Skip to main content

Table 4 Parameters used with the dropout regularized combination classifier for the lymphoma application

From: Robust identification of molecular phenotypes using semi-supervised learning

Method

Parameter

Value(s)

DRC classifier (applied to the lymphoma data set)

k (kNN sub-classifiers)

9

Subsets of features used in the sub-classifiers

Singles

Sub-classifier filtering criteria

OS HR between the two classification groups

Sub-classifier filtering range applied to the training set

[0.0; 100.0] (no filtering), [1.3; 100.0] (intermediate filtering) and [2.0; 100.0] (strong filtering)

Number of dropout iterations (in the boosting step)

100,000

Number of sub-classifiers kept in each dropout iteration

10

Number of training / test realizations

375

Number of samples included in the training subset, for each class

\( 2/3\times {N}_{\mathsf{S}} \), where \( {N}_{\mathsf{S}} \) is the number of samples in the smaller class. Remainder samples assigned to the test subset

Maximum number of refinement iterations

8