This is the beginning of several articles highlighting the architectural changes that will be occurring in Encog 3.0. Encog 3.0 is an upcoming release of Encog that will address several architectural issues that have crept into Encog as it has advanced. I am attempting to move Encog more in the direction of general machine learning, supporting much more than just neural networks.

Very often, I've been asked where is the best place to begin to learn neural networks. Especially, if the reader has no experience at all with them. I've written several very introductory articles. I would suggest you start with this article. It is the most introductory piece I've written on neural networks.

http://www.heatonresearch.com/content/non-mathematical-introduction-usin...

Online Videos

Then you might also try some of the intro videos. These videos are designed to show you how a neural network actually works, internally.

The TA-Lib DLL is a framework for technical analysis. This DLL contains over 200 technical indicators, such as ADX, MACD, RSI, Stochastic, and Bollinger Bands. It was originally written in C, but has been translated to 100% native Java and C#. TA-Lib contains little documentation, other than for the C implementation. If you are familiar with the C programming language, it is not too much of a stretch to get TA-Lib working for you. However, this article is meant to show how to create a simple moving average (SMA) with TA-Lib.

The TA-Lib JAR is a framework for technical analysis. This JAR contains over 200 technical indicators, such as ADX, MACD, RSI, Stochastic, and Bollinger Bands. It was originally written in C, but has been translated to 100% native Java and C#. TA-Lib contains little documentation, other than for the C implementation. If you are familiar with the C programming language, it is not too much of a stretch to get TA-Lib working for you. However, this article is meant to show how to create a simple moving average (SMA) with TA-Lib.

Neural networks and other machine learning constructs, such as support vector machines, want input data to be in the range between 0 to 1 or -1 to 1. This requires the data to be normalized. Normalization is a fairly simple mathematical process. You can also de-normalize data that the neural network outputs to get it into a usable form.

The following code shows how to normalize and de-normalize. I've added a fair amount of comments to this example, so I will let the comments speak for themselves. Basically we normalize 15 to 0.5 and then de-normalize 0.5 back to 15.

Encog v2.5.2 for Java has been released. This is a bug fix release. There will be a Encog v2.5.2 for .Net released soon. You can download from:

http://code.google.com/p/encog-java/downloads/list

Some of the bug fixes were:

* Fixes to the Maven POM
* Fixed array handling issue in generic persistor that caused issues with NEAT
* Fixed issue with NEAT genome handler

This version of Encog was also published to Maven Central. All future Encog Java releases will be pushed to Maven. Fore information about Encog and Maven, see: