path: root/lazar.rb

@@feature_generation_default = File.join(CONFIG[:services]["opentox-algorithm"],"fminer","bbrc")

# Get RDF/XML representation of the lazar algorithm
# @return [application/rdf+xml] OWL-DL representation of the lazar algorithm
get '/lazar/?' do
  algorithm = OpenTox::Algorithm::Generic.new(url_for('/lazar',:full))
  algorithm.metadata = {
    DC.title => 'lazar',
    DC.creator => "helma@in-silico.ch, andreas@maunz.de",
    DC.contributor => "vorgrimmlerdavid@gmx.de",
#    BO.instanceOf => "http://opentox.org/ontology/ist-algorithms.owl#lazar",
    OT.parameters => [
      { DC.description => "Dataset URI with the dependent variable", OT.paramScope => "mandatory", DC.title => "dataset_uri" },
      { DC.description => "Feature URI for dependent variable. Optional for datasets with only a single feature.", OT.paramScope => "optional", DC.title => "prediction_feature" },
      { DC.description => "URI of feature generation service. Default: #{@@feature_generation_default}", OT.paramScope => "optional", DC.title => "feature_generation_uri" },
      { DC.description => "URI of feature dataset. If this parameter is set no feature generation algorithm will be called", OT.paramScope => "optional", DC.title => "feature_dataset_uri" },
      { DC.description => "Further parameters for the feature generation service", OT.paramScope => "optional" }
    ]
  }
  case request.env['HTTP_ACCEPT']
  when /text\/html/
    content_type "text/html"
    OpenTox.text_to_html algorithm.to_yaml
  when /application\/x-yaml/
    content_type "application/x-yaml"
    algorithm.to_yaml
  else
    response['Content-Type'] = 'application/rdf+xml'  
    algorithm.to_rdfxml
  end
end

# Create a lazar prediction model
# @param [String] dataset_uri Training dataset URI
# @param [optional,String] prediction_feature URI of the feature to be predicted
# @param [optional,String] feature_generation_uri URI of the feature generation algorithm 
# @param [optional,String] - further parameters for the feature generation service 
# @return [text/uri-list] Task URI 
post '/lazar/?' do 

  params[:subjectid] = @subjectid
  raise OpenTox::NotFoundError.new "No dataset_uri parameter." unless params[:dataset_uri]
	dataset_uri = params[:dataset_uri]


  task = OpenTox::Task.create("Create lazar model",url_for('/lazar',:full)) do |task|

    # # # BASIC SETTINGS
    
    raise OpenTox::NotFoundError.new "Dataset #{dataset_uri} not found." unless training_dataset = OpenTox::Dataset.new(dataset_uri)
    training_dataset.load_all(@subjectid)

    # Prediction Feature
    prediction_feature = OpenTox::Feature.find(params[:prediction_feature],@subjectid)
    unless params[:prediction_feature] # try to read prediction_feature from dataset
      raise OpenTox::NotFoundError.new "#{training_dataset.features.size} features in dataset #{dataset_uri}. Please provide a prediction_feature parameter." unless training_dataset.features.size == 1
      prediction_feature = OpenTox::Feature.find(training_dataset.features.keys.first,@subjectid)
      params[:prediction_feature] = prediction_feature.uri # pass to feature mining service
    end
    raise OpenTox::NotFoundError.new "No feature #{prediction_feature.uri} in dataset #{params[:dataset_uri]}. (features: "+ training_dataset.features.inspect+")" unless training_dataset.features and training_dataset.features.include?(prediction_feature.uri)
    
    # Feature Generation URI
    feature_generation_uri = @@feature_generation_default unless ( (feature_generation_uri = params[:feature_generation_uri]) || (params[:feature_dataset_uri]) )

    # Create instance
		lazar = OpenTox::Model::Lazar.new
    
    # Classification: Weighted Majority, Substructure.match
    if prediction_feature.feature_type == "classification"
      lazar.value_map = training_dataset.value_map(params[:prediction_feature])

    # Regression: SVM, Substructure.match_hits
    elsif  prediction_feature.feature_type == "regression"
      lazar.feature_calculation_algorithm = "Substructure.match_hits" 
      lazar.prediction_algorithm = "Neighbors.local_svm_regression" 
    end


    # # # USER VALUES
    
    # Min Sim
    min_sim = params[:min_sim].to_f if params[:min_sim]
    min_sim = 0.3 unless params[:min_sim]

    # Algorithm
    lazar.prediction_algorithm = "Neighbors.#{params[:prediction_algorithm]}" if params[:prediction_algorithm]

    # Nr Hits
    nr_hits = false
    if params[:nr_hits] == "true" || lazar.prediction_algorithm.include?("local_svm")
      lazar.feature_calculation_algorithm = "Substructure.match_hits"
      nr_hits = true
    end
    params[:nr_hits] = "true" if lazar.feature_calculation_algorithm == "Substructure.match_hits" #not sure if this line in needed 

    # Propositionalization
    propositionalized = (lazar.prediction_algorithm=="Neighbors.weighted_majority_vote" ? false : true)
   
    # PC type
    pc_type = params[:pc_type] if params[:pc_type]
    lib = params[:lib] if params[:lib]

    # Min train performance
    min_train_performance = params[:min_train_performance].to_f if params[:min_train_performance]
    min_train_performance = 0.1 unless params[:min_train_performance]


    task.progress 10


    # # # Features

    # Read Features, currently only OT.NumericFeatures
    if params[:feature_dataset_uri]
      lazar.feature_calculation_algorithm = "Substructure.lookup"
      feature_dataset_uri = params[:feature_dataset_uri]
      training_features = OpenTox::Dataset.new(feature_dataset_uri)
      training_feature_types = training_features.feature_types(@subjectid)

      if training_feature_types.collect { |id, info| info.include? OT.NumericFeature }.include?(false) # <- extend this
        raise OpenTox::NotFoundError.new "Found a non-numeric feature in feature dataset"

      elsif training_dataset.compounds.size < training_feature_types.size
        raise OpenTox::BadRequestError.new "Number of training compounds (#{training_dataset.compounds.size}) smaller than number of non-missing features (#{training_feature_types.size})"

      else
        lazar.similarity_algorithm = "Similarity.cosine"
        min_sim = 0.7 unless params[:min_sim] 
        min_sim = (min_sim * 2.0 -1.0 ) # transform to cosine range [-1,1]

        training_features_tl = training_features.features.collect{|f,info|
          unless info[DC.description]
            [nil, nil]
          else
            info[DC.description].gsub(/.*\[/,"").chop.split(", ")
          end
        
        }
        training_features_pc_types = training_features_tl.collect{|info| info[0]}.flatten.uniq.compact
        training_features_lib = training_features_tl.collect{|info| info[1]}.flatten.uniq.compact
        unless (params[:pc_type] and params[:lib])
          
          if (!params[:pc_type] && training_features_pc_types.size>0)
            pc_type=training_features_pc_types.join(',')
            LOGGER.info "pc_type '#{pc_type}' auto-detected from feature dataset"
          end
          
          if (!params[:lib] && training_features_lib.size>0)
            lib=training_features_lib.join(',')
            LOGGER.info "lib '#{lib}' auto-detected from feature dataset"
          end
          
          unless (pc_type and lib)
            raise OpenTox::NotFoundError.new "No pc_type parameter given, and autodetection from feature dataset failed" unless pc_type
            raise OpenTox::NotFoundError.new "No lib parameter given, and autodetection from feature dataset failed" unless lib
          end
        
        end
      
      end

    # Create Features
    else 
      params[:feature_generation_uri] = feature_generation_uri
      params[:subjectid] = @subjectid
      prediction_feature = OpenTox::Feature.find params[:prediction_feature], @subjectid
      if prediction_feature.feature_type == "regression" && feature_generation_uri.match(/fminer/) 
        params[:feature_type] = "paths" unless params[:feature_type]
      end
      feature_dataset_uri = OpenTox::Algorithm::Generic.new(feature_generation_uri).run(params, OpenTox::SubTask.new(task,10,70)).to_s
      training_features = OpenTox::Dataset.new(feature_dataset_uri)
    end


    # # # Write fingerprints
    
    training_features.load_all(@subjectid)
		raise OpenTox::NotFoundError.new "Dataset #{feature_dataset_uri} not found." if training_features.nil?

    del_master_compounds = []
 
    # Creating InChi/URI Hash from trainig_feature for comparison with training_dataset to avoid missmatches caused by different URI authorities
    feature_compounds = {}
    training_features.compounds.each {|f_c_uri|
      f_compound = OpenTox::Compound.new(f_c_uri)
      feature_compounds[f_compound.to_inchi] = f_c_uri
    }
   
    training_dataset.compounds.each do |t_c_uri|

      t_compound = OpenTox::Compound.new(t_c_uri)
      entry = training_features.data_entries[feature_compounds[t_compound.to_inchi]]
      
      if entry.nil? # Training compound not found in feature dataset
        del_master_compounds << t_c_uri # Delete if training compound not found in feature dataset
      else
        lazar.fingerprints[t_c_uri] = {} unless lazar.fingerprints[t_c_uri]
        entry.keys.each do |feature|

          # CASE 1: Substructure
          if (lazar.feature_calculation_algorithm == "Substructure.match") || (lazar.feature_calculation_algorithm == "Substructure.match_hits")
            if training_features.features[feature]
              smarts = training_features.features[feature][OT.smarts]
              #lazar.fingerprints[compound] << smarts
              if lazar.feature_calculation_algorithm == "Substructure.match_hits"
                lazar.fingerprints[t_c_uri][smarts] = entry[feature].flatten.first * training_features.features[feature][OT.pValue]
              else
                lazar.fingerprints[t_c_uri][smarts] = 1 * training_features.features[feature][OT.pValue]
              end
              unless lazar.features.include? smarts
                lazar.features << smarts
                lazar.p_values[smarts] = training_features.features[feature][OT.pValue]
                lazar.effects[smarts] = training_features.features[feature][OT.effect]
              end
            end

          # CASE 2: Others
          elsif entry[feature].flatten.size == 1
            lazar.fingerprints[t_c_uri][feature] = entry[feature].flatten.first
            lazar.features << feature unless lazar.features.include? feature
          else
            LOGGER.warn "More than one entry (#{entry[feature].inspect}) for compound #{t_c_uri}, feature #{feature}"
          end
        end
      end
    end

    task.progress 80
    
    # Show compounds without feature information
    if del_master_compounds.size>0
      del_master_compounds.each{|compound| LOGGER.info "Compound: '#{compound.to_s}' not found in feature dataset and will be removed from compound list."}
    end
    # # # Compounds
    lazar.compounds=training_dataset.compounds.collect - del_master_compounds # Add only compounds with fingerprints 
    
    # # # Activities
    if prediction_feature.feature_type == "regression"
      lazar.compounds.each do |compound| 
        entry = training_dataset.data_entries[compound] 
        lazar.activities[compound] = [] unless lazar.activities[compound]
        unless entry[prediction_feature.uri].empty?
          entry[prediction_feature.uri].each do |value|
            lazar.activities[compound] << value
          end
        end
      end
    elsif prediction_feature.feature_type == "classification"
      lazar.compounds.each do |compound| 
        entry = training_dataset.data_entries[compound] 
        lazar.activities[compound] = [] unless lazar.activities[compound]
        unless entry[prediction_feature.uri].empty?
          entry[prediction_feature.uri].each do |value|
            lazar.activities[compound] << lazar.value_map.invert[value] # insert mapped values, not originals
          end
        end
      end
    end
    task.progress 90


    # # # Metadata
    lazar.metadata[DC.title] = "lazar model for #{URI.decode(File.basename(prediction_feature.uri))}"
    lazar.metadata[OT.dependentVariables] = prediction_feature.uri
    lazar.metadata[OT.trainingDataset] = dataset_uri
		lazar.metadata[OT.featureDataset] = feature_dataset_uri
    case prediction_feature.feature_type
    when "classification"
      lazar.metadata[RDF.type] = [OT.Model, OTA.ClassificationLazySingleTarget]
    when "regression"
      lazar.metadata[RDF.type] = [OT.Model, OTA.RegressionLazySingleTarget]
    end

    lazar.metadata[OT.parameters] = [
      {DC.title => "dataset_uri", OT.paramValue => dataset_uri},
      {DC.title => "prediction_feature", OT.paramValue => prediction_feature.uri},
      {DC.title => "feature_generation_uri", OT.paramValue => feature_generation_uri},
      {DC.title => "propositionalized", OT.paramValue => propositionalized},
      {DC.title => "pc_type", OT.paramValue => pc_type},
      {DC.title => "lib", OT.paramValue => lib},
      {DC.title => "nr_hits", OT.paramValue => nr_hits},
      {DC.title => "min_sim", OT.paramValue => min_sim},
      {DC.title => "min_train_performance", OT.paramValue => min_train_performance},

    ]
		
		model_uri = lazar.save(@subjectid)
		LOGGER.info model_uri + " created #{Time.now}"
    model_uri

	end
  response['Content-Type'] = 'text/uri-list' 
  raise OpenTox::ServiceUnavailableError.newtask.uri+"\n" if task.status == "Cancelled"
  halt 202,task.uri
end