isotree 0.1.3 → 0.1.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +5 -0
- data/README.md +6 -0
- data/ext/isotree/ext.cpp +80 -19
- data/lib/isotree/isolation_forest.rb +14 -5
- data/lib/isotree/version.rb +1 -1
- data/vendor/isotree/README.md +5 -1
- data/vendor/isotree/src/fit_model.cpp +19 -8
- metadata +2 -2
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 6121fa6837526e05ea0124e69c189dea6e0c94a303a003ed8f5db5e8469e54b3
|
4
|
+
data.tar.gz: eb4a955ffa47876af4ece7ce7006635af4b43e29475055cf0046d61cbb54d443
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 22eb873ac7cff7cd5a1e0e7c04e6c47e895c06d8e0e52cd5ee2c4ab31b652203737186fe2dfd84298ea9cdf5327bed0c4d22e67ec9fb787179eae978f922b127
|
7
|
+
data.tar.gz: bcc62a52fa71bf6e3175108a473e834b8a730ea0a94dd8adc017057a23a9bf56d451c3b955b3aafacd2cd9e6f4c237c31ce4f7a3eefd4d11aaed25183def3b92
|
data/CHANGELOG.md
CHANGED
data/README.md
CHANGED
@@ -58,12 +58,18 @@ IsoTree::IsolationForest.new(
|
|
58
58
|
prob_split_avg_gain: 0,
|
59
59
|
prob_split_pooled_gain: 0,
|
60
60
|
min_gain: 0,
|
61
|
+
missing_action: "impute",
|
62
|
+
new_categ_action: "smallest",
|
63
|
+
categ_split_type: "subset",
|
61
64
|
all_perm: false,
|
62
65
|
coef_by_prop: false,
|
63
66
|
sample_with_replacement: false,
|
64
67
|
penalize_range: true,
|
65
68
|
weigh_by_kurtosis: false,
|
69
|
+
coefs: "normal",
|
66
70
|
min_imp_obs: 3,
|
71
|
+
depth_imp: "higher",
|
72
|
+
weigh_imp_rows: "inverse",
|
67
73
|
random_seed: 1,
|
68
74
|
nthreads: -1
|
69
75
|
)
|
data/ext/isotree/ext.cpp
CHANGED
@@ -5,17 +5,77 @@
|
|
5
5
|
#include <rice/Array.hpp>
|
6
6
|
#include <rice/Hash.hpp>
|
7
7
|
#include <rice/Module.hpp>
|
8
|
+
#include <rice/Object.hpp>
|
8
9
|
#include <rice/String.hpp>
|
9
10
|
#include <rice/Symbol.hpp>
|
10
11
|
|
11
12
|
using Rice::Array;
|
12
13
|
using Rice::Hash;
|
13
14
|
using Rice::Module;
|
15
|
+
using Rice::Object;
|
14
16
|
using Rice::String;
|
15
17
|
using Rice::Symbol;
|
16
18
|
using Rice::define_class_under;
|
17
19
|
using Rice::define_module;
|
18
20
|
|
21
|
+
template<>
|
22
|
+
NewCategAction from_ruby<NewCategAction>(Object x)
|
23
|
+
{
|
24
|
+
auto value = x.to_s().str();
|
25
|
+
if (value == "weighted") return Weighted;
|
26
|
+
if (value == "smallest") return Smallest;
|
27
|
+
if (value == "random") return Random;
|
28
|
+
throw std::runtime_error("Unknown new categ action: " + value);
|
29
|
+
}
|
30
|
+
|
31
|
+
template<>
|
32
|
+
MissingAction from_ruby<MissingAction>(Object x)
|
33
|
+
{
|
34
|
+
auto value = x.to_s().str();
|
35
|
+
if (value == "divide") return Divide;
|
36
|
+
if (value == "impute") return Impute;
|
37
|
+
if (value == "fail") return Fail;
|
38
|
+
throw std::runtime_error("Unknown missing action: " + value);
|
39
|
+
}
|
40
|
+
|
41
|
+
template<>
|
42
|
+
CategSplit from_ruby<CategSplit>(Object x)
|
43
|
+
{
|
44
|
+
auto value = x.to_s().str();
|
45
|
+
if (value == "subset") return SubSet;
|
46
|
+
if (value == "single_categ") return SingleCateg;
|
47
|
+
throw std::runtime_error("Unknown categ split: " + value);
|
48
|
+
}
|
49
|
+
|
50
|
+
template<>
|
51
|
+
CoefType from_ruby<CoefType>(Object x)
|
52
|
+
{
|
53
|
+
auto value = x.to_s().str();
|
54
|
+
if (value == "uniform") return Uniform;
|
55
|
+
if (value == "normal") return Normal;
|
56
|
+
throw std::runtime_error("Unknown coef type: " + value);
|
57
|
+
}
|
58
|
+
|
59
|
+
template<>
|
60
|
+
UseDepthImp from_ruby<UseDepthImp>(Object x)
|
61
|
+
{
|
62
|
+
auto value = x.to_s().str();
|
63
|
+
if (value == "lower") return Lower;
|
64
|
+
if (value == "higher") return Higher;
|
65
|
+
if (value == "same") return Same;
|
66
|
+
throw std::runtime_error("Unknown depth imp: " + value);
|
67
|
+
}
|
68
|
+
|
69
|
+
template<>
|
70
|
+
WeighImpRows from_ruby<WeighImpRows>(Object x)
|
71
|
+
{
|
72
|
+
auto value = x.to_s().str();
|
73
|
+
if (value == "inverse") return Inverse;
|
74
|
+
if (value == "prop") return Prop;
|
75
|
+
if (value == "flat") return Flat;
|
76
|
+
throw std::runtime_error("Unknown weight imp rows: " + value);
|
77
|
+
}
|
78
|
+
|
19
79
|
extern "C"
|
20
80
|
void Init_ext()
|
21
81
|
{
|
@@ -54,25 +114,7 @@ void Init_ext()
|
|
54
114
|
sparse_ix* Xc_indptr = NULL;
|
55
115
|
|
56
116
|
// options
|
57
|
-
|
58
|
-
double* sample_weights = NULL;
|
59
|
-
bool weight_as_sample = false;
|
60
|
-
size_t max_depth = 0;
|
61
|
-
bool limit_depth = true;
|
62
|
-
bool standardize_dist = false;
|
63
|
-
double* tmat = NULL;
|
64
|
-
double* output_depths = NULL;
|
65
|
-
bool standardize_depth = false;
|
66
|
-
double* col_weights = NULL;
|
67
|
-
MissingAction missing_action = Impute;
|
68
|
-
CategSplit cat_split_type = SubSet;
|
69
|
-
NewCategAction new_cat_action = Smallest;
|
70
|
-
Imputer *imputer = NULL;
|
71
|
-
UseDepthImp depth_imp = Higher;
|
72
|
-
WeighImpRows weigh_imp_rows = Inverse;
|
73
|
-
bool impute_at_fit = false;
|
74
|
-
|
75
|
-
// Rice has limit of 14 arguments, so use hash for options
|
117
|
+
// Rice has limit of 14 arguments, so use hash
|
76
118
|
size_t sample_size = options.get<size_t, Symbol>("sample_size");
|
77
119
|
size_t ndim = options.get<size_t, Symbol>("ndim");
|
78
120
|
size_t ntrees = options.get<size_t, Symbol>("ntrees");
|
@@ -82,15 +124,34 @@ void Init_ext()
|
|
82
124
|
double prob_pick_by_gain_pl = options.get<double, Symbol>("prob_pick_pooled_gain");
|
83
125
|
double prob_split_by_gain_pl = options.get<double, Symbol>("prob_split_pooled_gain");
|
84
126
|
double min_gain = options.get<double, Symbol>("min_gain");
|
127
|
+
MissingAction missing_action = options.get<MissingAction, Symbol>("missing_action");
|
128
|
+
CategSplit cat_split_type = options.get<CategSplit, Symbol>("categ_split_type");
|
129
|
+
NewCategAction new_cat_action = options.get<NewCategAction, Symbol>("new_categ_action");
|
85
130
|
bool all_perm = options.get<bool, Symbol>("all_perm");
|
86
131
|
bool coef_by_prop = options.get<bool, Symbol>("coef_by_prop");
|
87
132
|
bool with_replacement = options.get<bool, Symbol>("sample_with_replacement");
|
88
133
|
bool penalize_range = options.get<bool, Symbol>("penalize_range");
|
89
134
|
bool weigh_by_kurt = options.get<bool, Symbol>("weigh_by_kurtosis");
|
135
|
+
CoefType coef_type = options.get<CoefType, Symbol>("coefs");
|
90
136
|
size_t min_imp_obs = options.get<size_t, Symbol>("min_imp_obs");
|
137
|
+
UseDepthImp depth_imp = options.get<UseDepthImp, Symbol>("depth_imp");
|
138
|
+
WeighImpRows weigh_imp_rows = options.get<WeighImpRows, Symbol>("weigh_imp_rows");
|
91
139
|
uint64_t random_seed = options.get<uint64_t, Symbol>("random_seed");
|
92
140
|
int nthreads = options.get<int, Symbol>("nthreads");
|
93
141
|
|
142
|
+
// TODO options
|
143
|
+
double* sample_weights = NULL;
|
144
|
+
bool weight_as_sample = false;
|
145
|
+
size_t max_depth = 0;
|
146
|
+
bool limit_depth = true;
|
147
|
+
bool standardize_dist = false;
|
148
|
+
double* tmat = NULL;
|
149
|
+
double* output_depths = NULL;
|
150
|
+
bool standardize_depth = false;
|
151
|
+
double* col_weights = NULL;
|
152
|
+
Imputer *imputer = NULL;
|
153
|
+
bool impute_at_fit = false;
|
154
|
+
|
94
155
|
fit_iforest(
|
95
156
|
NULL,
|
96
157
|
&iso,
|
@@ -4,9 +4,11 @@ module IsoTree
|
|
4
4
|
sample_size: nil, ntrees: 500, ndim: 3, ntry: 3,
|
5
5
|
prob_pick_avg_gain: 0, prob_pick_pooled_gain: 0,
|
6
6
|
prob_split_avg_gain: 0, prob_split_pooled_gain: 0,
|
7
|
-
min_gain: 0,
|
7
|
+
min_gain: 0, missing_action: "impute", new_categ_action: "smallest",
|
8
|
+
categ_split_type: "subset", all_perm: false, coef_by_prop: false,
|
8
9
|
sample_with_replacement: false, penalize_range: true,
|
9
|
-
weigh_by_kurtosis: false,
|
10
|
+
weigh_by_kurtosis: false, coefs: "normal", min_imp_obs: 3, depth_imp: "higher",
|
11
|
+
weigh_imp_rows: "inverse", random_seed: 1, nthreads: -1
|
10
12
|
)
|
11
13
|
|
12
14
|
@sample_size = sample_size
|
@@ -18,12 +20,18 @@ module IsoTree
|
|
18
20
|
@prob_split_avg_gain = prob_split_avg_gain
|
19
21
|
@prob_split_pooled_gain = prob_split_pooled_gain
|
20
22
|
@min_gain = min_gain
|
23
|
+
@missing_action = missing_action
|
24
|
+
@new_categ_action = new_categ_action
|
25
|
+
@categ_split_type = categ_split_type
|
21
26
|
@all_perm = all_perm
|
22
27
|
@coef_by_prop = coef_by_prop
|
23
28
|
@sample_with_replacement = sample_with_replacement
|
24
29
|
@penalize_range = penalize_range
|
25
30
|
@weigh_by_kurtosis = weigh_by_kurtosis
|
31
|
+
@coefs = coefs
|
26
32
|
@min_imp_obs = min_imp_obs
|
33
|
+
@depth_imp = depth_imp
|
34
|
+
@weigh_imp_rows = weigh_imp_rows
|
27
35
|
@random_seed = random_seed
|
28
36
|
|
29
37
|
# etc module returns virtual cores
|
@@ -138,10 +146,11 @@ module IsoTree
|
|
138
146
|
sample_size ntrees ndim ntry
|
139
147
|
prob_pick_avg_gain prob_pick_pooled_gain
|
140
148
|
prob_split_avg_gain prob_split_pooled_gain
|
141
|
-
min_gain
|
149
|
+
min_gain missing_action new_categ_action
|
150
|
+
categ_split_type all_perm coef_by_prop
|
142
151
|
sample_with_replacement penalize_range
|
143
|
-
weigh_by_kurtosis min_imp_obs
|
144
|
-
random_seed nthreads
|
152
|
+
weigh_by_kurtosis coefs min_imp_obs depth_imp
|
153
|
+
weigh_imp_rows random_seed nthreads
|
145
154
|
)
|
146
155
|
options = {}
|
147
156
|
keys.each do |k|
|
data/lib/isotree/version.rb
CHANGED
data/vendor/isotree/README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1
1
|
# IsoTree
|
2
2
|
|
3
|
-
Fast and multi-threaded implementation of Extended Isolation Forest, Fair-Cut Forest, SCiForest (a.k.a. Split-Criterion iForest), and regular Isolation Forest, for outlier/anomaly detection, plus additions for imputation of missing values, distance/similarity calculation between observations, and handling of categorical data. Written in C++ with interfaces for Python and R.
|
3
|
+
Fast and multi-threaded implementation of Extended Isolation Forest, Fair-Cut Forest, SCiForest (a.k.a. Split-Criterion iForest), and regular Isolation Forest, for outlier/anomaly detection, plus additions for imputation of missing values, distance/similarity calculation between observations, and handling of categorical data. Written in C++ with interfaces for Python and R. An additional wrapper for Ruby can be found [here](https://github.com/ankane/isotree).
|
4
4
|
|
5
5
|
The new concepts in this software are described in:
|
6
6
|
* [Distance approximation using Isolation Forests](https://arxiv.org/abs/1910.12362)
|
@@ -82,6 +82,10 @@ sudo ldconfig
|
|
82
82
|
|
83
83
|
(Will build as a shared object - linkage is then done with `-lisotree`)
|
84
84
|
|
85
|
+
* Ruby
|
86
|
+
|
87
|
+
See [external repository with wrapper](https://github.com/ankane/isotree).
|
88
|
+
|
85
89
|
# Sample usage
|
86
90
|
|
87
91
|
**Warning: default parameters in this implementation are very different from default parameters in others such as SciKit-Learn's, and these defaults won't scale to large datasets (see documentation for details).**
|
@@ -1,7 +1,7 @@
|
|
1
1
|
/* Isolation forests and variations thereof, with adjustments for incorporation
|
2
2
|
* of categorical variables and missing values.
|
3
3
|
* Writen for C++11 standard and aimed at being used in R and Python.
|
4
|
-
*
|
4
|
+
*
|
5
5
|
* This library is based on the following works:
|
6
6
|
* [1] Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou.
|
7
7
|
* "Isolation forest."
|
@@ -20,7 +20,7 @@
|
|
20
20
|
* [7] Quinlan, J. Ross. C4. 5: programs for machine learning. Elsevier, 2014.
|
21
21
|
* [8] Cortes, David. "Distance approximation using Isolation Forests." arXiv preprint arXiv:1910.12362 (2019).
|
22
22
|
* [9] Cortes, David. "Imputing missing values with unsupervised random trees." arXiv preprint arXiv:1911.06646 (2019).
|
23
|
-
*
|
23
|
+
*
|
24
24
|
* BSD 2-Clause License
|
25
25
|
* Copyright (c) 2019, David Cortes
|
26
26
|
* All rights reserved.
|
@@ -47,7 +47,7 @@
|
|
47
47
|
bool interrupt_switch;
|
48
48
|
|
49
49
|
/* Fit Isolation Forest model, or variant of it such as SCiForest
|
50
|
-
*
|
50
|
+
*
|
51
51
|
* Parameters:
|
52
52
|
* ===========
|
53
53
|
* - model_outputs (out)
|
@@ -291,7 +291,7 @@ bool interrupt_switch;
|
|
291
291
|
* Number of parallel threads to use. Note that, the more threads, the more memory will be
|
292
292
|
* allocated, even if the thread does not end up being used. Ignored when not building with
|
293
293
|
* OpenMP support.
|
294
|
-
*
|
294
|
+
*
|
295
295
|
* Returns
|
296
296
|
* =======
|
297
297
|
* Will return macro 'EXIT_SUCCESS' (typically =0) upon completion.
|
@@ -300,7 +300,7 @@ bool interrupt_switch;
|
|
300
300
|
* what these values correspond to, you can use the functions
|
301
301
|
* 'return_EXIT_SUCESS' and 'return_EXIT_FAILURE', which will return them
|
302
302
|
* as integers.
|
303
|
-
*
|
303
|
+
*
|
304
304
|
* References
|
305
305
|
* ==========
|
306
306
|
* [1] Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou.
|
@@ -418,6 +418,13 @@ int fit_iforest(IsoForest *model_outputs, ExtIsoForest *model_outputs_ext,
|
|
418
418
|
|
419
419
|
/* Global variable that determines if the procedure receives a stop signal */
|
420
420
|
interrupt_switch = false;
|
421
|
+
/* TODO: find a better way of handling interrupt signals when calling in Python/R.
|
422
|
+
The following will still change the behavior of interrupts when called through e.g. Flask */
|
423
|
+
#if !defined(_WIN32) && !defined(_WIN64) && !defined(_MSC_VER)
|
424
|
+
struct sigaction sig_handle;
|
425
|
+
sig_handle.sa_flags = SA_RESETHAND;
|
426
|
+
sig_handle.sa_handler = set_interrup_global_variable;
|
427
|
+
#endif
|
421
428
|
|
422
429
|
/* grow trees */
|
423
430
|
#pragma omp parallel for num_threads(nthreads) schedule(dynamic) shared(model_outputs, model_outputs_ext, worker_memory, input_data, model_params)
|
@@ -461,7 +468,11 @@ int fit_iforest(IsoForest *model_outputs, ExtIsoForest *model_outputs_ext,
|
|
461
468
|
else
|
462
469
|
model_outputs_ext->hplanes[tree].shrink_to_fit();
|
463
470
|
|
464
|
-
|
471
|
+
#if !defined(_WIN32) && !defined(_WIN64) && !defined(_MSC_VER)
|
472
|
+
// sigaction(SIGINT, &sig_handle, NULL);
|
473
|
+
#else
|
474
|
+
// signal(SIGINT, set_interrup_global_variable);
|
475
|
+
#endif
|
465
476
|
}
|
466
477
|
|
467
478
|
/* check if the procedure got interrupted */
|
@@ -545,7 +556,7 @@ int fit_iforest(IsoForest *model_outputs, ExtIsoForest *model_outputs_ext,
|
|
545
556
|
|
546
557
|
|
547
558
|
/* Add additional trees to already-fitted isolation forest model
|
548
|
-
*
|
559
|
+
*
|
549
560
|
* Parameters
|
550
561
|
* ==========
|
551
562
|
* - model_outputs
|
@@ -1001,7 +1012,7 @@ void fit_itree(std::vector<IsoTree> *tree_root,
|
|
1001
1012
|
if (
|
1002
1013
|
model_params.cat_split_type == SubSet &&
|
1003
1014
|
(
|
1004
|
-
model_params.prob_pick_by_gain_avg ||
|
1015
|
+
model_params.prob_pick_by_gain_avg ||
|
1005
1016
|
model_params.prob_pick_by_gain_pl
|
1006
1017
|
)
|
1007
1018
|
)
|
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: isotree
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.1.
|
4
|
+
version: 0.1.4
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Andrew Kane
|
8
8
|
autorequire:
|
9
9
|
bindir: bin
|
10
10
|
cert_chain: []
|
11
|
-
date: 2020-08-
|
11
|
+
date: 2020-08-22 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: rice
|