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
|
module OpenTox
class CrossValidation
field :validation_ids, type: Array, default: []
field :model_id, type: BSON::ObjectId
field :folds, type: Integer
field :nr_instances, type: Integer
field :nr_unpredicted, type: Integer
field :predictions, type: Array, default: []
field :finished_at, type: Time
def time
finished_at - created_at
end
def validations
validation_ids.collect{|vid| Validation.find vid}
end
def model
Model::Lazar.find model_id
end
def self.create model, n=10
cv = self.new(
name: model.name,
model_id: model.id,
folds: n
)
cv.save # set created_at
nr_instances = 0
nr_unpredicted = 0
predictions = []
validation_class = Object.const_get(self.to_s.sub(/Cross/,''))
training_dataset = Dataset.find model.training_dataset_id
training_dataset.folds(n).each_with_index do |fold,fold_nr|
fork do # parallel execution of validations
$logger.debug "Dataset #{training_dataset.name}: Fold #{fold_nr} started"
t = Time.now
#p validation_class#.create(model, fold[0], fold[1],cv)
validation = validation_class.create(model, fold[0], fold[1],cv)
#p validation
$logger.debug "Dataset #{training_dataset.name}, Fold #{fold_nr}: #{Time.now-t} seconds"
end
end
Process.waitall
cv.validation_ids = Validation.where(:crossvalidation_id => cv.id).distinct(:_id)
cv.validations.each do |validation|
nr_instances += validation.nr_instances
nr_unpredicted += validation.nr_unpredicted
predictions += validation.predictions
end
cv.update_attributes(
nr_instances: nr_instances,
nr_unpredicted: nr_unpredicted,
predictions: predictions
)
cv
end
end
class ClassificationCrossValidation < CrossValidation
field :accept_values, type: Array
field :confusion_matrix, type: Array
field :weighted_confusion_matrix, type: Array
field :accuracy, type: Float
field :weighted_accuracy, type: Float
field :true_rate, type: Hash
field :predictivity, type: Hash
# TODO auc, f-measure (usability??)
def self.create model, n=10
cv = super model, n
accept_values = Feature.find(model.prediction_feature_id).accept_values
confusion_matrix = Array.new(accept_values.size,0){Array.new(accept_values.size,0)}
weighted_confusion_matrix = Array.new(accept_values.size,0){Array.new(accept_values.size,0)}
true_rate = {}
predictivity = {}
cv.predictions.each do |pred|
compound_id,activity,prediction,confidence = pred
if activity and prediction and confidence.numeric?
if prediction == activity
if prediction == accept_values[0]
confusion_matrix[0][0] += 1
weighted_confusion_matrix[0][0] += confidence
elsif prediction == accept_values[1]
confusion_matrix[1][1] += 1
weighted_confusion_matrix[1][1] += confidence
end
elsif prediction != activity
if prediction == accept_values[0]
confusion_matrix[0][1] += 1
weighted_confusion_matrix[0][1] += confidence
elsif prediction == accept_values[1]
confusion_matrix[1][0] += 1
weighted_confusion_matrix[1][0] += confidence
end
end
else
nr_unpredicted += 1 if prediction.nil?
end
end
true_rate = {}
predictivity = {}
accept_values.each_with_index do |v,i|
true_rate[v] = confusion_matrix[i][i]/confusion_matrix[i].reduce(:+).to_f
predictivity[v] = confusion_matrix[i][i]/confusion_matrix.collect{|n| n[i]}.reduce(:+).to_f
end
confidence_sum = 0
weighted_confusion_matrix.each do |r|
r.each do |c|
confidence_sum += c
end
end
cv.update_attributes(
accept_values: accept_values,
confusion_matrix: confusion_matrix,
weighted_confusion_matrix: weighted_confusion_matrix,
accuracy: (confusion_matrix[0][0]+confusion_matrix[1][1])/(cv.nr_instances-cv.nr_unpredicted).to_f,
weighted_accuracy: (weighted_confusion_matrix[0][0]+weighted_confusion_matrix[1][1])/confidence_sum.to_f,
true_rate: true_rate,
predictivity: predictivity,
finished_at: Time.now
)
cv.save
cv
end
#Average area under roc 0.646
#Area under roc 0.646
#F measure carcinogen: 0.769, noncarcinogen: 0.348
end
class RegressionCrossValidation < CrossValidation
field :rmse, type: Float
field :mae, type: Float
field :weighted_rmse, type: Float
field :weighted_mae, type: Float
field :r_squared, type: Float
field :correlation_plot_id, type: BSON::ObjectId
field :confidence_plot_id, type: BSON::ObjectId
def self.create model, n=10
cv = super model, n
rmse = 0
weighted_rmse = 0
rse = 0
weighted_rse = 0
mae = 0
weighted_mae = 0
rae = 0
weighted_rae = 0
confidence_sum = 0
cv.predictions.each do |pred|
compound_id,activity,prediction,confidence = pred
if activity and prediction
error = Math.log10(prediction)-Math.log10(activity)
rmse += error**2
weighted_rmse += confidence*error**2
mae += error.abs
weighted_mae += confidence*error.abs
confidence_sum += confidence
else
cv.warnings << "No training activities for #{Compound.find(compound_id).smiles} in training dataset #{cv.model.training_dataset_id}."
$logger.debug "No training activities for #{Compound.find(compound_id).smiles} in training dataset #{cv.model.training_dataset_id}."
end
end
x = cv.predictions.collect{|p| p[1]}
y = cv.predictions.collect{|p| p[2]}
R.assign "measurement", x
R.assign "prediction", y
R.eval "r <- cor(-log(measurement),-log(prediction))"
r = R.eval("r").to_ruby
mae = mae/cv.predictions.size
weighted_mae = weighted_mae/confidence_sum
rmse = Math.sqrt(rmse/cv.predictions.size)
weighted_rmse = Math.sqrt(weighted_rmse/confidence_sum)
# TODO check!!
=begin
cv.predictions.sort! do |a,b|
relative_error_a = (a[1]-a[2]).abs/a[1].to_f
relative_error_a = 1/relative_error_a if relative_error_a < 1
relative_error_b = (b[1]-b[2]).abs/b[1].to_f
relative_error_b = 1/relative_error_b if relative_error_b < 1
[relative_error_b,b[3]] <=> [relative_error_a,a[3]]
end
=end
cv.update_attributes(
mae: mae,
rmse: rmse,
weighted_mae: weighted_mae,
weighted_rmse: weighted_rmse,
r_squared: r**2
)
cv.save
cv
end
def misclassifications n=nil
#n = predictions.size unless n
n = 20 unless n
model = Model::Lazar.find(self.model_id)
training_dataset = Dataset.find(model.training_dataset_id)
prediction_feature = training_dataset.features.first
predictions[0..n-1].collect do |p|
compound = Compound.find(p[0])
neighbors = compound.neighbors.collect do |n|
neighbor = Compound.find(n[0])
values = training_dataset.values(neighbor,prediction_feature)
{ :smiles => neighbor.smiles, :fingerprint => neighbor.fp4.collect{|id| Smarts.find(id).name},:similarity => n[1], :measurements => values}
end
{
:smiles => compound.smiles,
:fingerprint => compound.fp4.collect{|id| Smarts.find(id).name},
:measured => p[1],
:predicted => p[2],
:relative_error => (p[1]-p[2]).abs/p[1].to_f,
:confidence => p[3],
:neighbors => neighbors
}
end
end
def confidence_plot
tmpfile = "/tmp/#{id.to_s}_confidence.svg"
sorted_predictions = predictions.sort{|a,b| b[3]<=>a[3]}.collect{|p| [(Math.log10(p[1])-Math.log10(p[2]))**2,p[3]]}
R.assign "error", sorted_predictions.collect{|p| p[0]}
#R.assign "p", predictions.collect{|p| p[2]}
R.assign "confidence", predictions.collect{|p| p[2]}
#R.eval "diff = log(m)-log(p)"
R.eval "library(ggplot2)"
R.eval "svg(filename='#{tmpfile}')"
R.eval "image = qplot(confidence,error)"#,main='#{self.name}',asp=1,xlim=range, ylim=range)"
R.eval "ggsave(file='#{tmpfile}', plot=image)"
R.eval "dev.off()"
file = Mongo::Grid::File.new(File.read(tmpfile), :filename => "#{self.id.to_s}_confidence_plot.svg")
plot_id = $gridfs.insert_one(file)
update(:confidence_plot_id => plot_id)
$gridfs.find_one(_id: confidence_plot_id).data
end
def correlation_plot
unless correlation_plot_id
tmpfile = "/tmp/#{id.to_s}_correlation.svg"
x = predictions.collect{|p| p[1]}
y = predictions.collect{|p| p[2]}
attributes = Model::Lazar.find(self.model_id).attributes
attributes.delete_if{|key,_| key.match(/_id|_at/) or ["_id","creator","name"].include? key}
attributes = attributes.values.collect{|v| v.is_a?(String) ? v.sub(/OpenTox::/,'') : v}.join("\n")
p "'"+attributes
R.eval "library(ggplot2)"
R.eval "library(grid)"
R.eval "library(gridExtra)"
R.assign "measurement", x
R.assign "prediction", y
#R.eval "error <- log(Measurement)-log(Prediction)"
#R.eval "rmse <- sqrt(mean(error^2, na.rm=T))"
#R.eval "mae <- mean(abs(error), na.rm=T)"
#R.eval "r <- cor(-log(prediction),-log(measurement))"
R.eval "svg(filename='#{tmpfile}')"
R.eval "all = c(-log(measurement),-log(prediction))"
R.eval "range = c(min(all), max(all))"
R.eval "image = qplot(-log(prediction),-log(measurement),main='#{self.name}',asp=1,xlim=range, ylim=range)"
R.eval "image = image + geom_abline(intercept=0, slope=1) + stat_smooth(method='lm', se=FALSE)"
R.eval "text = textGrob(paste('RMSE: ', '#{rmse.round(2)},','MAE:','#{mae.round(2)},','r^2: ','#{r_squared.round(2)}','\n\n','#{attributes}'),just=c('left','top'),check.overlap = T)"
R.eval "grid.arrange(image, text, ncol=2)"
R.eval "dev.off()"
file = Mongo::Grid::File.new(File.read(tmpfile), :filename => "#{self.id.to_s}_correlation_plot.svg")
plot_id = $gridfs.insert_one(file)
update(:correlation_plot_id => plot_id)
end
p correlation_plot_id
$gridfs.find_one(_id: correlation_plot_id).data
end
end
end
|