rumale 0.17.3 → 0.18.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +5 -0
- data/README.md +3 -4
- data/lib/rumale.rb +3 -0
- data/lib/rumale/metric_learning/fisher_discriminant_analysis.rb +113 -0
- data/lib/rumale/metric_learning/neighbourhood_component_analysis.rb +183 -0
- data/lib/rumale/model_selection/function.rb +42 -0
- data/lib/rumale/pairwise_metric.rb +9 -8
- data/lib/rumale/version.rb +1 -1
- data/rumale.gemspec +5 -2
- metadata +24 -5
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 7caefe8d0715984980daede523f1944460bcff7dd782c431b7e95d851ddfbf82
|
4
|
+
data.tar.gz: 05ba1863410dcaabd31dc281c921011a9625f75980c137ab8c88f302c33042b5
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 06a95f1fda83e546954e235f06c5b487cc393a340c48b1316b9ef2d0fafbb5be4bb4de02681c6b7b013915695ecd06a04a8692c9df2ba9822e3ba4e01925c0df
|
7
|
+
data.tar.gz: 0e5e1ae19fbb420cba2914f251b8ba44de29f4eb9feed31d5bcdb0dc4003a9221053bedddc3214faf8803282965c9dbd51c1e3e81df94fe6bb4fa8c6afbfe277
|
data/CHANGELOG.md
CHANGED
@@ -1,3 +1,8 @@
|
|
1
|
+
# 0.18.0
|
2
|
+
- Add transformer class for [FisherDiscriminantAnalysis](https://yoshoku.github.io/rumale/doc/Rumale/MetricLearning/FisherDiscriminantAnalysis.html).
|
3
|
+
- Add transformer class for [NeighbourhoodComponentAnalysis](https://yoshoku.github.io/rumale/doc/Rumale/MetricLearning/NeighbourhoodComponentAnalysis.html).
|
4
|
+
- Add [module function](https://yoshoku.github.io/rumale/doc/Rumale/ModelSelection.html#train_test_split-class_method) for hold-out validation.
|
5
|
+
|
1
6
|
# 0.17.3
|
2
7
|
- Add pipeline class for [FeatureUnion](https://yoshoku.github.io/rumale/doc/Rumale/Pipeline/FeatureUnion.html).
|
3
8
|
- Fix to use mmh3 gem for generating hash value on [FeatureHasher](https://yoshoku.github.io/rumale/doc/Rumale/FeatureExtraction/FeatureHasher.html).
|
data/README.md
CHANGED
@@ -15,12 +15,11 @@ Logistic Regression, Ridge, Lasso, Factorization Machine,
|
|
15
15
|
Multi-layer Perceptron,
|
16
16
|
Naive Bayes, Decision Tree, Gradient Tree Boosting, Random Forest,
|
17
17
|
K-Means, Gaussian Mixture Model, DBSCAN, Spectral Clustering,
|
18
|
-
Mutidimensional Scaling, t-SNE,
|
18
|
+
Mutidimensional Scaling, t-SNE,
|
19
|
+
Fisher Discriminant Analysis, Neighbourhood Component Analysis,
|
20
|
+
Principal Component Analysis, Non-negative Matrix Factorization,
|
19
21
|
and many other algorithms.
|
20
22
|
|
21
|
-
This project was formerly known as "SVMKit".
|
22
|
-
If you are using SVMKit, please install Rumale and replace `SVMKit` constants with `Rumale`.
|
23
|
-
|
24
23
|
## Installation
|
25
24
|
|
26
25
|
Add this line to your application's Gemfile:
|
data/lib/rumale.rb
CHANGED
@@ -78,6 +78,8 @@ require 'rumale/decomposition/factor_analysis'
|
|
78
78
|
require 'rumale/decomposition/fast_ica'
|
79
79
|
require 'rumale/manifold/tsne'
|
80
80
|
require 'rumale/manifold/mds'
|
81
|
+
require 'rumale/metric_learning/fisher_discriminant_analysis'
|
82
|
+
require 'rumale/metric_learning/neighbourhood_component_analysis.rb'
|
81
83
|
require 'rumale/neural_network/adam'
|
82
84
|
require 'rumale/neural_network/base_mlp'
|
83
85
|
require 'rumale/neural_network/mlp_regressor'
|
@@ -100,6 +102,7 @@ require 'rumale/model_selection/shuffle_split'
|
|
100
102
|
require 'rumale/model_selection/stratified_shuffle_split'
|
101
103
|
require 'rumale/model_selection/cross_validation'
|
102
104
|
require 'rumale/model_selection/grid_search_cv'
|
105
|
+
require 'rumale/model_selection/function'
|
103
106
|
require 'rumale/evaluation_measure/accuracy'
|
104
107
|
require 'rumale/evaluation_measure/precision'
|
105
108
|
require 'rumale/evaluation_measure/recall'
|
@@ -0,0 +1,113 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/base/base_estimator'
|
4
|
+
require 'rumale/base/transformer'
|
5
|
+
|
6
|
+
module Rumale
|
7
|
+
# Module for metric learning algorithms.
|
8
|
+
module MetricLearning
|
9
|
+
# FisherDiscriminantAnalysis is a class that implements Fisher Discriminant Analysis.
|
10
|
+
#
|
11
|
+
# @example
|
12
|
+
# transformer = Rumale::MetricLearning::FisherDiscriminantAnalysis.new
|
13
|
+
# transformer.fit(training_samples, traininig_labels)
|
14
|
+
# low_samples = transformer.transform(testing_samples)
|
15
|
+
#
|
16
|
+
# *Reference*
|
17
|
+
# - Fisher, R. A., "The use of multiple measurements in taxonomic problems," Annals of Eugenics, vol. 7, pp. 179--188, 1936.
|
18
|
+
# - Sugiyama, M., "Local Fisher Discriminant Analysis for Supervised Dimensionality Reduction," Proc. ICML'06, pp. 905--912, 2006.
|
19
|
+
class FisherDiscriminantAnalysis
|
20
|
+
include Base::BaseEstimator
|
21
|
+
include Base::Transformer
|
22
|
+
|
23
|
+
# Returns the transform matrix.
|
24
|
+
# @return [Numo::DFloat] (shape: [n_components, n_features])
|
25
|
+
attr_reader :components
|
26
|
+
|
27
|
+
# Returns the mean vector.
|
28
|
+
# @return [Numo::DFloat] (shape: [n_features])
|
29
|
+
attr_reader :mean
|
30
|
+
|
31
|
+
# Returns the class mean vectors.
|
32
|
+
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
33
|
+
attr_reader :class_means
|
34
|
+
|
35
|
+
# Return the class labels.
|
36
|
+
# @return [Numo::Int32] (shape: [n_classes])
|
37
|
+
attr_reader :classes
|
38
|
+
|
39
|
+
# Create a new transformer with FisherDiscriminantAnalysis.
|
40
|
+
#
|
41
|
+
# @param n_components [Integer] The number of components.
|
42
|
+
# If nil is given, the number of components will be set to [n_features, n_classes - 1].min
|
43
|
+
def initialize(n_components: nil)
|
44
|
+
check_params_numeric_or_nil(n_components: n_components)
|
45
|
+
@params = {}
|
46
|
+
@params[:n_components] = n_components
|
47
|
+
end
|
48
|
+
|
49
|
+
# Fit the model with given training data.
|
50
|
+
#
|
51
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
52
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used for fitting the model.
|
53
|
+
# @return [FisherDiscriminantAnalysis] The learned classifier itself.
|
54
|
+
def fit(x, y)
|
55
|
+
x = check_convert_sample_array(x)
|
56
|
+
y = check_convert_label_array(y)
|
57
|
+
check_sample_label_size(x, y)
|
58
|
+
raise 'FisherDiscriminatAnalysis#fit requires Numo::Linalg but that is not loaded.' unless enable_linalg?
|
59
|
+
|
60
|
+
# initialize some variables.
|
61
|
+
n_features = x.shape[1]
|
62
|
+
@classes = Numo::Int32[*y.to_a.uniq.sort]
|
63
|
+
n_classes = @classes.size
|
64
|
+
n_components = if @params[:n_components].nil?
|
65
|
+
[n_features, n_classes - 1].min
|
66
|
+
else
|
67
|
+
[n_features, @params[:n_components]].min
|
68
|
+
end
|
69
|
+
|
70
|
+
# calculate within and between scatter matricies.
|
71
|
+
within_mat = Numo::DFloat.zeros(n_features, n_features)
|
72
|
+
between_mat = Numo::DFloat.zeros(n_features, n_features)
|
73
|
+
@class_means = Numo::DFloat.zeros(n_classes, n_features)
|
74
|
+
@mean = x.mean(0)
|
75
|
+
@classes.each_with_index do |label, i|
|
76
|
+
mask_vec = y.eq(label)
|
77
|
+
sz_class = mask_vec.count
|
78
|
+
class_samples = x[mask_vec, true]
|
79
|
+
class_mean = class_samples.mean(0)
|
80
|
+
within_mat += (class_samples - class_mean).transpose.dot(class_samples - class_mean)
|
81
|
+
between_mat += sz_class * (class_mean - @mean).expand_dims(1) * (class_mean - @mean)
|
82
|
+
@class_means[i, true] = class_mean
|
83
|
+
end
|
84
|
+
|
85
|
+
# calculate components.
|
86
|
+
_, evecs = Numo::Linalg.eigh(between_mat, within_mat, vals_range: (n_features - n_components)...n_features)
|
87
|
+
comps = evecs.reverse(1).transpose.dup
|
88
|
+
@components = n_components == 1 ? comps[0, true].dup : comps.dup
|
89
|
+
self
|
90
|
+
end
|
91
|
+
|
92
|
+
# Fit the model with training data, and then transform them with the learned model.
|
93
|
+
#
|
94
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
95
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used for fitting the model.
|
96
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data
|
97
|
+
def fit_transform(x, y)
|
98
|
+
x = check_convert_sample_array(x)
|
99
|
+
y = check_convert_label_array(y)
|
100
|
+
fit(x, y).transform(x)
|
101
|
+
end
|
102
|
+
|
103
|
+
# Transform the given data with the learned model.
|
104
|
+
#
|
105
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The data to be transformed with the learned model.
|
106
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data.
|
107
|
+
def transform(x)
|
108
|
+
x = check_convert_sample_array(x)
|
109
|
+
x.dot(@components.transpose)
|
110
|
+
end
|
111
|
+
end
|
112
|
+
end
|
113
|
+
end
|
@@ -0,0 +1,183 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/base/base_estimator'
|
4
|
+
require 'rumale/base/transformer'
|
5
|
+
require 'mopti/scaled_conjugate_gradient'
|
6
|
+
|
7
|
+
module Rumale
|
8
|
+
module MetricLearning
|
9
|
+
# NeighbourhoodComponentAnalysis is a class that implements Neighbourhood Component Analysis.
|
10
|
+
#
|
11
|
+
# @example
|
12
|
+
# transformer = Rumale::MetricLearning::NeighbourhoodComponentAnalysis.new
|
13
|
+
# transformer.fit(training_samples, traininig_labels)
|
14
|
+
# low_samples = transformer.transform(testing_samples)
|
15
|
+
#
|
16
|
+
# *Reference*
|
17
|
+
# - Goldberger, J., Roweis, S., Hinton, G., and Salakhutdinov, R., "Neighbourhood Component Analysis," Advances in NIPS'17, pp. 513--520, 2005.
|
18
|
+
class NeighbourhoodComponentAnalysis
|
19
|
+
include Base::BaseEstimator
|
20
|
+
include Base::Transformer
|
21
|
+
|
22
|
+
# Returns the neighbourhood components.
|
23
|
+
# @return [Numo::DFloat] (shape: [n_components, n_features])
|
24
|
+
attr_reader :components
|
25
|
+
|
26
|
+
# Return the number of iterations run for optimization
|
27
|
+
# @return [Integer]
|
28
|
+
attr_reader :n_iter
|
29
|
+
|
30
|
+
# Return the random generator.
|
31
|
+
# @return [Random]
|
32
|
+
attr_reader :rng
|
33
|
+
|
34
|
+
# Create a new transformer with NeighbourhoodComponentAnalysis.
|
35
|
+
#
|
36
|
+
# @param n_components [Integer] The number of components.
|
37
|
+
# @param init [String] The initialization method for components ('random' or 'pca').
|
38
|
+
# @param max_iter [Integer] The maximum number of iterations.
|
39
|
+
# @param tol [Float] The tolerance of termination criterion.
|
40
|
+
# @param verbose [Boolean] The flag indicating whether to output loss during iteration.
|
41
|
+
# @param random_seed [Integer] The seed value using to initialize the random generator.
|
42
|
+
def initialize(n_components: nil, init: 'random', max_iter: 100, tol: 1e-6, verbose: false, random_seed: nil)
|
43
|
+
check_params_numeric_or_nil(n_components: n_components, random_seed: random_seed)
|
44
|
+
check_params_numeric(max_iter: max_iter, tol: tol)
|
45
|
+
check_params_string(init: init)
|
46
|
+
check_params_boolean(verbose: verbose)
|
47
|
+
@params = {}
|
48
|
+
@params[:n_components] = n_components
|
49
|
+
@params[:init] = init
|
50
|
+
@params[:max_iter] = max_iter
|
51
|
+
@params[:tol] = tol
|
52
|
+
@params[:verbose] = verbose
|
53
|
+
@params[:random_seed] = random_seed
|
54
|
+
@params[:random_seed] ||= srand
|
55
|
+
@components = nil
|
56
|
+
@n_iter = nil
|
57
|
+
@rng = Random.new(@params[:random_seed])
|
58
|
+
end
|
59
|
+
|
60
|
+
# Fit the model with given training data.
|
61
|
+
#
|
62
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
63
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used for fitting the model.
|
64
|
+
# @return [NeighbourhoodComponentAnalysis] The learned classifier itself.
|
65
|
+
def fit(x, y)
|
66
|
+
x = check_convert_sample_array(x)
|
67
|
+
y = check_convert_label_array(y)
|
68
|
+
check_sample_label_size(x, y)
|
69
|
+
n_features = x.shape[1]
|
70
|
+
n_components = if @params[:n_components].nil?
|
71
|
+
n_features
|
72
|
+
else
|
73
|
+
[n_features, @params[:n_components]].min
|
74
|
+
end
|
75
|
+
@components, @n_iter = optimize_components(x, y, n_features, n_components)
|
76
|
+
self
|
77
|
+
end
|
78
|
+
|
79
|
+
# Fit the model with training data, and then transform them with the learned model.
|
80
|
+
#
|
81
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
|
82
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used for fitting the model.
|
83
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data
|
84
|
+
def fit_transform(x, y)
|
85
|
+
x = check_convert_sample_array(x)
|
86
|
+
y = check_convert_label_array(y)
|
87
|
+
fit(x, y).transform(x)
|
88
|
+
end
|
89
|
+
|
90
|
+
# Transform the given data with the learned model.
|
91
|
+
#
|
92
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The data to be transformed with the learned model.
|
93
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data.
|
94
|
+
def transform(x)
|
95
|
+
x = check_convert_sample_array(x)
|
96
|
+
x.dot(@components.transpose)
|
97
|
+
end
|
98
|
+
|
99
|
+
private
|
100
|
+
|
101
|
+
def init_components(x, n_features, n_components)
|
102
|
+
if @params[:init] == 'pca'
|
103
|
+
pca = Rumale::Decomposition::PCA.new(n_components: n_components, solver: 'evd')
|
104
|
+
pca.fit(x).components.flatten.dup
|
105
|
+
else
|
106
|
+
Rumale::Utils.rand_normal([n_features, n_components], @rng.dup).flatten.dup
|
107
|
+
end
|
108
|
+
end
|
109
|
+
|
110
|
+
def optimize_components(x, y, n_features, n_components)
|
111
|
+
# initialize components.
|
112
|
+
comp_init = init_components(x, n_features, n_components)
|
113
|
+
# initialize optimization results.
|
114
|
+
res = {}
|
115
|
+
res[:x] = comp_init
|
116
|
+
res[:n_iter] = 0
|
117
|
+
# perform optimization.
|
118
|
+
optimizer = Mopti::ScaledConjugateGradient.new(
|
119
|
+
fnc: method(:nca_loss), jcb: method(:nca_dloss),
|
120
|
+
x_init: comp_init, args: [x, y],
|
121
|
+
max_iter: @params[:max_iter], ftol: @params[:tol]
|
122
|
+
)
|
123
|
+
fold = 0.0
|
124
|
+
dold = 0.0
|
125
|
+
optimizer.each do |prm|
|
126
|
+
res = prm
|
127
|
+
puts "[NeighbourhoodComponentAnalysis] The value of objective function after #{res[:n_iter]} epochs: #{x.shape[0] - res[:fnc]}" if @params[:verbose]
|
128
|
+
break if (fold - res[:fnc]).abs <= @params[:tol] && (dold - res[:jcb]).abs <= @params[:tol]
|
129
|
+
fold = res[:fnc]
|
130
|
+
dold = res[:jcb]
|
131
|
+
end
|
132
|
+
# return the results.
|
133
|
+
n_iter = res[:n_iter]
|
134
|
+
comps = n_components == 1 ? res[:x].dup : res[:x].reshape(n_components, n_features)
|
135
|
+
[comps, n_iter]
|
136
|
+
end
|
137
|
+
|
138
|
+
def nca_loss(w, x, y)
|
139
|
+
# initialize some variables.
|
140
|
+
n_samples, n_features = x.shape
|
141
|
+
n_components = w.size / n_features
|
142
|
+
# projection.
|
143
|
+
w = w.reshape(n_components, n_features)
|
144
|
+
z = x.dot(w.transpose)
|
145
|
+
# calculate probability matrix.
|
146
|
+
prob_mat = probability_matrix(z)
|
147
|
+
# calculate loss.
|
148
|
+
# NOTE:
|
149
|
+
# NCA attempts to maximize its objective function.
|
150
|
+
# For the minization algorithm, the objective function value is subtracted from the maixmum value (n_samples).
|
151
|
+
mask_mat = y.expand_dims(1).eq(y)
|
152
|
+
masked_prob_mat = prob_mat * mask_mat
|
153
|
+
n_samples - masked_prob_mat.sum
|
154
|
+
end
|
155
|
+
|
156
|
+
def nca_dloss(w, x, y)
|
157
|
+
# initialize some variables.
|
158
|
+
n_features = x.shape[1]
|
159
|
+
n_components = w.size / n_features
|
160
|
+
# projection.
|
161
|
+
w = w.reshape(n_components, n_features)
|
162
|
+
z = x.dot(w.transpose)
|
163
|
+
# calculate probability matrix.
|
164
|
+
prob_mat = probability_matrix(z)
|
165
|
+
# calculate gradient.
|
166
|
+
mask_mat = y.expand_dims(1).eq(y)
|
167
|
+
masked_prob_mat = prob_mat * mask_mat
|
168
|
+
weighted_prob_mat = masked_prob_mat - prob_mat * masked_prob_mat.sum(1).expand_dims(1)
|
169
|
+
weighted_prob_mat += weighted_prob_mat.transpose
|
170
|
+
weighted_prob_mat[weighted_prob_mat.diag_indices] = -weighted_prob_mat.sum(0)
|
171
|
+
gradient = 2 * z.transpose.dot(weighted_prob_mat).dot(x)
|
172
|
+
-gradient.flatten.dup
|
173
|
+
end
|
174
|
+
|
175
|
+
def probability_matrix(z)
|
176
|
+
prob_mat = Numo::NMath.exp(-Rumale::PairwiseMetric.squared_error(z))
|
177
|
+
prob_mat[prob_mat.diag_indices] = 0.0
|
178
|
+
prob_mat /= prob_mat.sum(1).expand_dims(1)
|
179
|
+
prob_mat
|
180
|
+
end
|
181
|
+
end
|
182
|
+
end
|
183
|
+
end
|
@@ -0,0 +1,42 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/model_selection/shuffle_split'
|
4
|
+
require 'rumale/model_selection/stratified_shuffle_split'
|
5
|
+
|
6
|
+
module Rumale
|
7
|
+
module ModelSelection
|
8
|
+
module_function
|
9
|
+
|
10
|
+
# Split randomly data set into test and train data.
|
11
|
+
#
|
12
|
+
# @example
|
13
|
+
# x_train, x_test, y_train, y_test = Rumale::ModelSelection.train_test_split(x, y, test_size: 0.2, stratify: true, random_seed: 1)
|
14
|
+
#
|
15
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The dataset to be used to generate data indices.
|
16
|
+
# @param y [Numo::Int32] (shape: [n_samples]) The labels to be used to generate data indices for stratified random permutation.
|
17
|
+
# If stratify = false, this parameter is ignored.
|
18
|
+
# @param test_size [Float] The ratio of number of samples for test data.
|
19
|
+
# @param train_size [Float] The ratio of number of samples for train data.
|
20
|
+
# If nil is given, it sets to 1 - test_size.
|
21
|
+
# @param stratify [Boolean] The flag indicating whether to perform stratify split.
|
22
|
+
# @param random_seed [Integer] The seed value using to initialize the random generator.
|
23
|
+
# @return [Array<Numo::NArray>] The set of training and testing data.
|
24
|
+
def train_test_split(x, y = nil, test_size: 0.1, train_size: nil, stratify: false, random_seed: nil)
|
25
|
+
splitter = if stratify
|
26
|
+
Rumale::ModelSelection::StratifiedShuffleSplit.new(
|
27
|
+
n_splits: 1, test_size: test_size, train_size: train_size, random_seed: random_seed
|
28
|
+
)
|
29
|
+
else
|
30
|
+
Rumale::ModelSelection::ShuffleSplit.new(
|
31
|
+
n_splits: 1, test_size: test_size, train_size: train_size, random_seed: random_seed
|
32
|
+
)
|
33
|
+
end
|
34
|
+
train_ids, test_ids = splitter.split(x, y).first
|
35
|
+
x_train = x[train_ids, true].dup
|
36
|
+
y_train = y[train_ids].dup
|
37
|
+
x_test = x[test_ids, true].dup
|
38
|
+
y_test = y[test_ids].dup
|
39
|
+
[x_train, x_test, y_train, y_test]
|
40
|
+
end
|
41
|
+
end
|
42
|
+
end
|
@@ -42,15 +42,16 @@ module Rumale
|
|
42
42
|
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
43
43
|
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
44
44
|
def squared_error(x, y = nil)
|
45
|
-
|
45
|
+
y_not_given = y.nil?
|
46
|
+
y = x if y_not_given
|
46
47
|
x = Rumale::Validation.check_convert_sample_array(x)
|
47
|
-
y = Rumale::Validation.check_convert_sample_array(y)
|
48
|
-
|
49
|
-
|
50
|
-
|
51
|
-
|
52
|
-
|
53
|
-
|
48
|
+
y = Rumale::Validation.check_convert_sample_array(y) unless y_not_given
|
49
|
+
sum_x_vec = (x**2).sum(1).expand_dims(1)
|
50
|
+
sum_y_vec = y_not_given ? sum_x_vec.transpose : (y**2).sum(1).expand_dims(1).transpose
|
51
|
+
err_mat = -2 * x.dot(y.transpose)
|
52
|
+
err_mat += sum_x_vec
|
53
|
+
err_mat += sum_y_vec
|
54
|
+
err_mat.class.maximum(err_mat, 0)
|
54
55
|
end
|
55
56
|
|
56
57
|
# Calculate the rbf kernel between x and y.
|
data/lib/rumale/version.rb
CHANGED
data/rumale.gemspec
CHANGED
@@ -20,7 +20,9 @@ Gem::Specification.new do |spec|
|
|
20
20
|
Multi-layer Perceptron,
|
21
21
|
Naive Bayes, Decision Tree, Gradient Tree Boosting, Random Forest,
|
22
22
|
K-Means, Gaussian Mixture Model, DBSCAN, Spectral Clustering,
|
23
|
-
Mutidimensional Scaling, t-SNE,
|
23
|
+
Mutidimensional Scaling, t-SNE,
|
24
|
+
Fisher Discriminant Analysis, Neighbourhood Component Analysis,
|
25
|
+
Principal Component Analysis, Non-negative Matrix Factorization,
|
24
26
|
and many other algorithms.
|
25
27
|
MSG
|
26
28
|
spec.homepage = 'https://github.com/yoshoku/rumale'
|
@@ -45,13 +47,14 @@ Gem::Specification.new do |spec|
|
|
45
47
|
spec.required_ruby_version = '>= 2.3'
|
46
48
|
|
47
49
|
spec.add_runtime_dependency 'numo-narray', '>= 0.9.1'
|
50
|
+
spec.add_runtime_dependency 'mopti', '~> 0.1'
|
48
51
|
spec.add_runtime_dependency 'mmh3', '~> 0.1'
|
49
52
|
|
50
53
|
spec.add_development_dependency 'bundler', '~> 2.0'
|
51
54
|
spec.add_development_dependency 'coveralls', '~> 0.8'
|
52
55
|
spec.add_development_dependency 'numo-linalg', '>= 0.1.4'
|
53
56
|
spec.add_development_dependency 'parallel', '>= 1.17.0'
|
54
|
-
spec.add_development_dependency 'rake', '~>
|
57
|
+
spec.add_development_dependency 'rake', '~> 12.0'
|
55
58
|
spec.add_development_dependency 'rake-compiler', '~> 1.0'
|
56
59
|
spec.add_development_dependency 'rspec', '~> 3.0'
|
57
60
|
end
|
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.
|
4
|
+
version: 0.18.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- yoshoku
|
8
8
|
autorequire:
|
9
9
|
bindir: exe
|
10
10
|
cert_chain: []
|
11
|
-
date: 2020-02
|
11
|
+
date: 2020-03-02 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: numo-narray
|
@@ -24,6 +24,20 @@ dependencies:
|
|
24
24
|
- - ">="
|
25
25
|
- !ruby/object:Gem::Version
|
26
26
|
version: 0.9.1
|
27
|
+
- !ruby/object:Gem::Dependency
|
28
|
+
name: mopti
|
29
|
+
requirement: !ruby/object:Gem::Requirement
|
30
|
+
requirements:
|
31
|
+
- - "~>"
|
32
|
+
- !ruby/object:Gem::Version
|
33
|
+
version: '0.1'
|
34
|
+
type: :runtime
|
35
|
+
prerelease: false
|
36
|
+
version_requirements: !ruby/object:Gem::Requirement
|
37
|
+
requirements:
|
38
|
+
- - "~>"
|
39
|
+
- !ruby/object:Gem::Version
|
40
|
+
version: '0.1'
|
27
41
|
- !ruby/object:Gem::Dependency
|
28
42
|
name: mmh3
|
29
43
|
requirement: !ruby/object:Gem::Requirement
|
@@ -100,14 +114,14 @@ dependencies:
|
|
100
114
|
requirements:
|
101
115
|
- - "~>"
|
102
116
|
- !ruby/object:Gem::Version
|
103
|
-
version: '
|
117
|
+
version: '12.0'
|
104
118
|
type: :development
|
105
119
|
prerelease: false
|
106
120
|
version_requirements: !ruby/object:Gem::Requirement
|
107
121
|
requirements:
|
108
122
|
- - "~>"
|
109
123
|
- !ruby/object:Gem::Version
|
110
|
-
version: '
|
124
|
+
version: '12.0'
|
111
125
|
- !ruby/object:Gem::Dependency
|
112
126
|
name: rake-compiler
|
113
127
|
requirement: !ruby/object:Gem::Requirement
|
@@ -144,7 +158,9 @@ description: |
|
|
144
158
|
Multi-layer Perceptron,
|
145
159
|
Naive Bayes, Decision Tree, Gradient Tree Boosting, Random Forest,
|
146
160
|
K-Means, Gaussian Mixture Model, DBSCAN, Spectral Clustering,
|
147
|
-
Mutidimensional Scaling, t-SNE,
|
161
|
+
Mutidimensional Scaling, t-SNE,
|
162
|
+
Fisher Discriminant Analysis, Neighbourhood Component Analysis,
|
163
|
+
Principal Component Analysis, Non-negative Matrix Factorization,
|
148
164
|
and many other algorithms.
|
149
165
|
email:
|
150
166
|
- yoshoku@outlook.com
|
@@ -239,7 +255,10 @@ files:
|
|
239
255
|
- lib/rumale/linear_model/svr.rb
|
240
256
|
- lib/rumale/manifold/mds.rb
|
241
257
|
- lib/rumale/manifold/tsne.rb
|
258
|
+
- lib/rumale/metric_learning/fisher_discriminant_analysis.rb
|
259
|
+
- lib/rumale/metric_learning/neighbourhood_component_analysis.rb
|
242
260
|
- lib/rumale/model_selection/cross_validation.rb
|
261
|
+
- lib/rumale/model_selection/function.rb
|
243
262
|
- lib/rumale/model_selection/grid_search_cv.rb
|
244
263
|
- lib/rumale/model_selection/k_fold.rb
|
245
264
|
- lib/rumale/model_selection/shuffle_split.rb
|