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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
|
module OpenTox
module Algorithm
class Regression
def self.weighted_average compound, params
weighted_sum = 0.0
sim_sum = 0.0
confidence = 0.0
neighbors = params[:neighbors]
neighbors.each do |row|
sim = row["tanimoto"]
confidence = sim if sim > confidence # distance to nearest neighbor
# TODO add LOO errors
row["features"][params[:prediction_feature_id].to_s].each do |act|
weighted_sum += sim*Math.log10(act)
#activities << act # TODO: Transformation??
sim_sum += sim
end
end
#R.assign "activities", activities
#R.eval "cv = cv(activities)"
#confidence /= activities.standard_deviation#/activities.mean
#confidence = sim_sum*neighbors.size.to_f/params[:training_dataset_size]
#confidence = sim_sum/neighbors.size.to_f
#confidence = neighbors.size.to_f
confidence = 0 if confidence.nan?
sim_sum == 0 ? prediction = nil : prediction = 10**(weighted_sum/sim_sum)
{:value => prediction,:confidence => confidence}
end
def self.local_pls_regression compound, params
neighbors = params[:neighbors]
return {:value => nil, :confidence => nil, :warning => "No similar compounds in the training data"} unless neighbors.size > 0
activities = []
fingerprints = {}
weights = []
fingerprint_ids = neighbors.collect{|row| Compound.find(row["_id"]).fingerprint}.flatten.uniq.sort
neighbors.each_with_index do |row,i|
neighbor = Compound.find row["_id"]
fingerprint = neighbor.fingerprint
row["features"][params[:prediction_feature_id].to_s].each do |act|
activities << Math.log10(act)
weights << row["tanimoto"]
fingerprint_ids.each_with_index do |id,j|
fingerprints[id] ||= []
fingerprints[id] << fingerprint.include?(id)
end
end
end
name = Feature.find(params[:prediction_feature_id]).name
R.assign "activities", activities
R.assign "weights", weights
variables = []
data_frame = ["c(#{activities.join ","})"]
fingerprints.each do |k,v|
unless v.uniq.size == 1
data_frame << "factor(c(#{v.collect{|m| m ? "T" : "F"}.join ","}))"
variables << k
end
end
if variables.empty?
result = weighted_average(compound, params)
result[:warning] = "No variables for regression model. Using weighted average of similar compounds."
return result
return {:value => nil, :confidence => nil} # TODO confidence
else
R.eval "data <- data.frame(#{data_frame.join ","})"
R.assign "features", variables
R.eval "names(data) <- append(c('activities'),features)" #
begin
R.eval "model <- plsr(activities ~ .,data = data, ncomp = 4, weights = weights)"
rescue # fall back to weighted average
result = weighted_average(compound, params)
result[:warning] = "Could not create local PLS model. Using weighted average of similar compounds."
return result
end
#begin
#compound_features = fingerprint_ids.collect{|f| compound.fingerprint.include? f } # FIX
compound_features = variables.collect{|f| compound.fingerprint.include? f }
R.eval "fingerprint <- rbind(c(#{compound_features.collect{|f| f ? "T" : "F"}.join ','}))"
R.eval "names(fingerprint) <- features" #
R.eval "prediction <- predict(model,fingerprint)"
prediction = 10**R.eval("prediction").to_f
return {:value => prediction, :confidence => 1} # TODO confidence
#rescue
#p "Prediction failed"
#return {:value => nil, :confidence => nil} # TODO confidence
#end
end
end
def self.local_physchem_regression compound, params
neighbors = params[:neighbors]
return {:value => nil, :confidence => nil, :warning => "No similar compounds in the training data"} unless neighbors.size > 0
activities = []
fingerprints = {}
weights = []
fingerprint_ids = neighbors.collect{|row| Compound.find(row["_id"]).fingerprint}.flatten.uniq.sort
neighbors.each_with_index do |row,i|
neighbor = Compound.find row["_id"]
fingerprint = neighbor.fingerprint
row["features"][params[:prediction_feature_id].to_s].each do |act|
activities << Math.log10(act)
weights << row["tanimoto"]
fingerprint_ids.each_with_index do |id,j|
fingerprints[id] ||= []
fingerprints[id] << fingerprint.include?(id)
end
end
end
name = Feature.find(params[:prediction_feature_id]).name
R.assign "activities", activities
R.assign "weights", weights
variables = []
data_frame = ["c(#{activities.join ","})"]
fingerprints.each do |k,v|
unless v.uniq.size == 1
data_frame << "factor(c(#{v.collect{|m| m ? "T" : "F"}.join ","}))"
variables << k
end
end
if variables.empty?
result = weighted_average(compound, params)
result[:warning] = "No variables for regression model. Using weighted average of similar compounds."
return result
return {:value => nil, :confidence => nil} # TODO confidence
else
R.eval "data <- data.frame(#{data_frame.join ","})"
R.assign "features", variables
R.eval "names(data) <- append(c('activities'),features)" #
begin
R.eval "model <- plsr(activities ~ .,data = data, ncomp = 4, weights = weights)"
rescue # fall back to weighted average
result = weighted_average(compound, params)
result[:warning] = "Could not create local PLS model. Using weighted average of similar compounds."
return result
end
#begin
#compound_features = fingerprint_ids.collect{|f| compound.fingerprint.include? f } # FIX
compound_features = variables.collect{|f| compound.fingerprint.include? f }
R.eval "fingerprint <- rbind(c(#{compound_features.collect{|f| f ? "T" : "F"}.join ','}))"
R.eval "names(fingerprint) <- features" #
R.eval "prediction <- predict(model,fingerprint)"
prediction = 10**R.eval("prediction").to_f
return {:value => prediction, :confidence => 1} # TODO confidence
#rescue
#p "Prediction failed"
#return {:value => nil, :confidence => nil} # TODO confidence
#end
end
end
def self.weighted_average_with_relevant_fingerprints neighbors
weighted_sum = 0.0
sim_sum = 0.0
fingerprint_features = []
neighbors.each do |row|
n,sim,acts = row
neighbor = Compound.find n
fingerprint_features += neighbor.fp4
end
fingerprint_features.uniq!
p fingerprint_features
=begin
p n
acts.each do |act|
weighted_sum += sim*Math.log10(act)
sim_sum += sim
end
end
=end
confidence = sim_sum/neighbors.size.to_f
sim_sum == 0 ? prediction = nil : prediction = 10**(weighted_sum/sim_sum)
{:value => prediction,:confidence => confidence}
end
# 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
|