Module: OpenTox::Algorithm::FeatureSelection

Includes:

OpenTox::Algorithm

Defined in:

lib/algorithm.rb

Instance Attribute Summary

Attributes included from OpenTox

#metadata, #uri

Class Method Summary

• .rfe(params) ⇒ String

  Recursive Feature Elimination using caret.

Methods included from OpenTox::Algorithm

effect, gauss, get_cdk_descriptors, get_jl_descriptors, get_ob_descriptors, isnull_or_singular?, load_ds_csv, min_frequency, numeric?, pc_descriptors, #run, sum_size, #to_rdfxml, zero_variance?

Methods included from OpenTox

#add_metadata, all, #delete, #initialize, #load_metadata, sign_in, text_to_html, #to_rdfxml

Class Method Details

.rfe(params) ⇒ String

Recursive Feature Elimination using caret

Parameters:

• required (Hash) —

  keys: ds_csv_file, prediction_feature, fds_csv_file (dataset CSV file, prediction feature column name, and feature dataset CSV file), optional: del_missing (delete rows with missing values).

Returns:

• (String) —

  feature dataset CSV file composed of selected features.

# File 'lib/algorithm.rb', line 516

def self.rfe(params)
  @r=RinRuby.new(false,false)
  @r.ds_csv_file = params[:ds_csv_file].to_s
  @r.prediction_feature = params[:prediction_feature].to_s
  @r.fds_csv_file = params[:fds_csv_file].to_s
  @r.del_missing = params[:del_missing] == true ? 1 : 0
  r_result_file = params[:fds_csv_file].sub("rfe_", "rfe_R_")
  @r.f_fds_r = r_result_file.to_s

  # need packs 'randomForest', 'RANN'
  @r.eval <<-EOR
    suppressPackageStartupMessages(library('caret'))
    suppressPackageStartupMessages(library('randomForest'))
    suppressPackageStartupMessages(library('RANN'))
    suppressPackageStartupMessages(library('doMC'))
    registerDoMC()
    set.seed(1)

    acts = read.csv(ds_csv_file, check.names=F)
    feats = read.csv(fds_csv_file, check.names=F)
    ds = merge(acts, feats, by="SMILES") # duplicates features for duplicate SMILES :-)

    features = ds[,(dim(acts)[2]+1):(dim(ds)[2])]
    y = ds[,which(names(ds) == prediction_feature)] 

    # assumes a data matrix 'features' and a vector 'y' of target values
    row.names(features)=NULL

    # features with all values missing removed
    na_col = names ( which ( apply ( features, 2, function(x) all ( is.na ( x ) ) ) ) )
    features = features[,!names(features) %in% na_col]

    # features with infinite values removed
    inf_col = names ( which ( apply ( features, 2, function(x) any ( is.infinite ( x ) ) ) ) )
    features = features[,!names(features) %in% inf_col]

    # features with zero variance removed
    zero_var = names ( which ( apply ( features, 2, function(x) var(x, na.rm=T) ) == 0 ) )
    features = features[,!names(features) %in% zero_var]

    pp = NULL
    if (del_missing) {
      # needed if rows should be removed
      na_ids = apply ( features,1,function(x) any ( is.na ( x ) ) )
      features = features[!na_ids,]
      y = y[!na_ids]
      pp = preProcess(features, method=c("scale", "center"))
    } else {
      # Use imputation if NA's random (only then!)
      pp = preProcess(features, method=c("scale", "center", "knnImpute"))
    }
    features = predict(pp, features)

    # features with nan values removed (sometimes preProcess return NaN values)
    nan_col = names ( which ( apply ( features, 2, function(x) any ( is.nan ( x ) ) ) ) )
    features = features[,!names(features) %in% nan_col]

    # determine subsets
    subsets = dim(features)[2]*c(0.3, 0.32, 0.34, 0.36, 0.38, 0.4, 0.42, 0.44, 0.46, 0.48, 0.5, 0.52, 0.54, 0.56, 0.58, 0.6, 0.62, 0.64, 0.66, 0.68, 0.7)
    #subsets = dim(features)[2]*c(0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7)
    #subsets = c(2,3,4,5,7,10,subsets)
    #subsets = c(2,3,4,5,7,10,13,16,19,22,25,28,30)
    subsets = unique(sort(round(subsets))) 
    subsets = subsets[subsets<=dim(features)[2]]
    subsets = subsets[subsets>1] 

    # Recursive feature elimination
    rfProfile = rfe( x=features, y=y, rfeControl=rfeControl(functions=rfFuncs, number=150), sizes=subsets)

    # read existing dataset and select most useful features
    csv=feats[,c("SMILES", rfProfile$optVariables)]
    write.csv(x=csv,file=f_fds_r, row.names=F, quote=F, na='')
  EOR
  r_result_file
end