Biomedical Engineering Reference
In-Depth Information
in which the repeated execution of parts of the workfl ow with changed
parameters is desirable. This can range from simple iterations over several
input fi les, to cross-validation where a model is repeatedly trained and
evaluated with different distinct parts of data and can include even more
complex tasks such as feature elimination. In order to be able to model
such scenarios in KNIME, two special node types are available: loop start-
and loop end-nodes. In contrast to normal nodes (inside the loop), they
are not reset while the loop executes, both nodes have access to its
counterpart, and they can directly exchange information. For example,
the loop end-node can tell the start-node which column it should remove
at the next iteration or the start-node can tell the end-node whether or not
the current iteration is last. Figure 6.3 shows a feature elimination loop in
which the start- and end-nodes are visually distinguishable from normal
nodes. The feature elimination model can subsequently be used by the
feature elimination fi lter to remove attributes from the data table.
The node repository contains several predefi ned loops encapsulated in
meta-nodes, such as a simple 'for' loop, which is executed a certain
number of times, cross validation, iterative feature elimination, or looping
over a list of fi les.
6.1.4 Extensibility
KNIME is an Eclipse-based application, which has two major implications.
First, it is written in Java meaning that it runs on all major operating
systems (Windows, Linux, MacOS X). Second, the Eclipse platform was
designed to be highly extensible - as is KNIME. Developers can easily
Feature elimination is available as a loop inside a
meta-node
Figure 6.3
Search WWH ::
Custom Search