fix median for medianConfidence, add confidence to validation visualization, adjust to new super strat methods

3 files changed, 25 insertions, 22 deletions

diff --git a/lib/predictions.rb b/lib/predictions.rb
index eb13b31..63578fd 100755
--- a/lib/predictions.rb
+++ b/lib/predictions.rb
@@ -1,14 +1,6 @@
 
 require "lib/prediction_data.rb"
-
-
-class Array
-  
-  def median #array has to be sorted!     
-    (self[size/2] + self[(size+1)/2]) / 2.0
-  end 
-  
-end
+require "statsample"
 
 module Lib
 
@@ -668,7 +660,7 @@ module Lib
     # data for (roc-)plots ###################################################################################
     
     def median_confidence
-      @confidence_values.median if confidence_values_available?
+      @confidence_values.to_scale.median if confidence_values_available?
     end
     
      def get_roc_prediction_values(class_value)
diff --git a/validation/validation_application.rb b/validation/validation_application.rb
index f9a5b1c..689241c 100755
--- a/validation/validation_application.rb
+++ b/validation/validation_application.rb
@@ -9,8 +9,8 @@ require 'validation/validation_service.rb'
 helpers do
   def check_stratified(params)
     params[:stratified] = "false" unless params[:stratified]
-    raise OpenTox::BadRequestError.new "stratified != true|false|super|anti, is #{params[:stratified]}" unless
-      params[:stratified]=~/true|false|super|anti/
+    raise OpenTox::BadRequestError.new "stratified != true|false|super|super4|anti, is #{params[:stratified]}" unless
+      params[:stratified]=~/^(true|false|super|super4|anti)$/
   end
 end
 
@@ -583,8 +583,10 @@ post '/plain_training_test_split' do
   raise OpenTox::BadRequestError.new "dataset_uri missing" unless params[:dataset_uri]
   check_stratified(params)
   task = OpenTox::Task.create( "Create data-split", url_for("/plain_training_test_split", :full) ) do |task|
+    split_features = nil
+    split_features = params[:split_features].split(";") if params[:split_features]
     result = Validation::Util.train_test_dataset_split(params[:dataset_uri], params[:prediction_feature], @subjectid,
-      params[:stratified], params[:split_ratio], params[:random_seed], params[:missing_values], task)
+      params[:stratified], params[:split_ratio], params[:random_seed], params[:missing_values], task, split_features)
     content_type "text/uri-list"
     res = result[:training_dataset_uri]+"\n"+result[:test_dataset_uri]+"\n"
     LOGGER.info "plain training test split done #{res.to_s.gsub("\n"," \\n ")}"      
@@ -651,6 +653,7 @@ get '/:id/viz' do
   
   m = OpenTox::Model::Generic.find(validation.model_uri)
   predicted_feature = m.predicted_variable(nil)
+  confidence_feature = m.predicted_confidence(nil)
   actual_feature = validation.prediction_feature
   
   d = OpenTox::Dataset.create
@@ -665,6 +668,8 @@ get '/:id/viz' do
   d.add_feature(correct_classified_feature)
   predicted_nice_feature = "http://predicted"
   d.add_feature(predicted_nice_feature)
+  confidence_nice_feature = "http://confidence"
+  d.add_feature(confidence_nice_feature)
     
   [training, test].each do |data|
     data.compounds.each do |c|
@@ -689,6 +694,11 @@ get '/:id/viz' do
       d.add(c,predicted_nice_feature,p[0],true)
       d.add(c,correct_classified_feature,p[0]==a[0] ? "correct" : "miss",true)
     end 
+    if prediction.data_entries[c][confidence_feature]
+      conf = prediction.data_entries[c][confidence_feature]
+      raise if conf.size!=1
+      d.add(c,confidence_nice_feature,conf[0],true)
+    end
   end
   d.to_csv
 end
diff --git a/validation/validation_service.rb b/validation/validation_service.rb
index 5fded24..1b29630 100755
--- a/validation/validation_service.rb
+++ b/validation/validation_service.rb
@@ -500,8 +500,8 @@ module Validation
             meta, self.subjectid ).uri
           test_dataset_uris << test_dataset_uri
         end
-      when /true|super/
-        raise "DO NOT USED SUPER-STRATIFICATION FOR VAL-EXPERIMENTS AND CV, IF SO SOLVE _MISSING_VAULE_NA_ PROBLEM" if stratified=="super"
+      when /^(true|super|super4)$/
+        raise "DO NOT USED SUPER-STRATIFICATION FOR VAL-EXPERIMENTS AND CV, IF SO SOLVE _MISSING_VAULE_NA_ PROBLEM" if stratified=~/super/
         if stratified=="true"
           features = [ self.prediction_feature ] 
         else
@@ -627,7 +627,7 @@ module Validation
     # splits a dataset into test and training dataset
     # returns map with training_dataset_uri and test_dataset_uri
     def self.train_test_dataset_split( orig_dataset_uri, prediction_feature, subjectid, stratified="false",
-      split_ratio=nil, random_seed=nil, missing_values=nil, task=nil )
+      split_ratio=nil, random_seed=nil, missing_values=nil, task=nil, features=nil )
       
       split_ratio=0.67 unless split_ratio
       split_ratio = split_ratio.to_f
@@ -651,30 +651,31 @@ module Validation
       meta = { DC.creator => $url_provider.url_for('/training_test_split',:full) }
       
       case stratified
-      when /true|super|anti/
+      when /^(true|super|super4|anti)$/
         if stratified=="true"
           raise OpenTox::BadRequestError.new "prediction feature required for stratified splits" unless prediction_feature
+          LOGGER.warn "split features are ignored for stratified splits (use super instead)" if features
           features = [prediction_feature]
         else
           LOGGER.warn "prediction feature is ignored for super- or anti-stratified splits" if prediction_feature
-          features = nil
         end
         r_util = OpenTox::RUtil.new 
-        train, test = r_util.stratified_split( orig_dataset, meta, missing_values, split_ratio, @subjectid, random_seed, features, stratified=="anti" )
+        train, test = r_util.stratified_split( orig_dataset, meta, missing_values, split_ratio, @subjectid, random_seed, features, stratified )
         r_util.quit_r
         result = {:training_dataset_uri => train.uri, :test_dataset_uri => test.uri}
       when "false"
+        LOGGER.warn "split features are ignored for non-stratified splits (use super instead)" if features
         compounds = orig_dataset.compounds
         raise OpenTox::BadRequestError.new "Cannot split datset, num compounds in dataset < 2 ("+compounds.size.to_s+")" if compounds.size<2
         split = (compounds.size*split_ratio).to_i
         split = [split,1].max
         split = [split,compounds.size-2].min
         LOGGER.debug "splitting dataset "+orig_dataset_uri+
-                    " into train:0-"+split.to_s+" and test:"+(split+1).to_s+"-"+(compounds.size-1).to_s+
+                    " into train:0-"+(split-1).to_s+" and test:"+split.to_s+"-"+(compounds.size-1).to_s+
                     " (shuffled with seed "+random_seed.to_s+")"
         compounds.shuffle!( random_seed )
-        training_compounds = compounds[0..split]
-        test_compounds = compounds[(split+1)..-1]
+        training_compounds = compounds[0..(split-1)]
+        test_compounds = compounds[split..-1]
         task.progress(33) if task
   
         meta[DC.title] = "Training dataset split of "+orig_dataset.uri