An Example Data Analysis Workflow
The KNIME workbench is shown above with what might be a typical data analysis workflow. Workflows are developed interactively in KNIME and are generally built incrementally rather than all at once. Prior to submitting data to the J48 and Decision Tree Prediction nodes, the author of this workflow probably first chose to visually inspect the training data with the Interactive Table, Scatter Plot, and Parallel Coordinates nodes. The modeling and prediction part of this workflow was probably added after the visual inspection helped encourage the author to believe that such modeling could bear fruit. As seen in the next screenshot, these visualization nodes prove useful again after executing the modeling and prediction part of this workflow.
An Example Data Analysis Workflow: Views
After executing the workflow, the visualization nodes are used again to open views for visualizing the training data alongside the new "test data" fed to the modeling and prediction nodes for classification.
An Example Data Analysis Workflow: Hiliting
To help with analyzing results from the model developed and shown in the prior screenshots, here the workflow's author can see how different classifications were devised in the model by hiliting (selecting) rows in a table view and seeing the corresponding data points in other visualizations instantly pop out (hilite). This hiliting functionality (always referred to in KNIME as "hiliting") works between views (visualization windows) to trace subsets of the data.
Extending KNIME with New Data Types
In addition to common, standard data types such as strings, integers, or double values, KNIME can be extended to understand complex data types like those used to represent molecules (used in the chemical and pharmaceutical industries). This screenshot indicates how, as part of its understanding of molecular data types, KNIME has the ability to graphically depict these molecules (a far user-friendlier way of representing the molecules than merely showing the raw data defining a molecular structure). KNIME can be taught about many other complex data types and can even be taught how best to graphically represent that data to the user.
Meta Nodes: Turning a Workflow into a Reusable Node
Meta nodes encapsulate workflows into a single node, which can then be used over and over again in other workflows. In this screenshot, the workflow in the top half of the screen contains a meta node labeled as MetaNode 1:1. Inside this meta node is an entire workflow, seen in the bottom half of the screen. Thus, new types of nodes can be authored once and saved for future repeated use via the creation of meta nodes, all without needing to write a single line of code.
Integration of 3rd Party Packages
KNIME has integrated R, Weka and the Chemistry Development Kit (CDK), enabling the use of functionality from these 3rd party packages through a collection of KNIME nodes. This screenshot depicts the use of views from R nodes, Weka nodes, and CDK nodes made available from this integration.
KNIME's Extension Wizard
KNIME's functionality is highly extensible and enables the creation of new, custom nodes that can be seamlessly imported into a KNIME installation. As seen in this screenshot, the extension wizard facilitates the development of new nodes.