rumale 0.18.1 → 0.18.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/.travis.yml +0 -1
- data/CHANGELOG.md +16 -4
- data/lib/rumale.rb +6 -1
- data/lib/rumale/clustering/dbscan.rb +0 -17
- data/lib/rumale/clustering/gaussian_mixture.rb +0 -21
- data/lib/rumale/clustering/hdbscan.rb +0 -15
- data/lib/rumale/clustering/k_means.rb +0 -17
- data/lib/rumale/clustering/k_medoids.rb +0 -19
- data/lib/rumale/clustering/power_iteration.rb +0 -19
- data/lib/rumale/clustering/single_linkage.rb +0 -17
- data/lib/rumale/clustering/spectral_clustering.rb +0 -17
- data/lib/rumale/evaluation_measure/function.rb +34 -0
- data/lib/rumale/kernel_approximation/rbf.rb +0 -19
- data/lib/rumale/kernel_machine/kernel_pca.rb +0 -21
- data/lib/rumale/kernel_machine/kernel_ridge.rb +0 -15
- data/lib/rumale/kernel_machine/kernel_svc.rb +0 -21
- data/lib/rumale/naive_bayes/base_naive_bayes.rb +47 -0
- data/lib/rumale/naive_bayes/bernoulli_nb.rb +82 -0
- data/lib/rumale/naive_bayes/complement_nb.rb +85 -0
- data/lib/rumale/naive_bayes/gaussian_nb.rb +69 -0
- data/lib/rumale/naive_bayes/multinomial_nb.rb +74 -0
- data/lib/rumale/naive_bayes/negation_nb.rb +71 -0
- data/lib/rumale/nearest_neighbors/k_neighbors_classifier.rb +0 -19
- data/lib/rumale/nearest_neighbors/k_neighbors_regressor.rb +0 -17
- data/lib/rumale/neural_network/adam.rb +0 -19
- data/lib/rumale/preprocessing/bin_discretizer.rb +0 -15
- data/lib/rumale/preprocessing/label_binarizer.rb +0 -15
- data/lib/rumale/preprocessing/label_encoder.rb +0 -15
- data/lib/rumale/preprocessing/max_abs_scaler.rb +0 -15
- data/lib/rumale/preprocessing/min_max_scaler.rb +0 -17
- data/lib/rumale/preprocessing/one_hot_encoder.rb +0 -19
- data/lib/rumale/preprocessing/ordinal_encoder.rb +0 -13
- data/lib/rumale/preprocessing/standard_scaler.rb +0 -15
- data/lib/rumale/version.rb +1 -1
- metadata +8 -3
- data/lib/rumale/naive_bayes/naive_bayes.rb +0 -250
@@ -88,21 +88,6 @@ module Rumale
|
|
88
88
|
end
|
89
89
|
transformed
|
90
90
|
end
|
91
|
-
|
92
|
-
# Dump marshal data.
|
93
|
-
# @return [Hash] The marshal data about BinDiscretizer
|
94
|
-
def marshal_dump
|
95
|
-
{ params: @params,
|
96
|
-
feature_steps: @feature_steps }
|
97
|
-
end
|
98
|
-
|
99
|
-
# Load marshal data.
|
100
|
-
# @return [nil]
|
101
|
-
def marshal_load(obj)
|
102
|
-
@params = obj[:params]
|
103
|
-
@feature_steps = obj[:feature_steps]
|
104
|
-
nil
|
105
|
-
end
|
106
91
|
end
|
107
92
|
end
|
108
93
|
end
|
@@ -84,21 +84,6 @@ module Rumale
|
|
84
84
|
n_samples = x.shape[0]
|
85
85
|
Array.new(n_samples) { |n| @classes[x[n, true].ne(@params[:neg_label]).where[0]] }
|
86
86
|
end
|
87
|
-
|
88
|
-
# Dump marshal data.
|
89
|
-
# @return [Hash] The marshal data about LabelBinarizer.
|
90
|
-
def marshal_dump
|
91
|
-
{ params: @params,
|
92
|
-
classes: @classes }
|
93
|
-
end
|
94
|
-
|
95
|
-
# Load marshal data.
|
96
|
-
# @return [nil]
|
97
|
-
def marshal_load(obj)
|
98
|
-
@params = obj[:params]
|
99
|
-
@classes = obj[:classes]
|
100
|
-
nil
|
101
|
-
end
|
102
87
|
end
|
103
88
|
end
|
104
89
|
end
|
@@ -74,21 +74,6 @@ module Rumale
|
|
74
74
|
x = check_convert_label_array(x)
|
75
75
|
x.to_a.map { |n| @classes[n] }
|
76
76
|
end
|
77
|
-
|
78
|
-
# Dump marshal data.
|
79
|
-
# @return [Hash] The marshal data about LabelEncoder
|
80
|
-
def marshal_dump
|
81
|
-
{ params: @params,
|
82
|
-
classes: @classes }
|
83
|
-
end
|
84
|
-
|
85
|
-
# Load marshal data.
|
86
|
-
# @return [nil]
|
87
|
-
def marshal_load(obj)
|
88
|
-
@params = obj[:params]
|
89
|
-
@classes = obj[:classes]
|
90
|
-
nil
|
91
|
-
end
|
92
77
|
end
|
93
78
|
end
|
94
79
|
end
|
@@ -56,21 +56,6 @@ module Rumale
|
|
56
56
|
x = check_convert_sample_array(x)
|
57
57
|
x / @max_abs_vec
|
58
58
|
end
|
59
|
-
|
60
|
-
# Dump marshal data.
|
61
|
-
# @return [Hash] The marshal data about MaxAbsScaler.
|
62
|
-
def marshal_dump
|
63
|
-
{ params: @params,
|
64
|
-
max_abs_vec: @max_abs_vec }
|
65
|
-
end
|
66
|
-
|
67
|
-
# Load marshal data.
|
68
|
-
# @return [nil]
|
69
|
-
def marshal_load(obj)
|
70
|
-
@params = obj[:params]
|
71
|
-
@max_abs_vec = obj[:max_abs_vec]
|
72
|
-
nil
|
73
|
-
end
|
74
59
|
end
|
75
60
|
end
|
76
61
|
end
|
@@ -71,23 +71,6 @@ module Rumale
|
|
71
71
|
nx = (x - @min_vec.tile(n_samples, 1)) / dif_vec.tile(n_samples, 1)
|
72
72
|
nx * (@params[:feature_range][1] - @params[:feature_range][0]) + @params[:feature_range][0]
|
73
73
|
end
|
74
|
-
|
75
|
-
# Dump marshal data.
|
76
|
-
# @return [Hash] The marshal data about MinMaxScaler.
|
77
|
-
def marshal_dump
|
78
|
-
{ params: @params,
|
79
|
-
min_vec: @min_vec,
|
80
|
-
max_vec: @max_vec }
|
81
|
-
end
|
82
|
-
|
83
|
-
# Load marshal data.
|
84
|
-
# @return [nil]
|
85
|
-
def marshal_load(obj)
|
86
|
-
@params = obj[:params]
|
87
|
-
@min_vec = obj[:min_vec]
|
88
|
-
@max_vec = obj[:max_vec]
|
89
|
-
nil
|
90
|
-
end
|
91
74
|
end
|
92
75
|
end
|
93
76
|
end
|
@@ -81,25 +81,6 @@ module Rumale
|
|
81
81
|
codes[true, @active_features].dup
|
82
82
|
end
|
83
83
|
|
84
|
-
# Dump marshal data.
|
85
|
-
# @return [Hash] The marshal data about OneHotEncoder.
|
86
|
-
def marshal_dump
|
87
|
-
{ params: @params,
|
88
|
-
n_values: @n_values,
|
89
|
-
active_features: @active_features,
|
90
|
-
feature_indices: @feature_indices }
|
91
|
-
end
|
92
|
-
|
93
|
-
# Load marshal data.
|
94
|
-
# @return [nil]
|
95
|
-
def marshal_load(obj)
|
96
|
-
@params = obj[:params]
|
97
|
-
@n_values = obj[:n_values]
|
98
|
-
@active_features = obj[:active_features]
|
99
|
-
@feature_indices = obj[:feature_indices]
|
100
|
-
nil
|
101
|
-
end
|
102
|
-
|
103
84
|
private
|
104
85
|
|
105
86
|
def encode(x, indices)
|
@@ -102,19 +102,6 @@ module Rumale
|
|
102
102
|
|
103
103
|
Numo::NArray.asarray(inv_transformed.transpose)
|
104
104
|
end
|
105
|
-
|
106
|
-
# Dump marshal data.
|
107
|
-
# @return [Hash] The marshal data about OrdinalEncoder.
|
108
|
-
def marshal_dump
|
109
|
-
{ categories: @categories }
|
110
|
-
end
|
111
|
-
|
112
|
-
# Load marshal data.
|
113
|
-
# @return [nil]
|
114
|
-
def marshal_load(obj)
|
115
|
-
@categories = obj[:categories]
|
116
|
-
nil
|
117
|
-
end
|
118
105
|
end
|
119
106
|
end
|
120
107
|
end
|
@@ -66,21 +66,6 @@ module Rumale
|
|
66
66
|
n_samples, = x.shape
|
67
67
|
(x - @mean_vec.tile(n_samples, 1)) / @std_vec.tile(n_samples, 1)
|
68
68
|
end
|
69
|
-
|
70
|
-
# Dump marshal data.
|
71
|
-
# @return [Hash] The marshal data about StandardScaler.
|
72
|
-
def marshal_dump
|
73
|
-
{ mean_vec: @mean_vec,
|
74
|
-
std_vec: @std_vec }
|
75
|
-
end
|
76
|
-
|
77
|
-
# Load marshal data.
|
78
|
-
# @return [nil]
|
79
|
-
def marshal_load(obj)
|
80
|
-
@mean_vec = obj[:mean_vec]
|
81
|
-
@std_vec = obj[:std_vec]
|
82
|
-
nil
|
83
|
-
end
|
84
69
|
end
|
85
70
|
end
|
86
71
|
end
|
data/lib/rumale/version.rb
CHANGED
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: rumale
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.18.
|
4
|
+
version: 0.18.2
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- yoshoku
|
8
8
|
autorequire:
|
9
9
|
bindir: exe
|
10
10
|
cert_chain: []
|
11
|
-
date: 2020-03-
|
11
|
+
date: 2020-03-21 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: numo-narray
|
@@ -266,7 +266,12 @@ files:
|
|
266
266
|
- lib/rumale/model_selection/stratified_k_fold.rb
|
267
267
|
- lib/rumale/model_selection/stratified_shuffle_split.rb
|
268
268
|
- lib/rumale/multiclass/one_vs_rest_classifier.rb
|
269
|
-
- lib/rumale/naive_bayes/
|
269
|
+
- lib/rumale/naive_bayes/base_naive_bayes.rb
|
270
|
+
- lib/rumale/naive_bayes/bernoulli_nb.rb
|
271
|
+
- lib/rumale/naive_bayes/complement_nb.rb
|
272
|
+
- lib/rumale/naive_bayes/gaussian_nb.rb
|
273
|
+
- lib/rumale/naive_bayes/multinomial_nb.rb
|
274
|
+
- lib/rumale/naive_bayes/negation_nb.rb
|
270
275
|
- lib/rumale/nearest_neighbors/k_neighbors_classifier.rb
|
271
276
|
- lib/rumale/nearest_neighbors/k_neighbors_regressor.rb
|
272
277
|
- lib/rumale/nearest_neighbors/vp_tree.rb
|
@@ -1,250 +0,0 @@
|
|
1
|
-
# frozen_string_literal: true
|
2
|
-
|
3
|
-
require 'rumale/base/base_estimator'
|
4
|
-
require 'rumale/base/classifier'
|
5
|
-
|
6
|
-
module Rumale
|
7
|
-
# This module consists of the classes that implement naive bayes models.
|
8
|
-
module NaiveBayes
|
9
|
-
# BaseNaiveBayes is a class that has methods for common processes of naive bayes classifier.
|
10
|
-
class BaseNaiveBayes
|
11
|
-
include Base::BaseEstimator
|
12
|
-
include Base::Classifier
|
13
|
-
|
14
|
-
# Predict class labels for samples.
|
15
|
-
#
|
16
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the labels.
|
17
|
-
# @return [Numo::Int32] (shape: [n_samples]) Predicted class label per sample.
|
18
|
-
def predict(x)
|
19
|
-
x = check_convert_sample_array(x)
|
20
|
-
n_samples = x.shape.first
|
21
|
-
decision_values = decision_function(x)
|
22
|
-
Numo::Int32.asarray(Array.new(n_samples) { |n| @classes[decision_values[n, true].max_index] })
|
23
|
-
end
|
24
|
-
|
25
|
-
# Predict log-probability for samples.
|
26
|
-
#
|
27
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the log-probailities.
|
28
|
-
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted log-probability of each class per sample.
|
29
|
-
def predict_log_proba(x)
|
30
|
-
x = check_convert_sample_array(x)
|
31
|
-
n_samples, = x.shape
|
32
|
-
log_likelihoods = decision_function(x)
|
33
|
-
log_likelihoods - Numo::NMath.log(Numo::NMath.exp(log_likelihoods).sum(1)).reshape(n_samples, 1)
|
34
|
-
end
|
35
|
-
|
36
|
-
# Predict probability for samples.
|
37
|
-
#
|
38
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the probailities.
|
39
|
-
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted probability of each class per sample.
|
40
|
-
def predict_proba(x)
|
41
|
-
x = check_convert_sample_array(x)
|
42
|
-
Numo::NMath.exp(predict_log_proba(x)).abs
|
43
|
-
end
|
44
|
-
end
|
45
|
-
|
46
|
-
# GaussianNB is a class that implements Gaussian Naive Bayes classifier.
|
47
|
-
#
|
48
|
-
# @example
|
49
|
-
# estimator = Rumale::NaiveBayes::GaussianNB.new
|
50
|
-
# estimator.fit(training_samples, training_labels)
|
51
|
-
# results = estimator.predict(testing_samples)
|
52
|
-
class GaussianNB < BaseNaiveBayes
|
53
|
-
# Return the class labels.
|
54
|
-
# @return [Numo::Int32] (size: n_classes)
|
55
|
-
attr_reader :classes
|
56
|
-
|
57
|
-
# Return the prior probabilities of the classes.
|
58
|
-
# @return [Numo::DFloat] (shape: [n_classes])
|
59
|
-
attr_reader :class_priors
|
60
|
-
|
61
|
-
# Return the mean vectors of the classes.
|
62
|
-
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
63
|
-
attr_reader :means
|
64
|
-
|
65
|
-
# Return the variance vectors of the classes.
|
66
|
-
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
67
|
-
attr_reader :variances
|
68
|
-
|
69
|
-
# Create a new classifier with Gaussian Naive Bayes.
|
70
|
-
def initialize
|
71
|
-
@params = {}
|
72
|
-
end
|
73
|
-
|
74
|
-
# Fit the model with given training data.
|
75
|
-
#
|
76
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
77
|
-
# @param y [Numo::Int32] (shape: [n_samples]) The categorical variables (e.g. labels)
|
78
|
-
# to be used for fitting the model.
|
79
|
-
# @return [GaussianNB] The learned classifier itself.
|
80
|
-
def fit(x, y)
|
81
|
-
x = check_convert_sample_array(x)
|
82
|
-
y = check_convert_label_array(y)
|
83
|
-
check_sample_label_size(x, y)
|
84
|
-
n_samples, = x.shape
|
85
|
-
@classes = Numo::Int32[*y.to_a.uniq.sort]
|
86
|
-
@class_priors = Numo::DFloat[*@classes.to_a.map { |l| y.eq(l).count / n_samples.to_f }]
|
87
|
-
@means = Numo::DFloat[*@classes.to_a.map { |l| x[y.eq(l).where, true].mean(0) }]
|
88
|
-
@variances = Numo::DFloat[*@classes.to_a.map { |l| x[y.eq(l).where, true].var(0) }]
|
89
|
-
self
|
90
|
-
end
|
91
|
-
|
92
|
-
# Calculate confidence scores for samples.
|
93
|
-
#
|
94
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
95
|
-
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Confidence scores per sample for each class.
|
96
|
-
def decision_function(x)
|
97
|
-
x = check_convert_sample_array(x)
|
98
|
-
n_classes = @classes.size
|
99
|
-
log_likelihoods = Array.new(n_classes) do |l|
|
100
|
-
Math.log(@class_priors[l]) - 0.5 * (
|
101
|
-
Numo::NMath.log(2.0 * Math::PI * @variances[l, true]) +
|
102
|
-
((x - @means[l, true])**2 / @variances[l, true])).sum(1)
|
103
|
-
end
|
104
|
-
Numo::DFloat[*log_likelihoods].transpose
|
105
|
-
end
|
106
|
-
end
|
107
|
-
|
108
|
-
# MultinomialNB is a class that implements Multinomial Naive Bayes classifier.
|
109
|
-
#
|
110
|
-
# @example
|
111
|
-
# estimator = Rumale::NaiveBayes::MultinomialNB.new(smoothing_param: 1.0)
|
112
|
-
# estimator.fit(training_samples, training_labels)
|
113
|
-
# results = estimator.predict(testing_samples)
|
114
|
-
#
|
115
|
-
# *Reference*
|
116
|
-
# - C D. Manning, P. Raghavan, and H. Schutze, "Introduction to Information Retrieval," Cambridge University Press., 2008.
|
117
|
-
class MultinomialNB < BaseNaiveBayes
|
118
|
-
# Return the class labels.
|
119
|
-
# @return [Numo::Int32] (size: n_classes)
|
120
|
-
attr_reader :classes
|
121
|
-
|
122
|
-
# Return the prior probabilities of the classes.
|
123
|
-
# @return [Numo::DFloat] (shape: [n_classes])
|
124
|
-
attr_reader :class_priors
|
125
|
-
|
126
|
-
# Return the conditional probabilities for features of each class.
|
127
|
-
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
128
|
-
attr_reader :feature_probs
|
129
|
-
|
130
|
-
# Create a new classifier with Multinomial Naive Bayes.
|
131
|
-
#
|
132
|
-
# @param smoothing_param [Float] The Laplace smoothing parameter.
|
133
|
-
def initialize(smoothing_param: 1.0)
|
134
|
-
check_params_numeric(smoothing_param: smoothing_param)
|
135
|
-
check_params_positive(smoothing_param: smoothing_param)
|
136
|
-
@params = {}
|
137
|
-
@params[:smoothing_param] = smoothing_param
|
138
|
-
end
|
139
|
-
|
140
|
-
# Fit the model with given training data.
|
141
|
-
#
|
142
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
143
|
-
# @param y [Numo::Int32] (shape: [n_samples]) The categorical variables (e.g. labels)
|
144
|
-
# to be used for fitting the model.
|
145
|
-
# @return [MultinomialNB] The learned classifier itself.
|
146
|
-
def fit(x, y)
|
147
|
-
x = check_convert_sample_array(x)
|
148
|
-
y = check_convert_label_array(y)
|
149
|
-
check_sample_label_size(x, y)
|
150
|
-
n_samples, = x.shape
|
151
|
-
@classes = Numo::Int32[*y.to_a.uniq.sort]
|
152
|
-
@class_priors = Numo::DFloat[*@classes.to_a.map { |l| y.eq(l).count / n_samples.to_f }]
|
153
|
-
count_features = Numo::DFloat[*@classes.to_a.map { |l| x[y.eq(l).where, true].sum(0) }]
|
154
|
-
count_features += @params[:smoothing_param]
|
155
|
-
n_classes = @classes.size
|
156
|
-
@feature_probs = count_features / count_features.sum(1).reshape(n_classes, 1)
|
157
|
-
self
|
158
|
-
end
|
159
|
-
|
160
|
-
# Calculate confidence scores for samples.
|
161
|
-
#
|
162
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
163
|
-
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Confidence scores per sample for each class.
|
164
|
-
def decision_function(x)
|
165
|
-
x = check_convert_sample_array(x)
|
166
|
-
n_classes = @classes.size
|
167
|
-
bin_x = x.gt(0)
|
168
|
-
log_likelihoods = Array.new(n_classes) do |l|
|
169
|
-
Math.log(@class_priors[l]) + (Numo::DFloat[*bin_x] * Numo::NMath.log(@feature_probs[l, true])).sum(1)
|
170
|
-
end
|
171
|
-
Numo::DFloat[*log_likelihoods].transpose
|
172
|
-
end
|
173
|
-
end
|
174
|
-
|
175
|
-
# BernoulliNB is a class that implements Bernoulli Naive Bayes classifier.
|
176
|
-
#
|
177
|
-
# @example
|
178
|
-
# estimator = Rumale::NaiveBayes::BernoulliNB.new(smoothing_param: 1.0, bin_threshold: 0.0)
|
179
|
-
# estimator.fit(training_samples, training_labels)
|
180
|
-
# results = estimator.predict(testing_samples)
|
181
|
-
#
|
182
|
-
# *Reference*
|
183
|
-
# - C D. Manning, P. Raghavan, and H. Schutze, "Introduction to Information Retrieval," Cambridge University Press., 2008.
|
184
|
-
class BernoulliNB < BaseNaiveBayes
|
185
|
-
# Return the class labels.
|
186
|
-
# @return [Numo::Int32] (size: n_classes)
|
187
|
-
attr_reader :classes
|
188
|
-
|
189
|
-
# Return the prior probabilities of the classes.
|
190
|
-
# @return [Numo::DFloat] (shape: [n_classes])
|
191
|
-
attr_reader :class_priors
|
192
|
-
|
193
|
-
# Return the conditional probabilities for features of each class.
|
194
|
-
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
195
|
-
attr_reader :feature_probs
|
196
|
-
|
197
|
-
# Create a new classifier with Bernoulli Naive Bayes.
|
198
|
-
#
|
199
|
-
# @param smoothing_param [Float] The Laplace smoothing parameter.
|
200
|
-
# @param bin_threshold [Float] The threshold for binarizing of features.
|
201
|
-
def initialize(smoothing_param: 1.0, bin_threshold: 0.0)
|
202
|
-
check_params_numeric(smoothing_param: smoothing_param, bin_threshold: bin_threshold)
|
203
|
-
check_params_positive(smoothing_param: smoothing_param)
|
204
|
-
@params = {}
|
205
|
-
@params[:smoothing_param] = smoothing_param
|
206
|
-
@params[:bin_threshold] = bin_threshold
|
207
|
-
end
|
208
|
-
|
209
|
-
# Fit the model with given training data.
|
210
|
-
#
|
211
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
212
|
-
# @param y [Numo::Int32] (shape: [n_samples]) The categorical variables (e.g. labels)
|
213
|
-
# to be used for fitting the model.
|
214
|
-
# @return [BernoulliNB] The learned classifier itself.
|
215
|
-
def fit(x, y)
|
216
|
-
x = check_convert_sample_array(x)
|
217
|
-
y = check_convert_label_array(y)
|
218
|
-
check_sample_label_size(x, y)
|
219
|
-
n_samples, = x.shape
|
220
|
-
bin_x = Numo::DFloat[*x.gt(@params[:bin_threshold])]
|
221
|
-
@classes = Numo::Int32[*y.to_a.uniq.sort]
|
222
|
-
n_samples_each_class = Numo::DFloat[*@classes.to_a.map { |l| y.eq(l).count.to_f }]
|
223
|
-
@class_priors = n_samples_each_class / n_samples
|
224
|
-
count_features = Numo::DFloat[*@classes.to_a.map { |l| bin_x[y.eq(l).where, true].sum(0) }]
|
225
|
-
count_features += @params[:smoothing_param]
|
226
|
-
n_samples_each_class += 2.0 * @params[:smoothing_param]
|
227
|
-
n_classes = @classes.size
|
228
|
-
@feature_probs = count_features / n_samples_each_class.reshape(n_classes, 1)
|
229
|
-
self
|
230
|
-
end
|
231
|
-
|
232
|
-
# Calculate confidence scores for samples.
|
233
|
-
#
|
234
|
-
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
235
|
-
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Confidence scores per sample for each class.
|
236
|
-
def decision_function(x)
|
237
|
-
x = check_convert_sample_array(x)
|
238
|
-
n_classes = @classes.size
|
239
|
-
bin_x = Numo::DFloat[*x.gt(@params[:bin_threshold])]
|
240
|
-
not_bin_x = Numo::DFloat[*x.le(@params[:bin_threshold])]
|
241
|
-
log_likelihoods = Array.new(n_classes) do |l|
|
242
|
-
Math.log(@class_priors[l]) + (
|
243
|
-
(Numo::DFloat[*bin_x] * Numo::NMath.log(@feature_probs[l, true])).sum(1)
|
244
|
-
(Numo::DFloat[*not_bin_x] * Numo::NMath.log(1.0 - @feature_probs[l, true])).sum(1))
|
245
|
-
end
|
246
|
-
Numo::DFloat[*log_likelihoods].transpose
|
247
|
-
end
|
248
|
-
end
|
249
|
-
end
|
250
|
-
end
|