1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
|
#require "rinruby"
# TODO install R packages kernlab, caret, doMC, class, e1071
# TODO use Rserve
module OpenTox
module Algorithm
class Regression
require "rserve"
# Local support vector regression from neighbors
# @param [Hash] params Keys `:props, :activities, :sims, :min_train_performance` are required
# @return [Numeric] A prediction value.
def self.local_svm_regression neighbors, params={:min_train_performance => 0.1}
confidence = 0.0
prediction = nil
$logger.debug "Local SVM."
props = neighbors.collect{|row| row[3] }
neighbors.shift
activities = neighbors.collect{|n| n[2]}
prediction = self.local_svm_prop( props, activities, params[:min_train_performance]) # params[:props].nil? signals non-prop setting
prediction = nil if (!prediction.nil? && prediction.infinite?)
$logger.debug "Prediction: '#{prediction}' ('#{prediction.class}')."
if prediction
confidence = get_confidence({:sims => neighbors.collect{|n| n[1]}, :activities => activities})
else
confidence = nil if prediction.nil?
end
[prediction, confidence]
end
# Local support vector prediction from neighbors.
# Uses propositionalized setting.
# Not to be called directly (use local_svm_regression or local_svm_classification).
# @param [Array] props, propositionalization of neighbors and query structure e.g. [ Array_for_q, two-nested-Arrays_for_n ]
# @param [Array] activities, activities for neighbors.
# @param [Float] min_train_performance, parameter to control censoring
# @return [Numeric] A prediction value.
def self.local_svm_prop(props, activities, min_train_performance)
$logger.debug "Local SVM (Propositionalization / Kernlab Kernel)."
n_prop = props[1..-1] # is a matrix, i.e. two nested Arrays.
q_prop = props[0] # is an Array.
prediction = nil
if activities.uniq.size == 1
prediction = activities[0]
else
t = Time.now
#$logger.debug gram_matrix.to_yaml
#@r = RinRuby.new(true,false) # global R instance leads to Socket errors after a large number of requests
@r = Rserve::Connection.new#(true,false) # global R instance leads to Socket errors after a large number of requests
rs = []
["caret", "doMC", "class"].each do |lib|
#raise "failed to load R-package #{lib}" unless @r.void_eval "suppressPackageStartupMessages(library('#{lib}'))"
rs << "suppressPackageStartupMessages(library('#{lib}'))"
end
#@r.eval "registerDoMC()" # switch on parallel processing
rs << "registerDoMC()" # switch on parallel processing
#@r.eval "set.seed(1)"
rs << "set.seed(1)"
$logger.debug "Loading R packages: #{Time.now-t}"
t = Time.now
p n_prop
begin
# set data
rs << "n_prop <- c(#{n_prop.flatten.join(',')})"
rs << "n_prop <- c(#{n_prop.flatten.join(',')})"
rs << "n_prop_x_size <- c(#{n_prop.size})"
rs << "n_prop_y_size <- c(#{n_prop[0].size})"
rs << "y <- c(#{activities.join(',')})"
rs << "q_prop <- c(#{q_prop.join(',')})"
rs << "y = matrix(y)"
rs << "prop_matrix = matrix(n_prop, n_prop_x_size, n_prop_y_size, byrow=T)"
rs << "q_prop = matrix(q_prop, 1, n_prop_y_size, byrow=T)"
$logger.debug "Setting R data: #{Time.now-t}"
t = Time.now
# prepare data
rs << "
weights=NULL
if (!(class(y) == 'numeric')) {
y = factor(y)
weights=unlist(as.list(prop.table(table(y))))
weights=(weights-1)^2
}
"
rs << "
rem = nearZeroVar(prop_matrix)
if (length(rem) > 0) {
prop_matrix = prop_matrix[,-rem,drop=F]
q_prop = q_prop[,-rem,drop=F]
}
rem = findCorrelation(cor(prop_matrix))
if (length(rem) > 0) {
prop_matrix = prop_matrix[,-rem,drop=F]
q_prop = q_prop[,-rem,drop=F]
}
"
#p @r.eval("y").to_ruby
#p "weights"
#p @r.eval("weights").to_ruby
$logger.debug "Preparing R data: #{Time.now-t}"
t = Time.now
# model + support vectors
#train_success = @r.eval <<-EOR
rs << '
model = train(prop_matrix,y,
method="svmRadial",
preProcess=c("center", "scale"),
class.weights=weights,
trControl=trainControl(method="LGOCV",number=10),
tuneLength=8
)
perf = ifelse ( class(y)!="numeric", max(model$results$Accuracy), model$results[which.min(model$results$RMSE),]$Rsquared )
'
File.open("/tmp/r.r","w+"){|f| f.puts rs.join("\n")}
p rs.join("\n")
p `Rscript /tmp/r.r`
=begin
@r.void_eval <<-EOR
model = train(prop_matrix,y,
method="svmRadial",
#preProcess=c("center", "scale"),
#class.weights=weights,
#trControl=trainControl(method="LGOCV",number=10),
#tuneLength=8
)
perf = ifelse ( class(y)!='numeric', max(model$results$Accuracy), model$results[which.min(model$results$RMSE),]$Rsquared )
EOR
=end
$logger.debug "Creating R SVM model: #{Time.now-t}"
t = Time.now
if train_success
# prediction
@r.eval "predict(model,q_prop); p = predict(model,q_prop)" # kernlab bug: predict twice
#@r.eval "p = predict(model,q_prop)" # kernlab bug: predict twice
@r.eval "if (class(y)!='numeric') p = as.character(p)"
prediction = @r.p
# censoring
prediction = nil if ( @r.perf.nan? || @r.perf < min_train_performance.to_f )
prediction = nil if prediction =~ /NA/
$logger.debug "Performance: '#{sprintf("%.2f", @r.perf)}'"
else
$logger.debug "Model creation failed."
prediction = nil
end
$logger.debug "R Prediction: #{Time.now-t}"
rescue Exception => e
$logger.debug "#{e.class}: #{e.message}"
$logger.debug "Backtrace:\n\t#{e.backtrace.join("\n\t")}"
ensure
#puts @r.inspect
#TODO: broken pipe
#@r.quit # free R
end
end
prediction
end
end
end
end
|
