svmkit 0.2.3 → 0.2.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/.rubocop.yml +1 -0
- data/HISTORY.md +5 -0
- data/README.md +5 -3
- data/lib/svmkit/base/evaluator.rb +12 -0
- data/lib/svmkit/evaluation_measure/accuracy.rb +24 -0
- data/lib/svmkit/evaluation_measure/f_score.rb +44 -0
- data/lib/svmkit/evaluation_measure/precision.rb +44 -0
- data/lib/svmkit/evaluation_measure/precision_recall.rb +89 -0
- data/lib/svmkit/evaluation_measure/recall.rb +44 -0
- data/lib/svmkit/linear_model/logistic_regression.rb +6 -2
- data/lib/svmkit/model_selection/cross_validation.rb +15 -4
- data/lib/svmkit/polynomial_model/factorization_machine_classifier.rb +190 -0
- data/lib/svmkit/version.rb +1 -1
- data/lib/svmkit.rb +6 -0
- data/svmkit.gemspec +7 -4
- metadata +16 -6
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA1:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 22a165e3fa579d5a5d0dbd9095d208dd09d5fbd0
|
4
|
+
data.tar.gz: c3f7fa2f4fe115dc821ec5050cf7c1f0fdd75a36
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 4002a03944871f2fd9cf305fd64b02d4a1899b291167289d2f78112f977c54b33fc60041b2d2c506d755fb938e4d226002b1d6dc41ea74ad28d0e215adc0ee2b
|
7
|
+
data.tar.gz: 59dc53541aee6832ea66b826cc5f92e1329adba108aa2683ddba9105d8ac28fba7fae483ce1d2c092c60d83e9d3b3adfb574bf11401ad74f24435f237d262871
|
data/.rubocop.yml
CHANGED
data/HISTORY.md
CHANGED
@@ -1,3 +1,8 @@
|
|
1
|
+
# 0.2.4
|
2
|
+
- Added class for Factorization Machine classifier.
|
3
|
+
- Added classes for evaluation measures.
|
4
|
+
- Fixed the method for prediction of class probability in Logistic Regression.
|
5
|
+
|
1
6
|
# 0.2.3
|
2
7
|
- Added class for cross validation.
|
3
8
|
- Added specs for base modules.
|
data/README.md
CHANGED
@@ -4,9 +4,11 @@
|
|
4
4
|
[![Gem Version](https://badge.fury.io/rb/svmkit.svg)](https://badge.fury.io/rb/svmkit)
|
5
5
|
[![BSD 2-Clause License](https://img.shields.io/badge/License-BSD%202--Clause-orange.svg)](https://github.com/yoshoku/SVMKit/blob/master/LICENSE.txt)
|
6
6
|
|
7
|
-
SVMKit is a
|
8
|
-
SVMKit implements machine learning algorithms with
|
9
|
-
|
7
|
+
SVMKit is a machine learninig library in Ruby.
|
8
|
+
SVMKit implements machine learning algorithms with interfaces similar to Scikit-Learn in Python.
|
9
|
+
Since the development of SVMKit has just getting started,
|
10
|
+
the methods implemented in SVMKit are only Linear / Kernel SVC,
|
11
|
+
Logistic Regression, Factorization Machine, K-nearest neighbor classifier, and cross-validation.
|
10
12
|
|
11
13
|
## Installation
|
12
14
|
|
@@ -0,0 +1,12 @@
|
|
1
|
+
|
2
|
+
module SVMKit
|
3
|
+
module Base
|
4
|
+
# Module for all evaluation measures in SVMKit.
|
5
|
+
module Evaluator
|
6
|
+
# An abstract method for evaluation of model.
|
7
|
+
def score
|
8
|
+
raise NotImplementedError, "#{__method__} has to be implemented in #{self.class}."
|
9
|
+
end
|
10
|
+
end
|
11
|
+
end
|
12
|
+
end
|
@@ -0,0 +1,24 @@
|
|
1
|
+
require 'svmkit/base/evaluator'
|
2
|
+
|
3
|
+
module SVMKit
|
4
|
+
# This module consists of the classes for model evaluation.
|
5
|
+
module EvaluationMeasure
|
6
|
+
# Accuracy is a class that calculates the accuracy of classifier from the predicted labels.
|
7
|
+
#
|
8
|
+
# @example
|
9
|
+
# evaluator = SVMKit::EvaluationMeasure::Accuracy.new
|
10
|
+
# puts evaluator.score(ground_truth, predicted)
|
11
|
+
class Accuracy
|
12
|
+
include Base::Evaluator
|
13
|
+
|
14
|
+
# Claculate mean accuracy.
|
15
|
+
#
|
16
|
+
# @param y_true [Numo::Int32] (shape: [n_samples]) Ground truth labels.
|
17
|
+
# @param y_pred [Numo::Int32] (shape: [n_samples]) Predicted labels.
|
18
|
+
# @return [Float] Mean accuracy
|
19
|
+
def score(y_true, y_pred)
|
20
|
+
(y_true.to_a.map.with_index { |label, n| label == y_pred[n] ? 1 : 0 }).inject(:+) / y_true.size.to_f
|
21
|
+
end
|
22
|
+
end
|
23
|
+
end
|
24
|
+
end
|
@@ -0,0 +1,44 @@
|
|
1
|
+
require 'svmkit/base/evaluator'
|
2
|
+
require 'svmkit/evaluation_measure/precision_recall'
|
3
|
+
|
4
|
+
module SVMKit
|
5
|
+
# This module consists of the classes for model evaluation.
|
6
|
+
module EvaluationMeasure
|
7
|
+
# FScore is a class that calculates the F1-score of the predicted labels.
|
8
|
+
#
|
9
|
+
# @example
|
10
|
+
# evaluator = SVMKit::EvaluationMeasure::FScore.new
|
11
|
+
# puts evaluator.score(ground_truth, predicted)
|
12
|
+
class FScore
|
13
|
+
include Base::Evaluator
|
14
|
+
include EvaluationMeasure::PrecisionRecall
|
15
|
+
|
16
|
+
# Return the average type for calculation of F1-score.
|
17
|
+
# @return [String] ('binary', 'micro', 'macro')
|
18
|
+
attr_reader :average
|
19
|
+
|
20
|
+
# Create a new evaluation measure calculater for F1-score.
|
21
|
+
#
|
22
|
+
# @param average [String] The average type ('binary', 'micro', 'macro')
|
23
|
+
def initialize(average: 'binary')
|
24
|
+
@average = average
|
25
|
+
end
|
26
|
+
|
27
|
+
# Claculate average F1-score
|
28
|
+
#
|
29
|
+
# @param y_true [Numo::Int32] (shape: [n_samples]) Ground truth labels.
|
30
|
+
# @param y_pred [Numo::Int32] (shape: [n_samples]) Predicted labels.
|
31
|
+
# @return [Float] Average F1-score
|
32
|
+
def score(y_true, y_pred)
|
33
|
+
case @average
|
34
|
+
when 'binary'
|
35
|
+
f_score_each_class(y_true, y_pred).last
|
36
|
+
when 'micro'
|
37
|
+
micro_average_f_score(y_true, y_pred)
|
38
|
+
when 'macro'
|
39
|
+
macro_average_f_score(y_true, y_pred)
|
40
|
+
end
|
41
|
+
end
|
42
|
+
end
|
43
|
+
end
|
44
|
+
end
|
@@ -0,0 +1,44 @@
|
|
1
|
+
require 'svmkit/base/evaluator'
|
2
|
+
require 'svmkit/evaluation_measure/precision_recall'
|
3
|
+
|
4
|
+
module SVMKit
|
5
|
+
# This module consists of the classes for model evaluation.
|
6
|
+
module EvaluationMeasure
|
7
|
+
# Precision is a class that calculates the preicision of the predicted labels.
|
8
|
+
#
|
9
|
+
# @example
|
10
|
+
# evaluator = SVMKit::EvaluationMeasure::Precision.new
|
11
|
+
# puts evaluator.score(ground_truth, predicted)
|
12
|
+
class Precision
|
13
|
+
include Base::Evaluator
|
14
|
+
include EvaluationMeasure::PrecisionRecall
|
15
|
+
|
16
|
+
# Return the average type for calculation of precision.
|
17
|
+
# @return [String] ('binary', 'micro', 'macro')
|
18
|
+
attr_reader :average
|
19
|
+
|
20
|
+
# Create a new evaluation measure calculater for precision score.
|
21
|
+
#
|
22
|
+
# @param average [String] The average type ('binary', 'micro', 'macro')
|
23
|
+
def initialize(average: 'binary')
|
24
|
+
@average = average
|
25
|
+
end
|
26
|
+
|
27
|
+
# Claculate average precision.
|
28
|
+
#
|
29
|
+
# @param y_true [Numo::Int32] (shape: [n_samples]) Ground truth labels.
|
30
|
+
# @param y_pred [Numo::Int32] (shape: [n_samples]) Predicted labels.
|
31
|
+
# @return [Float] Average precision
|
32
|
+
def score(y_true, y_pred)
|
33
|
+
case @average
|
34
|
+
when 'binary'
|
35
|
+
precision_each_class(y_true, y_pred).last
|
36
|
+
when 'micro'
|
37
|
+
micro_average_precision(y_true, y_pred)
|
38
|
+
when 'macro'
|
39
|
+
macro_average_precision(y_true, y_pred)
|
40
|
+
end
|
41
|
+
end
|
42
|
+
end
|
43
|
+
end
|
44
|
+
end
|
@@ -0,0 +1,89 @@
|
|
1
|
+
require 'svmkit/base/evaluator'
|
2
|
+
|
3
|
+
module SVMKit
|
4
|
+
# This module consists of the classes for model evaluation.
|
5
|
+
module EvaluationMeasure
|
6
|
+
# @!visibility private
|
7
|
+
module PrecisionRecall
|
8
|
+
module_function
|
9
|
+
|
10
|
+
# @!visibility private
|
11
|
+
def precision_each_class(y_true, y_pred)
|
12
|
+
y_true.sort.to_a.uniq.map do |label|
|
13
|
+
target_positions = y_pred.eq(label)
|
14
|
+
next 0.0 if y_pred[target_positions].empty?
|
15
|
+
n_true_positives = Numo::Int32.cast(y_true[target_positions].eq(y_pred[target_positions])).sum.to_f
|
16
|
+
n_false_positives = Numo::Int32.cast(y_true[target_positions].ne(y_pred[target_positions])).sum.to_f
|
17
|
+
n_true_positives / (n_true_positives + n_false_positives)
|
18
|
+
end
|
19
|
+
end
|
20
|
+
|
21
|
+
# @!visibility private
|
22
|
+
def recall_each_class(y_true, y_pred)
|
23
|
+
y_true.sort.to_a.uniq.map do |label|
|
24
|
+
target_positions = y_true.eq(label)
|
25
|
+
next 0.0 if y_pred[target_positions].empty?
|
26
|
+
n_true_positives = Numo::Int32.cast(y_true[target_positions].eq(y_pred[target_positions])).sum.to_f
|
27
|
+
n_false_negatives = Numo::Int32.cast(y_true[target_positions].ne(y_pred[target_positions])).sum.to_f
|
28
|
+
n_true_positives / (n_true_positives + n_false_negatives)
|
29
|
+
end
|
30
|
+
end
|
31
|
+
|
32
|
+
# @!visibility private
|
33
|
+
def f_score_each_class(y_true, y_pred)
|
34
|
+
precision_each_class(y_true, y_pred).zip(recall_each_class(y_true, y_pred)).map do |p, r|
|
35
|
+
next 0.0 if p.zero? && r.zero?
|
36
|
+
(2.0 * p * r) / (p + r)
|
37
|
+
end
|
38
|
+
end
|
39
|
+
|
40
|
+
# @!visibility private
|
41
|
+
def micro_average_precision(y_true, y_pred)
|
42
|
+
evaluated_values = y_true.sort.to_a.uniq.map do |label|
|
43
|
+
target_positions = y_pred.eq(label)
|
44
|
+
next [0.0, 0.0] if y_pred[target_positions].empty?
|
45
|
+
n_true_positives = Numo::Int32.cast(y_true[target_positions].eq(y_pred[target_positions])).sum.to_f
|
46
|
+
n_false_positives = Numo::Int32.cast(y_true[target_positions].ne(y_pred[target_positions])).sum.to_f
|
47
|
+
[n_true_positives, n_true_positives + n_false_positives]
|
48
|
+
end
|
49
|
+
res = evaluated_values.transpose.map { |v| v.inject(:+) }
|
50
|
+
res.first / res.last
|
51
|
+
end
|
52
|
+
|
53
|
+
# @!visibility private
|
54
|
+
def micro_average_recall(y_true, y_pred)
|
55
|
+
evaluated_values = y_true.sort.to_a.uniq.map do |label|
|
56
|
+
target_positions = y_true.eq(label)
|
57
|
+
next 0.0 if y_pred[target_positions].empty?
|
58
|
+
n_true_positives = Numo::Int32.cast(y_true[target_positions].eq(y_pred[target_positions])).sum.to_f
|
59
|
+
n_false_negatives = Numo::Int32.cast(y_true[target_positions].ne(y_pred[target_positions])).sum.to_f
|
60
|
+
[n_true_positives, n_true_positives + n_false_negatives]
|
61
|
+
end
|
62
|
+
res = evaluated_values.transpose.map { |v| v.inject(:+) }
|
63
|
+
res.first / res.last
|
64
|
+
end
|
65
|
+
|
66
|
+
# @!visibility private
|
67
|
+
def micro_average_f_score(y_true, y_pred)
|
68
|
+
p = micro_average_precision(y_true, y_pred)
|
69
|
+
r = micro_average_recall(y_true, y_pred)
|
70
|
+
(2.0 * p * r) / (p + r)
|
71
|
+
end
|
72
|
+
|
73
|
+
# @!visibility private
|
74
|
+
def macro_average_precision(y_true, y_pred)
|
75
|
+
precision_each_class(y_true, y_pred).inject(:+) / y_true.to_a.uniq.size
|
76
|
+
end
|
77
|
+
|
78
|
+
# @!visibility private
|
79
|
+
def macro_average_recall(y_true, y_pred)
|
80
|
+
recall_each_class(y_true, y_pred).inject(:+) / y_true.to_a.uniq.size
|
81
|
+
end
|
82
|
+
|
83
|
+
# @!visibility private
|
84
|
+
def macro_average_f_score(y_true, y_pred)
|
85
|
+
f_score_each_class(y_true, y_pred).inject(:+) / y_true.to_a.uniq.size
|
86
|
+
end
|
87
|
+
end
|
88
|
+
end
|
89
|
+
end
|
@@ -0,0 +1,44 @@
|
|
1
|
+
require 'svmkit/base/evaluator'
|
2
|
+
require 'svmkit/evaluation_measure/precision_recall'
|
3
|
+
|
4
|
+
module SVMKit
|
5
|
+
# This module consists of the classes for model evaluation.
|
6
|
+
module EvaluationMeasure
|
7
|
+
# Recall is a class that calculates the recall of the predicted labels.
|
8
|
+
#
|
9
|
+
# @example
|
10
|
+
# evaluator = SVMKit::EvaluationMeasure::Recall.new
|
11
|
+
# puts evaluator.score(ground_truth, predicted)
|
12
|
+
class Recall
|
13
|
+
include Base::Evaluator
|
14
|
+
include EvaluationMeasure::PrecisionRecall
|
15
|
+
|
16
|
+
# Return the average type for calculation of recall.
|
17
|
+
# @return [String] ('binary', 'micro', 'macro')
|
18
|
+
attr_reader :average
|
19
|
+
|
20
|
+
# Create a new evaluation measure calculater for recall score.
|
21
|
+
#
|
22
|
+
# @param average [String] The average type ('binary', 'micro', 'macro')
|
23
|
+
def initialize(average: 'binary')
|
24
|
+
@average = average
|
25
|
+
end
|
26
|
+
|
27
|
+
# Claculate average recall
|
28
|
+
#
|
29
|
+
# @param y_true [Numo::Int32] (shape: [n_samples]) Ground truth labels.
|
30
|
+
# @param y_pred [Numo::Int32] (shape: [n_samples]) Predicted labels.
|
31
|
+
# @return [Float] Average recall
|
32
|
+
def score(y_true, y_pred)
|
33
|
+
case @average
|
34
|
+
when 'binary'
|
35
|
+
recall_each_class(y_true, y_pred).last
|
36
|
+
when 'micro'
|
37
|
+
micro_average_recall(y_true, y_pred)
|
38
|
+
when 'macro'
|
39
|
+
macro_average_recall(y_true, y_pred)
|
40
|
+
end
|
41
|
+
end
|
42
|
+
end
|
43
|
+
end
|
44
|
+
end
|
@@ -127,9 +127,13 @@ module SVMKit
|
|
127
127
|
# Predict probability for samples.
|
128
128
|
#
|
129
129
|
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the probailities.
|
130
|
-
# @return [Numo::DFloat] (shape: [n_samples]) Predicted probability per sample.
|
130
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted probability of each class per sample.
|
131
131
|
def predict_proba(x)
|
132
|
-
|
132
|
+
n_samples, = x.shape
|
133
|
+
proba = Numo::DFloat.zeros(n_samples, 2)
|
134
|
+
proba[true, 1] = decision_function(x)
|
135
|
+
proba[true, 0] = 1.0 - proba[true, 1]
|
136
|
+
proba
|
133
137
|
end
|
134
138
|
|
135
139
|
# Claculate the mean accuracy of the given testing data.
|
@@ -21,6 +21,10 @@ module SVMKit
|
|
21
21
|
# @return [Splitter]
|
22
22
|
attr_reader :splitter
|
23
23
|
|
24
|
+
# Return the evaluator that calculates score.
|
25
|
+
# @return [Evaluator]
|
26
|
+
attr_reader :evaluator
|
27
|
+
|
24
28
|
# Return the flag indicating whether to caculate the score of training dataset.
|
25
29
|
# @return [Boolean]
|
26
30
|
attr_reader :return_train_score
|
@@ -29,10 +33,12 @@ module SVMKit
|
|
29
33
|
#
|
30
34
|
# @param estimator [Classifier] The classifier of which performance is evaluated.
|
31
35
|
# @param splitter [Splitter] The splitter that divides dataset to training and testing dataset.
|
36
|
+
# @param evaluator [Evaluator] The evaluator that calculates score of estimator results.
|
32
37
|
# @param return_train_score [Boolean] The flag indicating whether to calculate the score of training dataset.
|
33
|
-
def initialize(estimator: nil, splitter: nil, return_train_score: false)
|
38
|
+
def initialize(estimator: nil, splitter: nil, evaluator: nil, return_train_score: false)
|
34
39
|
@estimator = estimator
|
35
40
|
@splitter = splitter
|
41
|
+
@evaluator = evaluator
|
36
42
|
@return_train_score = return_train_score
|
37
43
|
end
|
38
44
|
|
@@ -49,7 +55,7 @@ module SVMKit
|
|
49
55
|
# the return_train_score is false.
|
50
56
|
def perform(x, y)
|
51
57
|
# Initialize the report of cross validation.
|
52
|
-
report = {test_score: [], train_score: nil, fit_time: []}
|
58
|
+
report = { test_score: [], train_score: nil, fit_time: [] }
|
53
59
|
report[:train_score] = [] if @return_train_score
|
54
60
|
# Evaluate the estimator on each split.
|
55
61
|
@splitter.split(x, y).each do |train_ids, test_ids|
|
@@ -64,8 +70,13 @@ module SVMKit
|
|
64
70
|
@estimator.fit(train_x, train_y)
|
65
71
|
# Calculate scores and prepare the report.
|
66
72
|
report[:fit_time].push(Time.now.to_i - start_time)
|
67
|
-
|
68
|
-
|
73
|
+
if @evaluator.nil?
|
74
|
+
report[:test_score].push(@estimator.score(test_x, test_y))
|
75
|
+
report[:train_score].push(@estimator.score(train_x, train_y)) if @return_train_score
|
76
|
+
else
|
77
|
+
report[:test_score].push(@evaluator.score(test_y, @estimator.predict(test_x)))
|
78
|
+
report[:train_score].push(@estimator.score(train_x, @estimator.predict(train_x))) if @return_train_score
|
79
|
+
end
|
69
80
|
end
|
70
81
|
report
|
71
82
|
end
|
@@ -0,0 +1,190 @@
|
|
1
|
+
require 'svmkit/base/base_estimator'
|
2
|
+
require 'svmkit/base/classifier'
|
3
|
+
|
4
|
+
module SVMKit
|
5
|
+
# This module consists of the classes that implemnt polynomial models.
|
6
|
+
module PolynomialModel
|
7
|
+
# FactorizationMachineClassifier is a class that
|
8
|
+
# implements Fatorization Machine for binary classification
|
9
|
+
# with (mini-batch) stochastic gradient descent optimization.
|
10
|
+
# Note that this implementation uses hinge loss for the loss function.
|
11
|
+
#
|
12
|
+
# @example
|
13
|
+
# estimator =
|
14
|
+
# SVMKit::PolynomialModel::FactorizationMachineClassifier.new(
|
15
|
+
# n_factors: 10, reg_param_bias: 0.001, reg_param_weight: 0.001, reg_param_factor: 0.001,
|
16
|
+
# max_iter: 5000, batch_size: 50, random_seed: 1)
|
17
|
+
# estimator.fit(training_samples, traininig_labels)
|
18
|
+
# results = estimator.predict(testing_samples)
|
19
|
+
#
|
20
|
+
# *Reference*
|
21
|
+
# - S. Rendle, "Factorization Machines with libFM," ACM Transactions on Intelligent Systems and Technology, vol. 3 (3), pp. 57:1--57:22, 2012.
|
22
|
+
# - S. Rendle, "Factorization Machines," Proceedings of the 10th IEEE International Conference on Data Mining (ICDM'10), pp. 995--1000, 2010.
|
23
|
+
class FactorizationMachineClassifier
|
24
|
+
include Base::BaseEstimator
|
25
|
+
include Base::Classifier
|
26
|
+
|
27
|
+
# Return the factor matrix for Factorization Machine.
|
28
|
+
# @return [Numo::DFloat] (shape: [n_factors, n_features])
|
29
|
+
attr_reader :factor_mat
|
30
|
+
|
31
|
+
# Return the weight vector for Factorization Machine.
|
32
|
+
# @return [Numo::DFloat] (shape: [n_features])
|
33
|
+
attr_reader :weight_vec
|
34
|
+
|
35
|
+
# Return the bias term for Factoriazation Machine.
|
36
|
+
# @return [Float]
|
37
|
+
attr_reader :bias_term
|
38
|
+
|
39
|
+
# Return the random generator for transformation.
|
40
|
+
# @return [Random]
|
41
|
+
attr_reader :rng
|
42
|
+
|
43
|
+
# Create a new classifier with Support Vector Machine by the Pegasos algorithm.
|
44
|
+
#
|
45
|
+
# @param n_factors [Integer] The maximum number of iterations.
|
46
|
+
# @param reg_param_bias [Float] The regularization parameter for bias term.
|
47
|
+
# @param reg_param_weight [Float] The regularization parameter for weight vector.
|
48
|
+
# @param reg_param_factor [Float] The regularization parameter for factor matrix.
|
49
|
+
# @param init_std [Float] The standard deviation of normal random number for initialization of factor matrix.
|
50
|
+
# @param max_iter [Integer] The maximum number of iterations.
|
51
|
+
# @param batch_size [Integer] The size of the mini batches.
|
52
|
+
# @param random_seed [Integer] The seed value using to initialize the random generator.
|
53
|
+
def initialize(n_factors: 2, reg_param_bias: 1.0, reg_param_weight: 1.0, reg_param_factor: 1.0,
|
54
|
+
init_std: 0.1, max_iter: 1000, batch_size: 10, random_seed: nil)
|
55
|
+
@params = {}
|
56
|
+
@params[:n_factors] = n_factors
|
57
|
+
@params[:reg_param_bias] = reg_param_bias
|
58
|
+
@params[:reg_param_weight] = reg_param_weight
|
59
|
+
@params[:reg_param_factor] = reg_param_factor
|
60
|
+
@params[:init_std] = init_std
|
61
|
+
@params[:max_iter] = max_iter
|
62
|
+
@params[:batch_size] = batch_size
|
63
|
+
@params[:random_seed] = random_seed
|
64
|
+
@params[:random_seed] ||= srand
|
65
|
+
@factor_mat = nil
|
66
|
+
@weight_vec = nil
|
67
|
+
@bias_term = 0.0
|
68
|
+
@rng = Random.new(@params[:random_seed])
|
69
|
+
end
|
70
|
+
|
71
|
+
# Fit the model with given training data.
|
72
|
+
#
|
73
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
74
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used for fitting the model.
|
75
|
+
# @return [FactorizationMachineClassifier] The learned classifier itself.
|
76
|
+
def fit(x, y)
|
77
|
+
# Generate binary labels.
|
78
|
+
negative_label = y.to_a.uniq.sort.shift
|
79
|
+
bin_y = y.map { |l| l != negative_label ? 1.0 : -1.0 }
|
80
|
+
# Initialize some variables.
|
81
|
+
n_samples, n_features = x.shape
|
82
|
+
rand_ids = [*0...n_samples].shuffle(random: @rng)
|
83
|
+
@factor_mat = rand_normal([@params[:n_factors], n_features], 0, @params[:init_std])
|
84
|
+
@weight_vec = Numo::DFloat.zeros(n_features)
|
85
|
+
@bias_term = 0.0
|
86
|
+
# Start optimization.
|
87
|
+
@params[:max_iter].times do |t|
|
88
|
+
# Random sampling.
|
89
|
+
subset_ids = rand_ids.shift(@params[:batch_size])
|
90
|
+
rand_ids.concat(subset_ids)
|
91
|
+
data = x[subset_ids, true]
|
92
|
+
label = bin_y[subset_ids]
|
93
|
+
# Calculate gradients for loss function.
|
94
|
+
loss_grad = loss_gradient(data, label)
|
95
|
+
next if loss_grad.ne(0.0).count.zero?
|
96
|
+
# Update each parameter.
|
97
|
+
@bias_term -= learning_rate(@params[:reg_param_bias], t) * bias_gradient(loss_grad)
|
98
|
+
@weight_vec -= learning_rate(@params[:reg_param_weight], t) * weight_gradient(loss_grad, data)
|
99
|
+
@params[:n_factors].times do |n|
|
100
|
+
@factor_mat[n, true] -= learning_rate(@params[:reg_param_factor], t) *
|
101
|
+
factor_gradient(loss_grad, data, @factor_mat[n, true])
|
102
|
+
end
|
103
|
+
end
|
104
|
+
self
|
105
|
+
end
|
106
|
+
|
107
|
+
# Calculate confidence scores for samples.
|
108
|
+
#
|
109
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
110
|
+
# @return [Numo::DFloat] (shape: [n_samples]) Confidence score per sample.
|
111
|
+
def decision_function(x)
|
112
|
+
linear_term = @bias_term + x.dot(@weight_vec)
|
113
|
+
factor_term = 0.5 * (@factor_mat.dot(x.transpose)**2 - (@factor_mat**2).dot(x.transpose**2)).sum
|
114
|
+
linear_term + factor_term
|
115
|
+
end
|
116
|
+
|
117
|
+
# Predict class labels for samples.
|
118
|
+
#
|
119
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the labels.
|
120
|
+
# @return [Numo::Int32] (shape: [n_samples]) Predicted class label per sample.
|
121
|
+
def predict(x)
|
122
|
+
Numo::Int32.cast(decision_function(x).map { |v| v >= 0.0 ? 1 : -1 })
|
123
|
+
end
|
124
|
+
|
125
|
+
# Claculate the mean accuracy of the given testing data.
|
126
|
+
#
|
127
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) Testing data.
|
128
|
+
# @param y [Numo::Int32] (shape: [n_samples]) True labels for testing data.
|
129
|
+
# @return [Float] Mean accuracy
|
130
|
+
def score(x, y)
|
131
|
+
p = predict(x)
|
132
|
+
n_hits = (y.to_a.map.with_index { |l, n| l == p[n] ? 1 : 0 }).inject(:+)
|
133
|
+
n_hits / y.size.to_f
|
134
|
+
end
|
135
|
+
|
136
|
+
# Dump marshal data.
|
137
|
+
# @return [Hash] The marshal data about FactorizationMachineClassifier
|
138
|
+
def marshal_dump
|
139
|
+
{ params: @params, factor_mat: @factor_mat, weight_vec: @weight_vec, bias_term: @bias_term, rng: @rng }
|
140
|
+
end
|
141
|
+
|
142
|
+
# Load marshal data.
|
143
|
+
# @return [nil]
|
144
|
+
def marshal_load(obj)
|
145
|
+
@params = obj[:params]
|
146
|
+
@factor_mat = obj[:factor_mat]
|
147
|
+
@weight_vec = obj[:weight_vec]
|
148
|
+
@bias_term = obj[:bias_term]
|
149
|
+
@rng = obj[:rng]
|
150
|
+
nil
|
151
|
+
end
|
152
|
+
|
153
|
+
private
|
154
|
+
|
155
|
+
def loss_gradient(x, y)
|
156
|
+
evaluated = y * decision_function(x)
|
157
|
+
gradient = Numo::DFloat.zeros(evaluated.size)
|
158
|
+
gradient[evaluated < 1.0] = -y[evaluated < 1.0]
|
159
|
+
gradient
|
160
|
+
end
|
161
|
+
|
162
|
+
def learning_rate(reg_param, iter)
|
163
|
+
1.0 / (reg_param * (iter + 1))
|
164
|
+
end
|
165
|
+
|
166
|
+
def bias_gradient(loss_grad)
|
167
|
+
loss_grad.mean + @params[:reg_param_bias] * @bias_term
|
168
|
+
end
|
169
|
+
|
170
|
+
def weight_gradient(loss_grad, data)
|
171
|
+
(loss_grad.expand_dims(1) * data).mean(0) + @params[:reg_param_weight] * @weight_vec
|
172
|
+
end
|
173
|
+
|
174
|
+
def factor_gradient(loss_grad, data, factor)
|
175
|
+
reg_term = @params[:reg_param_factor] * factor
|
176
|
+
(loss_grad.expand_dims(1) * (data * data.dot(factor).expand_dims(1) - factor * (data**2))).mean(0) + reg_term
|
177
|
+
end
|
178
|
+
|
179
|
+
def rand_uniform(shape)
|
180
|
+
Numo::DFloat[*Array.new(shape.inject(&:*)) { @rng.rand }].reshape(*shape)
|
181
|
+
end
|
182
|
+
|
183
|
+
def rand_normal(shape, mu, sigma)
|
184
|
+
a = rand_uniform(shape)
|
185
|
+
b = rand_uniform(shape)
|
186
|
+
mu + sigma * (Numo::NMath.sqrt(-2.0 * Numo::NMath.log(a)) * Numo::NMath.sin(2.0 * Math::PI * b))
|
187
|
+
end
|
188
|
+
end
|
189
|
+
end
|
190
|
+
end
|
data/lib/svmkit/version.rb
CHANGED
data/lib/svmkit.rb
CHANGED
@@ -8,10 +8,12 @@ require 'svmkit/base/base_estimator'
|
|
8
8
|
require 'svmkit/base/classifier'
|
9
9
|
require 'svmkit/base/transformer'
|
10
10
|
require 'svmkit/base/splitter'
|
11
|
+
require 'svmkit/base/evaluator'
|
11
12
|
require 'svmkit/kernel_approximation/rbf'
|
12
13
|
require 'svmkit/linear_model/svc'
|
13
14
|
require 'svmkit/linear_model/logistic_regression'
|
14
15
|
require 'svmkit/kernel_machine/kernel_svc'
|
16
|
+
require 'svmkit/polynomial_model/factorization_machine_classifier'
|
15
17
|
require 'svmkit/multiclass/one_vs_rest_classifier'
|
16
18
|
require 'svmkit/nearest_neighbors/k_neighbors_classifier'
|
17
19
|
require 'svmkit/preprocessing/l2_normalizer'
|
@@ -20,3 +22,7 @@ require 'svmkit/preprocessing/standard_scaler'
|
|
20
22
|
require 'svmkit/model_selection/k_fold'
|
21
23
|
require 'svmkit/model_selection/stratified_k_fold'
|
22
24
|
require 'svmkit/model_selection/cross_validation'
|
25
|
+
require 'svmkit/evaluation_measure/accuracy'
|
26
|
+
require 'svmkit/evaluation_measure/precision'
|
27
|
+
require 'svmkit/evaluation_measure/recall'
|
28
|
+
require 'svmkit/evaluation_measure/f_score'
|
data/svmkit.gemspec
CHANGED
@@ -11,12 +11,15 @@ Gem::Specification.new do |spec|
|
|
11
11
|
spec.email = ['yoshoku@outlook.com']
|
12
12
|
|
13
13
|
spec.summary = <<MSG
|
14
|
-
SVMKit is
|
14
|
+
SVMKit is a machine learninig library in Ruby.
|
15
|
+
SVMKit implements machine learning algorithms with interfaces similar to Scikit-Learn in Python.
|
15
16
|
MSG
|
16
17
|
spec.description = <<MSG
|
17
|
-
SVMKit is a
|
18
|
-
SVMKit implements machine learning algorithms with
|
19
|
-
|
18
|
+
SVMKit is a machine learninig library in Ruby.
|
19
|
+
SVMKit implements machine learning algorithms with interfaces similar to Scikit-Learn in Python.
|
20
|
+
Since the development of SVMKit has just getting started,
|
21
|
+
the methods implemented in SVMKit are only Linear / Kernel SVC,
|
22
|
+
Logistic Regression, Factorization Machine, K-nearest neighbor method, and cross-validation.
|
20
23
|
MSG
|
21
24
|
spec.homepage = 'https://github.com/yoshoku/svmkit'
|
22
25
|
spec.license = 'BSD-2-Clause'
|
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: svmkit
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.2.
|
4
|
+
version: 0.2.4
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- yoshoku
|
8
8
|
autorequire:
|
9
9
|
bindir: exe
|
10
10
|
cert_chain: []
|
11
|
-
date: 2018-
|
11
|
+
date: 2018-02-04 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: numo-narray
|
@@ -81,9 +81,11 @@ dependencies:
|
|
81
81
|
- !ruby/object:Gem::Version
|
82
82
|
version: '0.15'
|
83
83
|
description: |
|
84
|
-
SVMKit is a
|
85
|
-
SVMKit implements machine learning algorithms with
|
86
|
-
|
84
|
+
SVMKit is a machine learninig library in Ruby.
|
85
|
+
SVMKit implements machine learning algorithms with interfaces similar to Scikit-Learn in Python.
|
86
|
+
Since the development of SVMKit has just getting started,
|
87
|
+
the methods implemented in SVMKit are only Linear / Kernel SVC,
|
88
|
+
Logistic Regression, Factorization Machine, K-nearest neighbor method, and cross-validation.
|
87
89
|
email:
|
88
90
|
- yoshoku@outlook.com
|
89
91
|
executables: []
|
@@ -105,9 +107,15 @@ files:
|
|
105
107
|
- lib/svmkit.rb
|
106
108
|
- lib/svmkit/base/base_estimator.rb
|
107
109
|
- lib/svmkit/base/classifier.rb
|
110
|
+
- lib/svmkit/base/evaluator.rb
|
108
111
|
- lib/svmkit/base/splitter.rb
|
109
112
|
- lib/svmkit/base/transformer.rb
|
110
113
|
- lib/svmkit/dataset.rb
|
114
|
+
- lib/svmkit/evaluation_measure/accuracy.rb
|
115
|
+
- lib/svmkit/evaluation_measure/f_score.rb
|
116
|
+
- lib/svmkit/evaluation_measure/precision.rb
|
117
|
+
- lib/svmkit/evaluation_measure/precision_recall.rb
|
118
|
+
- lib/svmkit/evaluation_measure/recall.rb
|
111
119
|
- lib/svmkit/kernel_approximation/rbf.rb
|
112
120
|
- lib/svmkit/kernel_machine/kernel_svc.rb
|
113
121
|
- lib/svmkit/linear_model/logistic_regression.rb
|
@@ -118,6 +126,7 @@ files:
|
|
118
126
|
- lib/svmkit/multiclass/one_vs_rest_classifier.rb
|
119
127
|
- lib/svmkit/nearest_neighbors/k_neighbors_classifier.rb
|
120
128
|
- lib/svmkit/pairwise_metric.rb
|
129
|
+
- lib/svmkit/polynomial_model/factorization_machine_classifier.rb
|
121
130
|
- lib/svmkit/preprocessing/l2_normalizer.rb
|
122
131
|
- lib/svmkit/preprocessing/min_max_scaler.rb
|
123
132
|
- lib/svmkit/preprocessing/standard_scaler.rb
|
@@ -152,5 +161,6 @@ rubyforge_project:
|
|
152
161
|
rubygems_version: 2.4.5.4
|
153
162
|
signing_key:
|
154
163
|
specification_version: 4
|
155
|
-
summary: SVMKit is
|
164
|
+
summary: SVMKit is a machine learninig library in Ruby. SVMKit implements machine
|
165
|
+
learning algorithms with interfaces similar to Scikit-Learn in Python.
|
156
166
|
test_files: []
|