rumale 0.8.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.coveralls.yml +1 -0
- data/.gitignore +20 -0
- data/.rspec +3 -0
- data/.rubocop.yml +47 -0
- data/.rubocop_todo.yml +58 -0
- data/.travis.yml +13 -0
- data/CHANGELOG.md +2 -0
- data/CODE_OF_CONDUCT.md +74 -0
- data/Gemfile +4 -0
- data/LICENSE.txt +23 -0
- data/README.md +175 -0
- data/Rakefile +6 -0
- data/bin/console +14 -0
- data/bin/setup +8 -0
- data/lib/rumale.rb +70 -0
- data/lib/rumale/base/base_estimator.rb +13 -0
- data/lib/rumale/base/classifier.rb +36 -0
- data/lib/rumale/base/cluster_analyzer.rb +31 -0
- data/lib/rumale/base/evaluator.rb +17 -0
- data/lib/rumale/base/regressor.rb +36 -0
- data/lib/rumale/base/splitter.rb +21 -0
- data/lib/rumale/base/transformer.rb +22 -0
- data/lib/rumale/clustering/dbscan.rb +125 -0
- data/lib/rumale/clustering/k_means.rb +138 -0
- data/lib/rumale/dataset.rb +110 -0
- data/lib/rumale/decomposition/nmf.rb +141 -0
- data/lib/rumale/decomposition/pca.rb +148 -0
- data/lib/rumale/ensemble/ada_boost_classifier.rb +196 -0
- data/lib/rumale/ensemble/ada_boost_regressor.rb +178 -0
- data/lib/rumale/ensemble/random_forest_classifier.rb +180 -0
- data/lib/rumale/ensemble/random_forest_regressor.rb +141 -0
- data/lib/rumale/evaluation_measure/accuracy.rb +29 -0
- data/lib/rumale/evaluation_measure/f_score.rb +50 -0
- data/lib/rumale/evaluation_measure/log_loss.rb +45 -0
- data/lib/rumale/evaluation_measure/mean_absolute_error.rb +29 -0
- data/lib/rumale/evaluation_measure/mean_squared_error.rb +29 -0
- data/lib/rumale/evaluation_measure/normalized_mutual_information.rb +62 -0
- data/lib/rumale/evaluation_measure/precision.rb +50 -0
- data/lib/rumale/evaluation_measure/precision_recall.rb +91 -0
- data/lib/rumale/evaluation_measure/purity.rb +40 -0
- data/lib/rumale/evaluation_measure/r2_score.rb +43 -0
- data/lib/rumale/evaluation_measure/recall.rb +50 -0
- data/lib/rumale/kernel_approximation/rbf.rb +121 -0
- data/lib/rumale/kernel_machine/kernel_svc.rb +193 -0
- data/lib/rumale/linear_model/base_linear_model.rb +89 -0
- data/lib/rumale/linear_model/lasso.rb +136 -0
- data/lib/rumale/linear_model/linear_regression.rb +110 -0
- data/lib/rumale/linear_model/logistic_regression.rb +159 -0
- data/lib/rumale/linear_model/ridge.rb +110 -0
- data/lib/rumale/linear_model/svc.rb +183 -0
- data/lib/rumale/linear_model/svr.rb +122 -0
- data/lib/rumale/model_selection/cross_validation.rb +123 -0
- data/lib/rumale/model_selection/grid_search_cv.rb +247 -0
- data/lib/rumale/model_selection/k_fold.rb +76 -0
- data/lib/rumale/model_selection/stratified_k_fold.rb +94 -0
- data/lib/rumale/multiclass/one_vs_rest_classifier.rb +100 -0
- data/lib/rumale/naive_bayes/naive_bayes.rb +315 -0
- data/lib/rumale/nearest_neighbors/k_neighbors_classifier.rb +111 -0
- data/lib/rumale/nearest_neighbors/k_neighbors_regressor.rb +93 -0
- data/lib/rumale/optimizer/nadam.rb +90 -0
- data/lib/rumale/optimizer/rmsprop.rb +69 -0
- data/lib/rumale/optimizer/sgd.rb +65 -0
- data/lib/rumale/optimizer/yellow_fin.rb +144 -0
- data/lib/rumale/pairwise_metric.rb +91 -0
- data/lib/rumale/pipeline/pipeline.rb +197 -0
- data/lib/rumale/polynomial_model/base_factorization_machine.rb +99 -0
- data/lib/rumale/polynomial_model/factorization_machine_classifier.rb +197 -0
- data/lib/rumale/polynomial_model/factorization_machine_regressor.rb +131 -0
- data/lib/rumale/preprocessing/l2_normalizer.rb +62 -0
- data/lib/rumale/preprocessing/label_encoder.rb +94 -0
- data/lib/rumale/preprocessing/min_max_scaler.rb +92 -0
- data/lib/rumale/preprocessing/one_hot_encoder.rb +98 -0
- data/lib/rumale/preprocessing/standard_scaler.rb +86 -0
- data/lib/rumale/probabilistic_output.rb +112 -0
- data/lib/rumale/tree/base_decision_tree.rb +153 -0
- data/lib/rumale/tree/decision_tree_classifier.rb +163 -0
- data/lib/rumale/tree/decision_tree_regressor.rb +135 -0
- data/lib/rumale/tree/node.rb +70 -0
- data/lib/rumale/utils.rb +37 -0
- data/lib/rumale/validation.rb +79 -0
- data/lib/rumale/values.rb +13 -0
- data/lib/rumale/version.rb +6 -0
- data/rumale.gemspec +41 -0
- metadata +204 -0
@@ -0,0 +1,91 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/validation'
|
4
|
+
|
5
|
+
module Rumale
|
6
|
+
# Module for calculating pairwise distances, similarities, and kernels.
|
7
|
+
module PairwiseMetric
|
8
|
+
class << self
|
9
|
+
# Calculate the pairwise euclidean distances between x and y.
|
10
|
+
#
|
11
|
+
# @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
|
12
|
+
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
13
|
+
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
14
|
+
def euclidean_distance(x, y = nil)
|
15
|
+
y = x if y.nil?
|
16
|
+
Rumale::Validation.check_sample_array(x)
|
17
|
+
Rumale::Validation.check_sample_array(y)
|
18
|
+
sum_x_vec = (x**2).sum(1)
|
19
|
+
sum_y_vec = (y**2).sum(1)
|
20
|
+
dot_xy_mat = x.dot(y.transpose)
|
21
|
+
distance_matrix = dot_xy_mat * -2.0 +
|
22
|
+
sum_x_vec.tile(y.shape[0], 1).transpose +
|
23
|
+
sum_y_vec.tile(x.shape[0], 1)
|
24
|
+
Numo::NMath.sqrt(distance_matrix.abs)
|
25
|
+
end
|
26
|
+
|
27
|
+
# Calculate the rbf kernel between x and y.
|
28
|
+
#
|
29
|
+
# @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
|
30
|
+
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
31
|
+
# @param gamma [Float] The parameter of rbf kernel, if nil it is 1 / n_features.
|
32
|
+
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
33
|
+
def rbf_kernel(x, y = nil, gamma = nil)
|
34
|
+
y = x if y.nil?
|
35
|
+
gamma ||= 1.0 / x.shape[1]
|
36
|
+
Rumale::Validation.check_sample_array(x)
|
37
|
+
Rumale::Validation.check_sample_array(y)
|
38
|
+
Rumale::Validation.check_params_float(gamma: gamma)
|
39
|
+
distance_matrix = euclidean_distance(x, y)
|
40
|
+
Numo::NMath.exp((distance_matrix**2) * -gamma)
|
41
|
+
end
|
42
|
+
|
43
|
+
# Calculate the linear kernel between x and y.
|
44
|
+
#
|
45
|
+
# @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
|
46
|
+
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
47
|
+
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
48
|
+
def linear_kernel(x, y = nil)
|
49
|
+
y = x if y.nil?
|
50
|
+
Rumale::Validation.check_sample_array(x)
|
51
|
+
Rumale::Validation.check_sample_array(y)
|
52
|
+
x.dot(y.transpose)
|
53
|
+
end
|
54
|
+
|
55
|
+
# Calculate the polynomial kernel between x and y.
|
56
|
+
#
|
57
|
+
# @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
|
58
|
+
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
59
|
+
# @param degree [Integer] The parameter of polynomial kernel.
|
60
|
+
# @param gamma [Float] The parameter of polynomial kernel, if nil it is 1 / n_features.
|
61
|
+
# @param coef [Integer] The parameter of polynomial kernel.
|
62
|
+
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
63
|
+
def polynomial_kernel(x, y = nil, degree = 3, gamma = nil, coef = 1)
|
64
|
+
y = x if y.nil?
|
65
|
+
gamma ||= 1.0 / x.shape[1]
|
66
|
+
Rumale::Validation.check_sample_array(x)
|
67
|
+
Rumale::Validation.check_sample_array(y)
|
68
|
+
Rumale::Validation.check_params_float(gamma: gamma)
|
69
|
+
Rumale::Validation.check_params_integer(degree: degree, coef: coef)
|
70
|
+
(x.dot(y.transpose) * gamma + coef)**degree
|
71
|
+
end
|
72
|
+
|
73
|
+
# Calculate the sigmoid kernel between x and y.
|
74
|
+
#
|
75
|
+
# @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
|
76
|
+
# @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
|
77
|
+
# @param gamma [Float] The parameter of polynomial kernel, if nil it is 1 / n_features.
|
78
|
+
# @param coef [Integer] The parameter of polynomial kernel.
|
79
|
+
# @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
|
80
|
+
def sigmoid_kernel(x, y = nil, gamma = nil, coef = 1)
|
81
|
+
y = x if y.nil?
|
82
|
+
gamma ||= 1.0 / x.shape[1]
|
83
|
+
Rumale::Validation.check_sample_array(x)
|
84
|
+
Rumale::Validation.check_sample_array(y)
|
85
|
+
Rumale::Validation.check_params_float(gamma: gamma)
|
86
|
+
Rumale::Validation.check_params_integer(coef: coef)
|
87
|
+
Numo::NMath.tanh(x.dot(y.transpose) * gamma + coef)
|
88
|
+
end
|
89
|
+
end
|
90
|
+
end
|
91
|
+
end
|
@@ -0,0 +1,197 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/validation'
|
4
|
+
require 'rumale/base/base_estimator'
|
5
|
+
|
6
|
+
module Rumale
|
7
|
+
# Module implements utilities of pipeline that cosists of a chain of transfomers and estimators.
|
8
|
+
module Pipeline
|
9
|
+
# Pipeline is a class that implements the function to perform the transformers and estimators sequencially.
|
10
|
+
#
|
11
|
+
# @example
|
12
|
+
# rbf = Rumale::KernelApproximation::RBF.new(gamma: 1.0, n_coponents: 128, random_seed: 1)
|
13
|
+
# svc = Rumale::LinearModel::SVC.new(reg_param: 1.0, fit_bias: true, max_iter: 5000, random_seed: 1)
|
14
|
+
# pipeline = Rumale::Pipeline::Pipeline.new(steps: { trs: rbf, est: svc })
|
15
|
+
# pipeline.fit(training_samples, traininig_labels)
|
16
|
+
# results = pipeline.predict(testing_samples)
|
17
|
+
#
|
18
|
+
class Pipeline
|
19
|
+
include Base::BaseEstimator
|
20
|
+
include Validation
|
21
|
+
|
22
|
+
# Return the steps.
|
23
|
+
# @return [Hash]
|
24
|
+
attr_reader :steps
|
25
|
+
|
26
|
+
# Create a new pipeline.
|
27
|
+
#
|
28
|
+
# @param steps [Hash] List of transformers and estimators. The order of transforms follows the insertion order of hash keys.
|
29
|
+
# The last entry is considered an estimator.
|
30
|
+
def initialize(steps:)
|
31
|
+
check_params_type(Hash, steps: steps)
|
32
|
+
validate_steps(steps)
|
33
|
+
@params = {}
|
34
|
+
@steps = steps
|
35
|
+
end
|
36
|
+
|
37
|
+
# Fit the model with given training data.
|
38
|
+
#
|
39
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be transformed and used for fitting the model.
|
40
|
+
# @param y [Numo::NArray] (shape: [n_samples, n_outputs]) The target values or labels to be used for fitting the model.
|
41
|
+
# @return [Pipeline] The learned pipeline itself.
|
42
|
+
def fit(x, y)
|
43
|
+
check_sample_array(x)
|
44
|
+
trans_x = apply_transforms(x, y, fit: true)
|
45
|
+
last_estimator&.fit(trans_x, y)
|
46
|
+
self
|
47
|
+
end
|
48
|
+
|
49
|
+
# Call the fit_predict method of last estimator after applying all transforms.
|
50
|
+
#
|
51
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be transformed and used for fitting the model.
|
52
|
+
# @param y [Numo::NArray] (shape: [n_samples, n_outputs], default: nil) The target values or labels to be used for fitting the model.
|
53
|
+
# @return [Numo::NArray] The predicted results by last estimator.
|
54
|
+
def fit_predict(x, y = nil)
|
55
|
+
check_sample_array(x)
|
56
|
+
trans_x = apply_transforms(x, y, fit: true)
|
57
|
+
last_estimator.fit_predict(trans_x)
|
58
|
+
end
|
59
|
+
|
60
|
+
# Call the fit_transform method of last estimator after applying all transforms.
|
61
|
+
#
|
62
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be transformed and used for fitting the model.
|
63
|
+
# @param y [Numo::NArray] (shape: [n_samples, n_outputs], default: nil) The target values or labels to be used for fitting the model.
|
64
|
+
# @return [Numo::NArray] The predicted results by last estimator.
|
65
|
+
def fit_transform(x, y = nil)
|
66
|
+
check_sample_array(x)
|
67
|
+
trans_x = apply_transforms(x, y, fit: true)
|
68
|
+
last_estimator.fit_transform(trans_x, y)
|
69
|
+
end
|
70
|
+
|
71
|
+
# Call the decision_function method of last estimator after applying all transforms.
|
72
|
+
#
|
73
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
74
|
+
# @return [Numo::DFloat] (shape: [n_samples]) Confidence score per sample.
|
75
|
+
def decision_function(x)
|
76
|
+
check_sample_array(x)
|
77
|
+
trans_x = apply_transforms(x)
|
78
|
+
last_estimator.decision_function(trans_x)
|
79
|
+
end
|
80
|
+
|
81
|
+
# Call the predict method of last estimator after applying all transforms.
|
82
|
+
#
|
83
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to obtain prediction result.
|
84
|
+
# @return [Numo::NArray] The predicted results by last estimator.
|
85
|
+
def predict(x)
|
86
|
+
check_sample_array(x)
|
87
|
+
trans_x = apply_transforms(x)
|
88
|
+
last_estimator.predict(trans_x)
|
89
|
+
end
|
90
|
+
|
91
|
+
# Call the predict_log_proba method of last estimator after applying all transforms.
|
92
|
+
#
|
93
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the log-probailities.
|
94
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted log-probability of each class per sample.
|
95
|
+
def predict_log_proba(x)
|
96
|
+
check_sample_array(x)
|
97
|
+
trans_x = apply_transforms(x)
|
98
|
+
last_estimator.predict_log_proba(trans_x)
|
99
|
+
end
|
100
|
+
|
101
|
+
# Call the predict_proba method of last estimator after applying all transforms.
|
102
|
+
#
|
103
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the probailities.
|
104
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted probability of each class per sample.
|
105
|
+
def predict_proba(x)
|
106
|
+
check_sample_array(x)
|
107
|
+
trans_x = apply_transforms(x)
|
108
|
+
last_estimator.predict_proba(trans_x)
|
109
|
+
end
|
110
|
+
|
111
|
+
# Call the transform method of last estimator after applying all transforms.
|
112
|
+
#
|
113
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to be transformed.
|
114
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed samples.
|
115
|
+
def transform(x)
|
116
|
+
check_sample_array(x)
|
117
|
+
trans_x = apply_transforms(x)
|
118
|
+
last_estimator.nil? ? trans_x : last_estimator.transform(trans_x)
|
119
|
+
end
|
120
|
+
|
121
|
+
# Call the inverse_transform method in reverse order.
|
122
|
+
#
|
123
|
+
# @param z [Numo::DFloat] (shape: [n_samples, n_components]) The transformed samples to be restored into original space.
|
124
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_featuress]) The restored samples.
|
125
|
+
def inverse_transform(z)
|
126
|
+
check_sample_array(z)
|
127
|
+
itrans_z = z
|
128
|
+
@steps.keys.reverse_each do |name|
|
129
|
+
transformer = @steps[name]
|
130
|
+
next if transformer.nil?
|
131
|
+
itrans_z = transformer.inverse_transform(itrans_z)
|
132
|
+
end
|
133
|
+
itrans_z
|
134
|
+
end
|
135
|
+
|
136
|
+
# Call the score method of last estimator after applying all transforms.
|
137
|
+
#
|
138
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) Testing data.
|
139
|
+
# @param y [Numo::NArray] (shape: [n_samples, n_outputs]) True target values or labels for testing data.
|
140
|
+
# @return [Float] The score of last estimator
|
141
|
+
def score(x, y)
|
142
|
+
check_sample_array(x)
|
143
|
+
trans_x = apply_transforms(x)
|
144
|
+
last_estimator.score(trans_x, y)
|
145
|
+
end
|
146
|
+
|
147
|
+
# Dump marshal data.
|
148
|
+
# @return [Hash] The marshal data about Pipeline.
|
149
|
+
def marshal_dump
|
150
|
+
{ params: @params,
|
151
|
+
steps: @steps }
|
152
|
+
end
|
153
|
+
|
154
|
+
# Load marshal data.
|
155
|
+
# @return [nil]
|
156
|
+
def marshal_load(obj)
|
157
|
+
@params = obj[:params]
|
158
|
+
@steps = obj[:steps]
|
159
|
+
nil
|
160
|
+
end
|
161
|
+
|
162
|
+
private
|
163
|
+
|
164
|
+
def validate_steps(steps)
|
165
|
+
steps.keys[0...-1].each do |name|
|
166
|
+
transformer = steps[name]
|
167
|
+
next if transformer.nil? || %i[fit transform].all? { |m| transformer.class.method_defined?(m) }
|
168
|
+
raise TypeError,
|
169
|
+
'Class of intermediate step in pipeline should be implemented fit and transform methods: ' \
|
170
|
+
"#{name} => #{transformer.class}"
|
171
|
+
end
|
172
|
+
|
173
|
+
estimator = steps[steps.keys.last]
|
174
|
+
unless estimator.nil? || estimator.class.method_defined?(:fit) # rubocop:disable Style/GuardClause
|
175
|
+
raise TypeError,
|
176
|
+
'Class of last step in pipeline should be implemented fit method: ' \
|
177
|
+
"#{steps.keys.last} => #{estimator.class}"
|
178
|
+
end
|
179
|
+
end
|
180
|
+
|
181
|
+
def apply_transforms(x, y = nil, fit: false)
|
182
|
+
trans_x = x
|
183
|
+
@steps.keys[0...-1].each do |name|
|
184
|
+
transformer = @steps[name]
|
185
|
+
next if transformer.nil?
|
186
|
+
transformer.fit(trans_x, y) if fit
|
187
|
+
trans_x = transformer.transform(trans_x)
|
188
|
+
end
|
189
|
+
trans_x
|
190
|
+
end
|
191
|
+
|
192
|
+
def last_estimator
|
193
|
+
@steps[@steps.keys.last]
|
194
|
+
end
|
195
|
+
end
|
196
|
+
end
|
197
|
+
end
|
@@ -0,0 +1,99 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/base/base_estimator'
|
4
|
+
require 'rumale/optimizer/nadam'
|
5
|
+
|
6
|
+
module Rumale
|
7
|
+
# This module consists of the classes that implement polynomial models.
|
8
|
+
module PolynomialModel
|
9
|
+
# BaseFactorizationMachine is an abstract class for implementation of Factorization Machine-based estimators.
|
10
|
+
# This class is used internally.
|
11
|
+
class BaseFactorizationMachine
|
12
|
+
include Base::BaseEstimator
|
13
|
+
|
14
|
+
# Initialize a Factorization Machine-based estimator.
|
15
|
+
#
|
16
|
+
# @param n_factors [Integer] The maximum number of iterations.
|
17
|
+
# @param loss [String] The loss function ('hinge' or 'logistic' or nil).
|
18
|
+
# @param reg_param_linear [Float] The regularization parameter for linear model.
|
19
|
+
# @param reg_param_factor [Float] The regularization parameter for factor matrix.
|
20
|
+
# @param max_iter [Integer] The maximum number of iterations.
|
21
|
+
# @param batch_size [Integer] The size of the mini batches.
|
22
|
+
# @param optimizer [Optimizer] The optimizer to calculate adaptive learning rate.
|
23
|
+
# If nil is given, Nadam is used.
|
24
|
+
# @param random_seed [Integer] The seed value using to initialize the random generator.
|
25
|
+
def initialize(n_factors: 2, loss: nil, reg_param_linear: 1.0, reg_param_factor: 1.0,
|
26
|
+
max_iter: 1000, batch_size: 10, optimizer: nil, random_seed: nil)
|
27
|
+
@params = {}
|
28
|
+
@params[:n_factors] = n_factors
|
29
|
+
@params[:loss] = loss unless loss.nil?
|
30
|
+
@params[:reg_param_linear] = reg_param_linear
|
31
|
+
@params[:reg_param_factor] = reg_param_factor
|
32
|
+
@params[:max_iter] = max_iter
|
33
|
+
@params[:batch_size] = batch_size
|
34
|
+
@params[:optimizer] = optimizer
|
35
|
+
@params[:optimizer] ||= Optimizer::Nadam.new
|
36
|
+
@params[:random_seed] = random_seed
|
37
|
+
@params[:random_seed] ||= srand
|
38
|
+
@factor_mat = nil
|
39
|
+
@weight_vec = nil
|
40
|
+
@bias_term = nil
|
41
|
+
@rng = Random.new(@params[:random_seed])
|
42
|
+
end
|
43
|
+
|
44
|
+
private
|
45
|
+
|
46
|
+
def partial_fit(x, y)
|
47
|
+
# Initialize some variables.
|
48
|
+
n_samples, n_features = x.shape
|
49
|
+
rand_ids = [*0...n_samples].shuffle(random: @rng)
|
50
|
+
weight_vec = Numo::DFloat.zeros(n_features + 1)
|
51
|
+
factor_mat = Numo::DFloat.zeros(@params[:n_factors], n_features)
|
52
|
+
weight_optimizer = @params[:optimizer].dup
|
53
|
+
factor_optimizers = Array.new(@params[:n_factors]) { @params[:optimizer].dup }
|
54
|
+
# Start optimization.
|
55
|
+
@params[:max_iter].times do |_t|
|
56
|
+
# Random sampling.
|
57
|
+
subset_ids = rand_ids.shift(@params[:batch_size])
|
58
|
+
rand_ids.concat(subset_ids)
|
59
|
+
data = x[subset_ids, true]
|
60
|
+
ex_data = expand_feature(data)
|
61
|
+
targets = y[subset_ids]
|
62
|
+
# Calculate gradients for loss function.
|
63
|
+
loss_grad = loss_gradient(data, ex_data, targets, factor_mat, weight_vec)
|
64
|
+
next if loss_grad.ne(0.0).count.zero?
|
65
|
+
# Update each parameter.
|
66
|
+
weight_vec = weight_optimizer.call(weight_vec, weight_gradient(loss_grad, ex_data, weight_vec))
|
67
|
+
@params[:n_factors].times do |n|
|
68
|
+
factor_mat[n, true] = factor_optimizers[n].call(factor_mat[n, true],
|
69
|
+
factor_gradient(loss_grad, data, factor_mat[n, true]))
|
70
|
+
end
|
71
|
+
end
|
72
|
+
[factor_mat, *split_weight_vec_bias(weight_vec)]
|
73
|
+
end
|
74
|
+
|
75
|
+
def loss_gradient(_x, _expanded_x, _y, _factor, _weight)
|
76
|
+
raise NotImplementedError, "#{__method__} has to be implemented in #{self.class}."
|
77
|
+
end
|
78
|
+
|
79
|
+
def weight_gradient(loss_grad, data, weight)
|
80
|
+
(loss_grad.expand_dims(1) * data).mean(0) + @params[:reg_param_linear] * weight
|
81
|
+
end
|
82
|
+
|
83
|
+
def factor_gradient(loss_grad, data, factor)
|
84
|
+
(loss_grad.expand_dims(1) * (data * data.dot(factor).expand_dims(1) - factor * (data**2))).mean(0) +
|
85
|
+
@params[:reg_param_factor] * factor
|
86
|
+
end
|
87
|
+
|
88
|
+
def expand_feature(x)
|
89
|
+
Numo::NArray.hstack([x, Numo::DFloat.ones([x.shape[0], 1])])
|
90
|
+
end
|
91
|
+
|
92
|
+
def split_weight_vec_bias(weight_vec)
|
93
|
+
weights = weight_vec[0...-1].dup
|
94
|
+
bias = weight_vec[-1]
|
95
|
+
[weights, bias]
|
96
|
+
end
|
97
|
+
end
|
98
|
+
end
|
99
|
+
end
|
@@ -0,0 +1,197 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'rumale/base/classifier'
|
4
|
+
require 'rumale/polynomial_model/base_factorization_machine'
|
5
|
+
|
6
|
+
module Rumale
|
7
|
+
# This module consists of the classes that implement polynomial models.
|
8
|
+
module PolynomialModel
|
9
|
+
# FactorizationMachineClassifier is a class that implements Factorization Machine
|
10
|
+
# with stochastic gradient descent (SGD) optimization.
|
11
|
+
# For multiclass classification problem, it uses one-vs-the-rest strategy.
|
12
|
+
#
|
13
|
+
# @example
|
14
|
+
# estimator =
|
15
|
+
# Rumale::PolynomialModel::FactorizationMachineClassifier.new(
|
16
|
+
# n_factors: 10, loss: 'hinge', reg_param_linear: 0.001, reg_param_factor: 0.001,
|
17
|
+
# max_iter: 5000, batch_size: 50, random_seed: 1)
|
18
|
+
# estimator.fit(training_samples, traininig_labels)
|
19
|
+
# results = estimator.predict(testing_samples)
|
20
|
+
#
|
21
|
+
# *Reference*
|
22
|
+
# - S. Rendle, "Factorization Machines with libFM," ACM TIST, vol. 3 (3), pp. 57:1--57:22, 2012.
|
23
|
+
# - S. Rendle, "Factorization Machines," Proc. ICDM'10, pp. 995--1000, 2010.
|
24
|
+
class FactorizationMachineClassifier < BaseFactorizationMachine
|
25
|
+
include Base::Classifier
|
26
|
+
|
27
|
+
# Return the factor matrix for Factorization Machine.
|
28
|
+
# @return [Numo::DFloat] (shape: [n_classes, n_factors, n_features])
|
29
|
+
attr_reader :factor_mat
|
30
|
+
|
31
|
+
# Return the weight vector for Factorization Machine.
|
32
|
+
# @return [Numo::DFloat] (shape: [n_classes, n_features])
|
33
|
+
attr_reader :weight_vec
|
34
|
+
|
35
|
+
# Return the bias term for Factoriazation Machine.
|
36
|
+
# @return [Numo::DFloat] (shape: [n_classes])
|
37
|
+
attr_reader :bias_term
|
38
|
+
|
39
|
+
# Return the class labels.
|
40
|
+
# @return [Numo::Int32] (shape: [n_classes])
|
41
|
+
attr_reader :classes
|
42
|
+
|
43
|
+
# Return the random generator for random sampling.
|
44
|
+
# @return [Random]
|
45
|
+
attr_reader :rng
|
46
|
+
|
47
|
+
# Create a new classifier with Factorization Machine.
|
48
|
+
#
|
49
|
+
# @param n_factors [Integer] The maximum number of iterations.
|
50
|
+
# @param loss [String] The loss function ('hinge' or 'logistic').
|
51
|
+
# @param reg_param_linear [Float] The regularization parameter for linear model.
|
52
|
+
# @param reg_param_factor [Float] The regularization parameter for factor matrix.
|
53
|
+
# @param max_iter [Integer] The maximum number of iterations.
|
54
|
+
# @param batch_size [Integer] The size of the mini batches.
|
55
|
+
# @param optimizer [Optimizer] The optimizer to calculate adaptive learning rate.
|
56
|
+
# If nil is given, Nadam is used.
|
57
|
+
# @param random_seed [Integer] The seed value using to initialize the random generator.
|
58
|
+
def initialize(n_factors: 2, loss: 'hinge', reg_param_linear: 1.0, reg_param_factor: 1.0,
|
59
|
+
max_iter: 1000, batch_size: 10, optimizer: nil, random_seed: nil)
|
60
|
+
check_params_float(reg_param_linear: reg_param_linear, reg_param_factor: reg_param_factor)
|
61
|
+
check_params_integer(n_factors: n_factors, max_iter: max_iter, batch_size: batch_size)
|
62
|
+
check_params_string(loss: loss)
|
63
|
+
check_params_type_or_nil(Integer, random_seed: random_seed)
|
64
|
+
check_params_positive(n_factors: n_factors,
|
65
|
+
reg_param_linear: reg_param_linear, reg_param_factor: reg_param_factor,
|
66
|
+
max_iter: max_iter, batch_size: batch_size)
|
67
|
+
super
|
68
|
+
@classes = nil
|
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
|
+
check_sample_array(x)
|
78
|
+
check_label_array(y)
|
79
|
+
check_sample_label_size(x, y)
|
80
|
+
|
81
|
+
@classes = Numo::Int32[*y.to_a.uniq.sort]
|
82
|
+
n_classes = @classes.size
|
83
|
+
_n_samples, n_features = x.shape
|
84
|
+
|
85
|
+
if n_classes > 2
|
86
|
+
@factor_mat = Numo::DFloat.zeros(n_classes, @params[:n_factors], n_features)
|
87
|
+
@weight_vec = Numo::DFloat.zeros(n_classes, n_features)
|
88
|
+
@bias_term = Numo::DFloat.zeros(n_classes)
|
89
|
+
n_classes.times do |n|
|
90
|
+
bin_y = Numo::Int32.cast(y.eq(@classes[n])) * 2 - 1
|
91
|
+
@factor_mat[n, true, true], @weight_vec[n, true], @bias_term[n] = partial_fit(x, bin_y)
|
92
|
+
end
|
93
|
+
else
|
94
|
+
negative_label = y.to_a.uniq.min
|
95
|
+
bin_y = Numo::Int32.cast(y.ne(negative_label)) * 2 - 1
|
96
|
+
@factor_mat, @weight_vec, @bias_term = partial_fit(x, bin_y)
|
97
|
+
end
|
98
|
+
|
99
|
+
self
|
100
|
+
end
|
101
|
+
|
102
|
+
# Calculate confidence scores for samples.
|
103
|
+
#
|
104
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to compute the scores.
|
105
|
+
# @return [Numo::DFloat] (shape: [n_samples]) Confidence score per sample.
|
106
|
+
def decision_function(x)
|
107
|
+
check_sample_array(x)
|
108
|
+
linear_term = @bias_term + x.dot(@weight_vec.transpose)
|
109
|
+
factor_term = if @classes.size <= 2
|
110
|
+
0.5 * (@factor_mat.dot(x.transpose)**2 - (@factor_mat**2).dot(x.transpose**2)).sum(0)
|
111
|
+
else
|
112
|
+
0.5 * (@factor_mat.dot(x.transpose)**2 - (@factor_mat**2).dot(x.transpose**2)).sum(1).transpose
|
113
|
+
end
|
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
|
+
check_sample_array(x)
|
123
|
+
return Numo::Int32.cast(decision_function(x).ge(0.0)) * 2 - 1 if @classes.size <= 2
|
124
|
+
|
125
|
+
n_samples, = x.shape
|
126
|
+
decision_values = decision_function(x)
|
127
|
+
Numo::Int32.asarray(Array.new(n_samples) { |n| @classes[decision_values[n, true].max_index] })
|
128
|
+
end
|
129
|
+
|
130
|
+
# Predict probability for samples.
|
131
|
+
#
|
132
|
+
# @param x [Numo::DFloat] (shape: [n_samples, n_features]) The samples to predict the probailities.
|
133
|
+
# @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted probability of each class per sample.
|
134
|
+
def predict_proba(x)
|
135
|
+
check_sample_array(x)
|
136
|
+
proba = 1.0 / (Numo::NMath.exp(-decision_function(x)) + 1.0)
|
137
|
+
return (proba.transpose / proba.sum(axis: 1)).transpose if @classes.size > 2
|
138
|
+
|
139
|
+
n_samples, = x.shape
|
140
|
+
probs = Numo::DFloat.zeros(n_samples, 2)
|
141
|
+
probs[true, 1] = proba
|
142
|
+
probs[true, 0] = 1.0 - proba
|
143
|
+
probs
|
144
|
+
end
|
145
|
+
|
146
|
+
# Dump marshal data.
|
147
|
+
# @return [Hash] The marshal data about FactorizationMachineClassifier.
|
148
|
+
def marshal_dump
|
149
|
+
{ params: @params,
|
150
|
+
factor_mat: @factor_mat,
|
151
|
+
weight_vec: @weight_vec,
|
152
|
+
bias_term: @bias_term,
|
153
|
+
classes: @classes,
|
154
|
+
rng: @rng }
|
155
|
+
end
|
156
|
+
|
157
|
+
# Load marshal data.
|
158
|
+
# @return [nil]
|
159
|
+
def marshal_load(obj)
|
160
|
+
@params = obj[:params]
|
161
|
+
@factor_mat = obj[:factor_mat]
|
162
|
+
@weight_vec = obj[:weight_vec]
|
163
|
+
@bias_term = obj[:bias_term]
|
164
|
+
@classes = obj[:classes]
|
165
|
+
@rng = obj[:rng]
|
166
|
+
nil
|
167
|
+
end
|
168
|
+
|
169
|
+
private
|
170
|
+
|
171
|
+
def bin_decision_function(x, ex_x, factor, weight)
|
172
|
+
ex_x.dot(weight) + 0.5 * (factor.dot(x.transpose)**2 - (factor**2).dot(x.transpose**2)).sum(0)
|
173
|
+
end
|
174
|
+
|
175
|
+
def hinge_loss_gradient(x, ex_x, y, factor, weight)
|
176
|
+
evaluated = y * bin_decision_function(x, ex_x, factor, weight)
|
177
|
+
gradient = Numo::DFloat.zeros(evaluated.size)
|
178
|
+
gradient[evaluated < 1.0] = -y[evaluated < 1.0]
|
179
|
+
gradient
|
180
|
+
end
|
181
|
+
|
182
|
+
def logistic_loss_gradient(x, ex_x, y, factor, weight)
|
183
|
+
evaluated = y * bin_decision_function(x, ex_x, factor, weight)
|
184
|
+
sigmoid_func = 1.0 / (Numo::NMath.exp(-evaluated) + 1.0)
|
185
|
+
(sigmoid_func - 1.0) * y
|
186
|
+
end
|
187
|
+
|
188
|
+
def loss_gradient(x, ex_x, y, factor, weight)
|
189
|
+
if @params[:loss] == 'hinge'
|
190
|
+
hinge_loss_gradient(x, ex_x, y, factor, weight)
|
191
|
+
else
|
192
|
+
logistic_loss_gradient(x, ex_x, y, factor, weight)
|
193
|
+
end
|
194
|
+
end
|
195
|
+
end
|
196
|
+
end
|
197
|
+
end
|