What is HierDAG library?

HierDAG is an R software library implementing two hierarchical ensemble methods for Directed Acyclic Graphs (DAGs):
1. Hierarchical Top-Down for DAG (HTD-DAG);
2. True-Path-Rule for DAG (TPR-DAG);

Files contained in the HierDAG library: brief description

1. Functions implementing Hierarchical Ensemble Methods for DAGs:

   • htd.R: implementation of the HTD algorithm;
   • tpr.R: implementation of the TPR algorithm and its variants;
   • Do.HTD.R: high level functions to compute the HTD-DAG algorithm;
   • Do.TPR.R: high level functions to compute TPR-DAG algorithm and its variants;

2. Utility Functions:

   • graph.utils.R: utility functions to process and analyze a graph;
   • IOgraph.R: IO functions to store and build graph both in plain text and in rda compressed format;
   • flat.score.norm.R: function to normalize the flat scores according to the maximum score of each class;
   • Do.full.annotations.table.R: high level functions to compute the full annotation table;
   • Do.best.F.score: high level functions to select the best F-score by choosing an appropriate threshold in a scores matrix;
   • Do.FLAT.scores.normalization: high level functions to normalize flat scores matrix w.r.t. MaxNorm or Qnorm;
   • F.hier.R: function to compute precision, recall, F-measure, specificity and accuracy for multi-class multi-label classification task (Kiritchenko-like multi-label F-scores)
   • AUPROC.R: function to compute AUROC and AUPRC through the R package precrec

Loading

Even if HTD-DAG and TPR-DAG algorithm depend on several other source codes, loading them in the R environment is straightforward. For instance to load the high-level function computing the HTD-DAG algorithm, just open the R environment in the same folder where you download the HierDAG library and type:

source("Do.HTD.R");

NB: to rightly load the HierDAG library, the following R libraries are required:

library("graph");
library("RBGL");
library("Rgraphviz");
library("PerfMeas"); 		
library("precrec");			
library("preprocessCore");