Preface

This document is for developers who want to extend the existing node functionality. If you need a custom sensor or effector to accept or to write different data than the default sensors and effectors, or if you want to modify the existing learning algorithm, you can write a node plugin.

You can write plugins in either C++ or Python. If you process them appropriately, they seamlessly integrate with NuPIC at runtime.

Topics

Scope of Document
Related Documentation
Document History
For More Information

Scope of Document

This document is meant for developers who want to create custom nodes for their NuPIC application. In many cases, the functionality of the nodes included with NuPIC works well. However, some developers need custom sensors or effectors or might want to modify the learning algorithm. This document discusses how to do this.

To create a custom node, you should be an experienced C++ or Python developer.

Document Overview

The document is useful regardless of whether you're developing a C++ plugin or a Python plugin.

Chapter 1, Getting Started With Numenta Node Plugins explains the process of plugin creation. It discusses the NodeSpec and also discusses the interaction between custom nodes and the NRE or Tools Framework. This chapter is relevant for both C++ and Python developers.
The next two chapters are mainly for C++ plugin developers:
Chapter 2, C++ Custom Node Quickstart, explains how you can use the quickstart.py script to create the set of files required for a C++ plugin. The files include boilerplate code, which is discussed as well.
Chapter 3, C++ Node Plugin API Reference, discusses the interfaces, classes, and methods on which you build your plugin. It also includes a discussion of the plugin node life cycle. This chapter is meant to help with an in-depth understanding of the plugin system.
Chapter 4, Using the Python Node Plugin API, is for Python developers. It gives an overview of the development process and briefly discusses the example included with the source.

Related Documentation

The Getting Started with NuPIC and Advanced NuPIC Programming manuals are for developers who want to implement an HTM Network. In addition, the following white papers are for developers who want background information on HTMs and for Zeta1Node.

The HTM Learning Algorithms is an in-depth discussion of how HTM Networks work.
Zeta1 Algorithms Reference discusses the learning algorithm implemented by the NuPIC Zeta1Node in detail. This includes discussion of all parameters.

Conventions

This document uses the following conventions:

All filenames, code examples, and names of code elements are displayed in code font.
Document names are given in italic font.

This document uses the following icons:

Table 1: Icons used in this document

Icon	Description
	Note. A noteworthy item. If you do not pay attention to a note, nothing bad is likely to happen.
	Warning. If you do not pay attention to a warning, data loss or other problems may result.
	Tip. Paying attention to a tip can make it easier and faster to use Numenta software.

Document History

This section lists changes made in the specified version of the document. Note that document version numbers do not always correspond to release numbers.

Release	Description
1.0 March 2007	First release (DRAFT).
1.0.1 March 15 2007	Minor bug fixes including update of white paper name, document name (Node Plugin Developer's Guide), and similar issues. Added Document History.
1.1 May 2007	First non-draft release. Added initial NodeSpec information. Added Python plugin information. Removed Zeta1Node walkthrough because it caused too much duplication.
1.1.1 August 2007	Minor documentation bug fixes.
1.2.0 January 2008	Added Windows build/install procedure. Minor bug fixes.
1.2.1 June 2008	Minor updates and corrections. Some additions to PyNode chapter.

For More Information

The Numenta website includes a variety of educational materials and forums to help you find answers to your questions. See http://numenta.com/for-developers.ph.