How to get the attribute involved into the classification of an ADTree

Question

My problem is the following :

I'm performing classification using weka's ADTree. I build a classifier on a dataset that have over 1700 attributes. The resulting ADTree only use a very little subset of them to classify the instances (near 10 attribute are used).

My question is, as the attribute computation for my instances is time consuming, could I retrieve the attribute's identifiers used by the ADTree? I aim to compute only the relevant attribute letting the other to a default value in order to avoid very long and useless computing.

Thanks in advance.

Answer 1

Finally find a solution. Had to extend the ADTree to achieve my goal

public class ExtendedADTree extends ADTree {

    private static final long   serialVersionUID    = 1L;

    /**
     * @param dataset
     *        The dataset to use to retrieve attributes labels.
     *
     * @return Returns the list of label used to predict the class of an instance knowing the
     *         dataset used to compute the tree.
     */

    public List<String> getPredictionNodeLabels(final Instances dataset) {
        final List<String> result = new LinkedList<>();

        // Initialize the list of splitter node to explore.
        final List<Splitter> nodesToExplore = new LinkedList<>();
        @SuppressWarnings("unchecked")
        final Enumeration<Splitter> rootChildrens = this.m_root.children();
        Collections.list(rootChildrens).forEach(child -> nodesToExplore.add(child));

        while (!(nodesToExplore.isEmpty())) {
            // while there is node to explore get the splitter childrens of the current node and add
            // them to the queue
            final Splitter currentNode = nodesToExplore.remove(0);

            // add the label of the splitter node to the result
            result.add(currentNode.attributeString(dataset));

            // add the childrens to the nodesToexplore list.
            for (int branch_number = 0; branch_number < currentNode.getNumOfBranches(); branch_number++) {
                final PredictionNode child = currentNode.getChildForBranch(branch_number);
                final Enumeration<Splitter> childChildren = child.children();
                Collections.list(childChildren).forEach(childOfChild -> nodesToExplore.add(childOfChild));
            }
        }

        return result;
    }
}

This method return the list of the attributes' label of decision nodes.