended for the author's health reasons. It will be resumed when the author recovers.**
|
4
|
+
|
3
5
|
![Rumale](https://dl.dropboxusercontent.com/s/joxruk2720ur66o/rumale_header_400.png)
|
4
6
|
|
5
|
-
[![Build Status](https://github.com/yoshoku/rumale/workflows/build/badge.svg)](https://github.com/yoshoku/rumale/actions
|
6
|
-
[![Coverage Status](https://coveralls.io/repos/github/yoshoku/rumale/badge.svg?branch=main)](https://coveralls.io/github/yoshoku/rumale?branch=main)
|
7
|
+
[![Build Status](https://github.com/yoshoku/rumale/actions/workflows/build.yml/badge.svg)](https://github.com/yoshoku/rumale/actions/workflows/build.yml)
|
7
8
|
[![Gem Version](https://badge.fury.io/rb/rumale.svg)](https://badge.fury.io/rb/rumale)
|
8
9
|
[![BSD 2-Clause License](https://img.shields.io/badge/License-BSD%202--Clause-orange.svg)](https://github.com/yoshoku/rumale/blob/main/LICENSE.txt)
|
9
10
|
[![Documentation](https://img.shields.io/badge/api-reference-blue.svg)](https://yoshoku.github.io/rumale/doc/)
|
@@ -189,6 +190,12 @@ Ubuntu:
|
|
189
190
|
$ sudo apt-get install libopenblas-dev liblapacke-dev
|
190
191
|
```
|
191
192
|
|
193
|
+
Fedora:
|
194
|
+
|
195
|
+
```bash
|
196
|
+
$ sudo dnf install openblas-devel lapack-devel
|
197
|
+
```
|
198
|
+
|
192
199
|
Windows (MSYS2):
|
193
200
|
|
194
201
|
```bash
|
@@ -226,6 +233,12 @@ Ubuntu:
|
|
226
233
|
$ sudo apt-get install gcc gfortran make
|
227
234
|
```
|
228
235
|
|
236
|
+
Fedora:
|
237
|
+
|
238
|
+
```bash
|
239
|
+
$ sudo dnf install gcc gcc-gfortran make
|
240
|
+
```
|
241
|
+
|
229
242
|
Install Numo::OpenBLAS gem.
|
230
243
|
|
231
244
|
```bash
|
@@ -239,6 +252,25 @@ require 'numo/openblas'
|
|
239
252
|
require 'rumale'
|
240
253
|
```
|
241
254
|
|
255
|
+
### Numo::BLIS
|
256
|
+
[Numo::BLIS](https://github.com/yoshoku/numo-blis) downloads and builds BLIS during installation
|
257
|
+
and uses that as a background library for Numo::Linalg.
|
258
|
+
BLIS is one of the high-performance BLAS as with OpenBLAS,
|
259
|
+
and using that can be expected to speed up of processing in Rumale.
|
260
|
+
|
261
|
+
Install Numo::BLIS gem.
|
262
|
+
|
263
|
+
```bash
|
264
|
+
$ gem install numo-blis
|
265
|
+
```
|
266
|
+
|
267
|
+
Load Numo::BLIS gem instead of Numo::Linalg.
|
268
|
+
|
269
|
+
```ruby
|
270
|
+
require 'numo/blis'
|
271
|
+
require 'rumale'
|
272
|
+
```
|
273
|
+
|
242
274
|
### Parallel
|
243
275
|
Several estimators in Rumale support parallel processing.
|
244
276
|
Parallel processing in Rumale is realized by [Parallel](https://github.com/grosser/parallel) gem,
|
data/ext/rumale/extconf.rb
CHANGED
@@ -1,18 +1,12 @@
|
|
1
|
-
#include "
|
1
|
+
#include "rumaleext.h"
|
2
2
|
|
3
|
-
|
4
|
-
|
5
|
-
double*
|
6
|
-
alloc_dbl_array(const long n_dimensions)
|
7
|
-
{
|
3
|
+
double* alloc_dbl_array(const long n_dimensions) {
|
8
4
|
double* arr = ALLOC_N(double, n_dimensions);
|
9
5
|
memset(arr, 0, n_dimensions * sizeof(double));
|
10
6
|
return arr;
|
11
7
|
}
|
12
8
|
|
13
|
-
double
|
14
|
-
calc_gini_coef(double* histogram, const long n_elements, const long n_classes)
|
15
|
-
{
|
9
|
+
double calc_gini_coef(double* histogram, const long n_elements, const long n_classes) {
|
16
10
|
long i;
|
17
11
|
double el;
|
18
12
|
double gini = 0.0;
|
@@ -25,9 +19,7 @@ calc_gini_coef(double* histogram, const long n_elements, const long n_classes)
|
|
25
19
|
return 1.0 - gini;
|
26
20
|
}
|
27
21
|
|
28
|
-
double
|
29
|
-
calc_entropy(double* histogram, const long n_elements, const long n_classes)
|
30
|
-
{
|
22
|
+
double calc_entropy(double* histogram, const long n_elements, const long n_classes) {
|
31
23
|
long i;
|
32
24
|
double el;
|
33
25
|
double entropy = 0.0;
|
@@ -41,8 +33,7 @@ calc_entropy(double* histogram, const long n_elements, const long n_classes)
|
|
41
33
|
}
|
42
34
|
|
43
35
|
VALUE
|
44
|
-
calc_mean_vec(double* sum_vec, const long n_dimensions, const long n_elements)
|
45
|
-
{
|
36
|
+
calc_mean_vec(double* sum_vec, const long n_dimensions, const long n_elements) {
|
46
37
|
long i;
|
47
38
|
VALUE mean_vec = rb_ary_new2(n_dimensions);
|
48
39
|
|
@@ -53,9 +44,7 @@ calc_mean_vec(double* sum_vec, const long n_dimensions, const long n_elements)
|
|
53
44
|
return mean_vec;
|
54
45
|
}
|
55
46
|
|
56
|
-
double
|
57
|
-
calc_vec_mae(VALUE vec_a, VALUE vec_b)
|
58
|
-
{
|
47
|
+
double calc_vec_mae(VALUE vec_a, VALUE vec_b) {
|
59
48
|
long i;
|
60
49
|
const long n_dimensions = RARRAY_LEN(vec_a);
|
61
50
|
double sum = 0.0;
|
@@ -69,9 +58,7 @@ calc_vec_mae(VALUE vec_a, VALUE vec_b)
|
|
69
58
|
return sum / n_dimensions;
|
70
59
|
}
|
71
60
|
|
72
|
-
double
|
73
|
-
calc_vec_mse(VALUE vec_a, VALUE vec_b)
|
74
|
-
{
|
61
|
+
double calc_vec_mse(VALUE vec_a, VALUE vec_b) {
|
75
62
|
long i;
|
76
63
|
const long n_dimensions = RARRAY_LEN(vec_a);
|
77
64
|
double sum = 0.0;
|
@@ -85,9 +72,7 @@ calc_vec_mse(VALUE vec_a, VALUE vec_b)
|
|
85
72
|
return sum / n_dimensions;
|
86
73
|
}
|
87
74
|
|
88
|
-
double
|
89
|
-
calc_mae(VALUE target_vecs, VALUE mean_vec)
|
90
|
-
{
|
75
|
+
double calc_mae(VALUE target_vecs, VALUE mean_vec) {
|
91
76
|
long i;
|
92
77
|
const long n_elements = RARRAY_LEN(target_vecs);
|
93
78
|
double sum = 0.0;
|
@@ -99,9 +84,7 @@ calc_mae(VALUE target_vecs, VALUE mean_vec)
|
|
99
84
|
return sum / n_elements;
|
100
85
|
}
|
101
86
|
|
102
|
-
double
|
103
|
-
calc_mse(VALUE target_vecs, VALUE mean_vec)
|
104
|
-
{
|
87
|
+
double calc_mse(VALUE target_vecs, VALUE mean_vec) {
|
105
88
|
long i;
|
106
89
|
const long n_elements = RARRAY_LEN(target_vecs);
|
107
90
|
double sum = 0.0;
|
@@ -113,18 +96,14 @@ calc_mse(VALUE target_vecs, VALUE mean_vec)
|
|
113
96
|
return sum / n_elements;
|
114
97
|
}
|
115
98
|
|
116
|
-
double
|
117
|
-
calc_impurity_cls(const char* criterion, double* histogram, const long n_elements, const long n_classes)
|
118
|
-
{
|
99
|
+
double calc_impurity_cls(const char* criterion, double* histogram, const long n_elements, const long n_classes) {
|
119
100
|
if (strcmp(criterion, "entropy") == 0) {
|
120
101
|
return calc_entropy(histogram, n_elements, n_classes);
|
121
102
|
}
|
122
103
|
return calc_gini_coef(histogram, n_elements, n_classes);
|
123
104
|
}
|
124
105
|
|
125
|
-
double
|
126
|
-
calc_impurity_reg(const char* criterion, VALUE target_vecs, double* sum_vec)
|
127
|
-
{
|
106
|
+
double calc_impurity_reg(const char* criterion, VALUE target_vecs, double* sum_vec) {
|
128
107
|
const long n_elements = RARRAY_LEN(target_vecs);
|
129
108
|
const long n_dimensions = RARRAY_LEN(rb_ary_entry(target_vecs, 0));
|
130
109
|
VALUE mean_vec = calc_mean_vec(sum_vec, n_dimensions, n_elements);
|
@@ -135,9 +114,7 @@ calc_impurity_reg(const char* criterion, VALUE target_vecs, double* sum_vec)
|
|
135
114
|
return calc_mse(target_vecs, mean_vec);
|
136
115
|
}
|
137
116
|
|
138
|
-
void
|
139
|
-
add_sum_vec(double* sum_vec, VALUE target)
|
140
|
-
{
|
117
|
+
void add_sum_vec(double* sum_vec, VALUE target) {
|
141
118
|
long i;
|
142
119
|
const long n_dimensions = RARRAY_LEN(target);
|
143
120
|
|
@@ -146,9 +123,7 @@ add_sum_vec(double* sum_vec, VALUE target)
|
|
146
123
|
}
|
147
124
|
}
|
148
125
|
|
149
|
-
void
|
150
|
-
sub_sum_vec(double* sum_vec, VALUE target)
|
151
|
-
{
|
126
|
+
void sub_sum_vec(double* sum_vec, VALUE target) {
|
152
127
|
long i;
|
153
128
|
const long n_dimensions = RARRAY_LEN(target);
|
154
129
|
|
@@ -168,9 +143,7 @@ typedef struct {
|
|
168
143
|
/**
|
169
144
|
* @!visibility private
|
170
145
|
*/
|
171
|
-
static void
|
172
|
-
iter_find_split_params_cls(na_loop_t const* lp)
|
173
|
-
{
|
146
|
+
static void iter_find_split_params_cls(na_loop_t const* lp) {
|
174
147
|
const int32_t* o = (int32_t*)NDL_PTR(lp, 0);
|
175
148
|
const double* f = (double*)NDL_PTR(lp, 1);
|
176
149
|
const int32_t* y = (int32_t*)NDL_PTR(lp, 2);
|
@@ -200,7 +173,9 @@ iter_find_split_params_cls(na_loop_t const* lp)
|
|
200
173
|
params[3] = 0.0; /* gain */
|
201
174
|
|
202
175
|
/* Initialize child node variables. */
|
203
|
-
for (i = 0; i < n_elements; i++) {
|
176
|
+
for (i = 0; i < n_elements; i++) {
|
177
|
+
r_histogram[y[o[i]]] += 1.0;
|
178
|
+
}
|
204
179
|
|
205
180
|
/* Find optimal parameters. */
|
206
181
|
while (curr_pos < n_elements && curr_el != last_el) {
|
@@ -246,14 +221,13 @@ iter_find_split_params_cls(na_loop_t const* lp)
|
|
246
221
|
* @param n_classes [Integer] The number of classes.
|
247
222
|
* @return [Array<Float>] The array consists of optimal parameters including impurities of child nodes, threshold, and gain.
|
248
223
|
*/
|
249
|
-
static VALUE
|
250
|
-
|
251
|
-
{
|
252
|
-
|
253
|
-
|
254
|
-
|
255
|
-
|
256
|
-
split_opts_cls opts = { StringValuePtr(criterion), NUM2LONG(n_classes), NUM2DBL(impurity) };
|
224
|
+
static VALUE find_split_params_cls(VALUE self, VALUE criterion, VALUE impurity, VALUE order, VALUE features, VALUE labels,
|
225
|
+
VALUE n_classes) {
|
226
|
+
ndfunc_arg_in_t ain[3] = {{numo_cInt32, 1}, {numo_cDFloat, 1}, {numo_cInt32, 1}};
|
227
|
+
size_t out_shape[1] = {4};
|
228
|
+
ndfunc_arg_out_t aout[1] = {{numo_cDFloat, 1, out_shape}};
|
229
|
+
ndfunc_t ndf = {(na_iter_func_t)iter_find_split_params_cls, NO_LOOP, 3, 1, ain, aout};
|
230
|
+
split_opts_cls opts = {StringValuePtr(criterion), NUM2LONG(n_classes), NUM2DBL(impurity)};
|
257
231
|
VALUE params = na_ndloop3(&ndf, &opts, 3, order, features, labels);
|
258
232
|
VALUE results = rb_ary_new2(4);
|
259
233
|
double* params_ptr = (double*)na_get_pointer_for_read(params);
|
@@ -276,9 +250,7 @@ typedef struct {
|
|
276
250
|
/**
|
277
251
|
* @!visibility private
|
278
252
|
*/
|
279
|
-
static void
|
280
|
-
iter_find_split_params_reg(na_loop_t const* lp)
|
281
|
-
{
|
253
|
+
static void iter_find_split_params_reg(na_loop_t const* lp) {
|
282
254
|
const int32_t* o = (int32_t*)NDL_PTR(lp, 0);
|
283
255
|
const double* f = (double*)NDL_PTR(lp, 1);
|
284
256
|
const double* y = (double*)NDL_PTR(lp, 2);
|
@@ -367,14 +339,12 @@ iter_find_split_params_reg(na_loop_t const* lp)
|
|
367
339
|
* @param targets [Numo::DFloat] (shape: [n_samples, n_outputs]) The target values.
|
368
340
|
* @return [Array<Float>] The array consists of optimal parameters including impurities of child nodes, threshold, and gain.
|
369
341
|
*/
|
370
|
-
static VALUE
|
371
|
-
|
372
|
-
{
|
373
|
-
|
374
|
-
|
375
|
-
|
376
|
-
ndfunc_t ndf = { (na_iter_func_t)iter_find_split_params_reg, NO_LOOP, 3, 1, ain, aout };
|
377
|
-
split_opts_reg opts = { StringValuePtr(criterion), NUM2DBL(impurity) };
|
342
|
+
static VALUE find_split_params_reg(VALUE self, VALUE criterion, VALUE impurity, VALUE order, VALUE features, VALUE targets) {
|
343
|
+
ndfunc_arg_in_t ain[3] = {{numo_cInt32, 1}, {numo_cDFloat, 1}, {numo_cDFloat, 2}};
|
344
|
+
size_t out_shape[1] = {4};
|
345
|
+
ndfunc_arg_out_t aout[1] = {{numo_cDFloat, 1, out_shape}};
|
346
|
+
ndfunc_t ndf = {(na_iter_func_t)iter_find_split_params_reg, NO_LOOP, 3, 1, ain, aout};
|
347
|
+
split_opts_reg opts = {StringValuePtr(criterion), NUM2DBL(impurity)};
|
378
348
|
VALUE params = na_ndloop3(&ndf, &opts, 3, order, features, targets);
|
379
349
|
VALUE results = rb_ary_new2(4);
|
380
350
|
double* params_ptr = (double*)na_get_pointer_for_read(params);
|
@@ -390,9 +360,7 @@ find_split_params_reg(VALUE self, VALUE criterion, VALUE impurity, VALUE order,
|
|
390
360
|
/**
|
391
361
|
* @!visibility private
|
392
362
|
*/
|
393
|
-
static void
|
394
|
-
iter_find_split_params_grad_reg(na_loop_t const* lp)
|
395
|
-
{
|
363
|
+
static void iter_find_split_params_grad_reg(na_loop_t const* lp) {
|
396
364
|
const int32_t* o = (int32_t*)NDL_PTR(lp, 0);
|
397
365
|
const double* f = (double*)NDL_PTR(lp, 1);
|
398
366
|
const double* g = (double*)NDL_PTR(lp, 2);
|
@@ -427,15 +395,16 @@ iter_find_split_params_grad_reg(na_loop_t const* lp)
|
|
427
395
|
/* Calculate gain of new split. */
|
428
396
|
r_grad = s_grad - l_grad;
|
429
397
|
r_hess = s_hess - l_hess;
|
430
|
-
gain = (l_grad * l_grad) / (l_hess + reg_lambda) +
|
431
|
-
(r_grad * r_grad) / (r_hess + reg_lambda) -
|
398
|
+
gain = (l_grad * l_grad) / (l_hess + reg_lambda) + (r_grad * r_grad) / (r_hess + reg_lambda) -
|
432
399
|
(s_grad * s_grad) / (s_hess + reg_lambda);
|
433
400
|
/* Update optimal parameters. */
|
434
401
|
if (gain > gain_max) {
|
435
402
|
threshold = 0.5 * (curr_el + next_el);
|
436
403
|
gain_max = gain;
|
437
404
|
}
|
438
|
-
if (next_pos == n_elements)
|
405
|
+
if (next_pos == n_elements) {
|
406
|
+
break;
|
407
|
+
}
|
439
408
|
curr_pos = next_pos;
|
440
409
|
curr_el = f[o[curr_pos]];
|
441
410
|
}
|
@@ -458,15 +427,13 @@ iter_find_split_params_grad_reg(na_loop_t const* lp)
|
|
458
427
|
* @param reg_lambda [Float] The L2 regularization term on weight.
|
459
428
|
* @return [Array<Float>] The array consists of optimal parameters including threshold and gain.
|
460
429
|
*/
|
461
|
-
static VALUE
|
462
|
-
|
463
|
-
|
464
|
-
{
|
465
|
-
|
466
|
-
|
467
|
-
|
468
|
-
ndfunc_t ndf = { (na_iter_func_t)iter_find_split_params_grad_reg, NO_LOOP, 4, 1, ain, aout };
|
469
|
-
double opts[3] = { NUM2DBL(sum_gradient), NUM2DBL(sum_hessian), NUM2DBL(reg_lambda) };
|
430
|
+
static VALUE find_split_params_grad_reg(VALUE self, VALUE order, VALUE features, VALUE gradients, VALUE hessians,
|
431
|
+
VALUE sum_gradient, VALUE sum_hessian, VALUE reg_lambda) {
|
432
|
+
ndfunc_arg_in_t ain[4] = {{numo_cInt32, 1}, {numo_cDFloat, 1}, {numo_cDFloat, 1}, {numo_cDFloat, 1}};
|
433
|
+
size_t out_shape[1] = {2};
|
434
|
+
ndfunc_arg_out_t aout[1] = {{numo_cDFloat, 1, out_shape}};
|
435
|
+
ndfunc_t ndf = {(na_iter_func_t)iter_find_split_params_grad_reg, NO_LOOP, 4, 1, ain, aout};
|
436
|
+
double opts[3] = {NUM2DBL(sum_gradient), NUM2DBL(sum_hessian), NUM2DBL(reg_lambda)};
|
470
437
|
VALUE params = na_ndloop3(&ndf, opts, 4, order, features, gradients, hessians);
|
471
438
|
VALUE results = rb_ary_new2(2);
|
472
439
|
double* params_ptr = (double*)na_get_pointer_for_read(params);
|
@@ -488,9 +455,7 @@ find_split_params_grad_reg
|
|
488
455
|
* @param n_classes_ [Integer] The number of classes.
|
489
456
|
* @return [Float] impurity
|
490
457
|
*/
|
491
|
-
static VALUE
|
492
|
-
node_impurity_cls(VALUE self, VALUE criterion, VALUE y_nary, VALUE n_elements_, VALUE n_classes_)
|
493
|
-
{
|
458
|
+
static VALUE node_impurity_cls(VALUE self, VALUE criterion, VALUE y_nary, VALUE n_elements_, VALUE n_classes_) {
|
494
459
|
long i;
|
495
460
|
const long n_classes = NUM2LONG(n_classes_);
|
496
461
|
const long n_elements = NUM2LONG(n_elements_);
|
@@ -498,7 +463,9 @@ node_impurity_cls(VALUE self, VALUE criterion, VALUE y_nary, VALUE n_elements_,
|
|
498
463
|
double* histogram = alloc_dbl_array(n_classes);
|
499
464
|
VALUE ret;
|
500
465
|
|
501
|
-
for (i = 0; i < n_elements; i++) {
|
466
|
+
for (i = 0; i < n_elements; i++) {
|
467
|
+
histogram[y[i]] += 1;
|
468
|
+
}
|
502
469
|
|
503
470
|
ret = DBL2NUM(calc_impurity_cls(StringValuePtr(criterion), histogram, n_elements, n_classes));
|
504
471
|
|
@@ -520,9 +487,7 @@ node_impurity_cls(VALUE self, VALUE criterion, VALUE y_nary, VALUE n_elements_,
|
|
520
487
|
* @param y [Array<Float>] (shape: [n_samples, n_outputs]) The taget values.
|
521
488
|
* @return [Float] impurity
|
522
489
|
*/
|
523
|
-
static VALUE
|
524
|
-
node_impurity_reg(VALUE self, VALUE criterion, VALUE y)
|
525
|
-
{
|
490
|
+
static VALUE node_impurity_reg(VALUE self, VALUE criterion, VALUE y) {
|
526
491
|
long i;
|
527
492
|
const long n_elements = RARRAY_LEN(y);
|
528
493
|
const long n_outputs = RARRAY_LEN(rb_ary_entry(y, 0));
|
@@ -546,9 +511,10 @@ node_impurity_reg(VALUE self, VALUE criterion, VALUE y)
|
|
546
511
|
return ret;
|
547
512
|
}
|
548
513
|
|
549
|
-
void
|
550
|
-
|
514
|
+
void Init_rumaleext(void) {
|
515
|
+
VALUE mRumale = rb_define_module("Rumale");
|
551
516
|
VALUE mTree = rb_define_module_under(mRumale, "Tree");
|
517
|
+
|
552
518
|
/**
|
553
519
|
* Document-module: Rumale::Tree::ExtDecisionTreeClassifier
|
554
520
|
* @!visibility private
|
@@ -1,12 +1,12 @@
|
|
1
|
-
#ifndef
|
2
|
-
#define
|
1
|
+
#ifndef RUMALEEXT_H
|
2
|
+
#define RUMALEEXT_H 1
|
3
3
|
|
4
4
|
#include <math.h>
|
5
5
|
#include <string.h>
|
6
|
+
|
6
7
|
#include <ruby.h>
|
8
|
+
|
7
9
|
#include <numo/narray.h>
|
8
10
|
#include <numo/template.h>
|
9
11
|
|
10
|
-
|
11
|
-
|
12
|
-
#endif /* RUMALE_TREE_H */
|
12
|
+
#endif /* RUMALEEXT_H */
|
@@ -1,6 +1,5 @@
|
|
1
1
|
# frozen_string_literal: true
|
2
2
|
|
3
|
-
require 'ostruct'
|
4
3
|
require 'rumale/base/base_estimator'
|
5
4
|
require 'rumale/base/cluster_analyzer'
|
6
5
|
require 'rumale/pairwise_metric'
|
@@ -108,7 +107,28 @@ module Rumale
|
|
108
107
|
end
|
109
108
|
end
|
110
109
|
|
111
|
-
|
110
|
+
# @!visibility private
|
111
|
+
class Node
|
112
|
+
# @!visibility private
|
113
|
+
attr_reader :x, :y, :weight, :n_elements
|
114
|
+
|
115
|
+
# @!visibility private
|
116
|
+
def initialize(x:, y:, weight:, n_elements: 0)
|
117
|
+
@x = x
|
118
|
+
@y = y
|
119
|
+
@weight = weight
|
120
|
+
@n_elements = n_elements
|
121
|
+
end
|
122
|
+
|
123
|
+
# @!visibility private
|
124
|
+
def ==(other)
|
125
|
+
# :nocov:
|
126
|
+
x == other.x && y == other.y && weight == other.weight && n_elements == other.n_elements
|
127
|
+
# :nocov:
|
128
|
+
end
|
129
|
+
end
|
130
|
+
|
131
|
+
private_constant :UnionFind, :Node
|
112
132
|
|
113
133
|
def partial_fit(distance_mat)
|
114
134
|
mr_distance_mat = mutual_reachability_distances(distance_mat, @params[:min_samples])
|
@@ -161,30 +181,30 @@ module Rumale
|
|
161
181
|
|
162
182
|
if n_x_elements >= min_cluster_size && n_y_elements >= min_cluster_size
|
163
183
|
relabel[edge.x] = next_label
|
164
|
-
res.push(
|
184
|
+
res.push(Node.new(x: relabel[n_id], y: relabel[edge.x], weight: density, n_elements: n_x_elements))
|
165
185
|
next_label += 1
|
166
186
|
relabel[edge.y] = next_label
|
167
|
-
res.push(
|
187
|
+
res.push(Node.new(x: relabel[n_id], y: relabel[edge.y], weight: density, n_elements: n_y_elements))
|
168
188
|
next_label += 1
|
169
189
|
elsif n_x_elements < min_cluster_size && n_y_elements < min_cluster_size
|
170
190
|
breadth_first_search_hierarchy(hierarchy, edge.x).each do |sn_id|
|
171
|
-
res.push(
|
191
|
+
res.push(Node.new(x: relabel[n_id], y: sn_id, weight: density, n_elements: 1)) if sn_id < n_points
|
172
192
|
visited[sn_id] = true
|
173
193
|
end
|
174
194
|
breadth_first_search_hierarchy(hierarchy, edge.y).each do |sn_id|
|
175
|
-
res.push(
|
195
|
+
res.push(Node.new(x: relabel[n_id], y: sn_id, weight: density, n_elements: 1)) if sn_id < n_points
|
176
196
|
visited[sn_id] = true
|
177
197
|
end
|
178
198
|
elsif n_x_elements < min_cluster_size
|
179
199
|
relabel[edge.y] = relabel[n_id]
|
180
200
|
breadth_first_search_hierarchy(hierarchy, edge.x).each do |sn_id|
|
181
|
-
res.push(
|
201
|
+
res.push(Node.new(x: relabel[n_id], y: sn_id, weight: density, n_elements: 1)) if sn_id < n_points
|
182
202
|
visited[sn_id] = true
|
183
203
|
end
|
184
204
|
elsif n_y_elements < min_cluster_size
|
185
205
|
relabel[edge.x] = relabel[n_id]
|
186
206
|
breadth_first_search_hierarchy(hierarchy, edge.y).each do |sn_id|
|
187
|
-
res.push(
|
207
|
+
res.push(Node.new(x: relabel[n_id], y: sn_id, weight: density, n_elements: 1)) if sn_id < n_points
|
188
208
|
visited[sn_id] = true
|
189
209
|
end
|
190
210
|
end
|
@@ -1,6 +1,5 @@
|
|
1
1
|
# frozen_string_literal: true
|
2
2
|
|
3
|
-
require 'ostruct'
|
4
3
|
require 'rumale/base/base_estimator'
|
5
4
|
require 'rumale/base/cluster_analyzer'
|
6
5
|
require 'rumale/pairwise_metric'
|
@@ -25,7 +24,7 @@ module Rumale
|
|
25
24
|
attr_reader :labels
|
26
25
|
|
27
26
|
# Return the hierarchical structure.
|
28
|
-
# @return [Array<
|
27
|
+
# @return [Array<SingleLinkage::Node>] (shape: [n_samples - 1])
|
29
28
|
attr_reader :hierarchy
|
30
29
|
|
31
30
|
# Create a new cluster analyzer with single linkage algorithm.
|
@@ -104,7 +103,26 @@ module Rumale
|
|
104
103
|
end
|
105
104
|
end
|
106
105
|
|
107
|
-
|
106
|
+
# @!visibility private
|
107
|
+
class Node
|
108
|
+
# @!visibility private
|
109
|
+
attr_reader :x, :y, :weight, :n_elements
|
110
|
+
|
111
|
+
# @!visibility private
|
112
|
+
def initialize(x:, y:, weight:, n_elements: 0)
|
113
|
+
@x = x
|
114
|
+
@y = y
|
115
|
+
@weight = weight
|
116
|
+
@n_elements = n_elements
|
117
|
+
end
|
118
|
+
|
119
|
+
# @!visibility private
|
120
|
+
def ==(other)
|
121
|
+
x == other.x && y == other.y && weight == other.weight && n_elements == other.n_elements
|
122
|
+
end
|
123
|
+
end
|
124
|
+
|
125
|
+
private_constant :UnionFind, :Node
|
108
126
|
|
109
127
|
def partial_fit(distance_mat)
|
110
128
|
mst = minimum_spanning_tree(distance_mat)
|
@@ -125,7 +143,7 @@ module Rumale
|
|
125
143
|
curr_weights = Numo::DFloat.minimum(curr_weights[target], complete_graph[curr_node, curr_labels])
|
126
144
|
next_node = curr_labels[curr_weights.min_index]
|
127
145
|
weight = curr_weights.min
|
128
|
-
|
146
|
+
Node.new(x: curr_node, y: next_node, weight: weight)
|
129
147
|
end
|
130
148
|
mst.sort! { |a, b| a.weight <=> b.weight }
|
131
149
|
end
|
@@ -140,7 +158,7 @@ module Rumale
|
|
140
158
|
x_root, y_root = [y_root, x_root] unless x_root < y_root
|
141
159
|
weight = mst[n].weight
|
142
160
|
n_samples = uf.union(x_root, y_root)
|
143
|
-
|
161
|
+
Node.new(x: x_root, y: y_root, weight: weight, n_elements: n_samples)
|
144
162
|
end
|
145
163
|
end
|
146
164
|
|
@@ -81,7 +81,7 @@ module Rumale
|
|
81
81
|
wx = @params[:whiten] ? (x - @mean).dot(whiten_mat.transpose) : x
|
82
82
|
unmixing, @n_iter = ica(wx, @params[:fun], @params[:max_iter], @params[:tol], @rng.dup)
|
83
83
|
@components = @params[:whiten] ? unmixing.dot(whiten_mat) : unmixing
|
84
|
-
@mixing = Numo::Linalg.pinv(@components)
|
84
|
+
@mixing = Numo::Linalg.pinv(@components).dup
|
85
85
|
if @params[:n_components] == 1
|
86
86
|
@components = @components.flatten.dup
|
87
87
|
@mixing = @mixing.flatten.dup
|
@@ -161,7 +161,7 @@ module Rumale
|
|
161
161
|
|
162
162
|
proba = 1.0 / (Numo::NMath.exp(-decision_function(x)) + 1.0)
|
163
163
|
|
164
|
-
return (proba.transpose / proba.sum(axis: 1)).transpose if @classes.size > 2
|
164
|
+
return (proba.transpose / proba.sum(axis: 1)).transpose.dup if @classes.size > 2
|
165
165
|
|
166
166
|
n_samples, = x.shape
|
167
167
|
probs = Numo::DFloat.zeros(n_samples, 2)
|
@@ -182,7 +182,7 @@ module Rumale
|
|
182
182
|
else
|
183
183
|
@estimators.map { |tree| tree.apply(x) }
|
184
184
|
end
|
185
|
-
Numo::Int32[*leaf_ids].transpose
|
185
|
+
Numo::Int32[*leaf_ids].transpose.dup
|
186
186
|
end
|
187
187
|
|
188
188
|
private
|
@@ -159,7 +159,7 @@ module Rumale
|
|
159
159
|
# @return [Numo::Int32] (shape: [n_samples, n_estimators]) Leaf index for sample.
|
160
160
|
def apply(x)
|
161
161
|
x = check_convert_sample_array(x)
|
162
|
-
Numo::Int32[*Array.new(@params[:n_estimators]) { |n| @estimators[n].apply(x) }].transpose
|
162
|
+
Numo::Int32[*Array.new(@params[:n_estimators]) { |n| @estimators[n].apply(x) }].transpose.dup
|
163
163
|
end
|
164
164
|
|
165
165
|
private
|
@@ -136,7 +136,7 @@ module Rumale
|
|
136
136
|
# @return [Numo::Int32] (shape: [n_samples, n_estimators]) Leaf index for sample.
|
137
137
|
def apply(x)
|
138
138
|
x = check_convert_sample_array(x)
|
139
|
-
Numo::Int32[*Array.new(@params[:n_estimators]) { |n| @estimators[n].apply(x) }].transpose
|
139
|
+
Numo::Int32[*Array.new(@params[:n_estimators]) { |n| @estimators[n].apply(x) }].transpose.dup
|
140
140
|
end
|
141
141
|
|
142
142
|
private
|
@@ -118,8 +118,7 @@ module Rumale
|
|
118
118
|
desc_y_true = Numo::Int32.cast(bin_y_true[desc_pred_ids])
|
119
119
|
desc_y_score = y_score[desc_pred_ids]
|
120
120
|
|
121
|
-
|
122
|
-
threshold_ids = dist_value_ids.append(desc_y_true.size - 1)
|
121
|
+
threshold_ids = Numo::Int32.cast(desc_y_score.diff.ne(0).where.to_a.append(desc_y_true.size - 1))
|
123
122
|
|
124
123
|
true_pos = desc_y_true.cumsum[threshold_ids]
|
125
124
|
false_pos = 1 + threshold_ids - true_pos
|
@@ -73,7 +73,7 @@ module Rumale
|
|
73
73
|
|
74
74
|
# random sampling.
|
75
75
|
@component_indices = Numo::Int32.cast(Array(0...n_samples).shuffle(random: sub_rng)[0...n_components])
|
76
|
-
@components = x[@component_indices, true]
|
76
|
+
@components = x[@component_indices, true].dup
|
77
77
|
|
78
78
|
# calculate normalizing factor.
|
79
79
|
kernel_mat = kernel_mat(@components)
|
@@ -152,7 +152,7 @@ module Rumale
|
|
152
152
|
|
153
153
|
if @classes.size > 2
|
154
154
|
probs = 1.0 / (Numo::NMath.exp(@prob_param[true, 0] * decision_function(x) + @prob_param[true, 1]) + 1.0)
|
155
|
-
return (probs.transpose / probs.sum(axis: 1)).transpose
|
155
|
+
return (probs.transpose / probs.sum(axis: 1)).transpose.dup
|
156
156
|
end
|
157
157
|
|
158
158
|
n_samples, = x.shape
|