You are here

Encog 2.5 for Java, .Net and Silverlight Released

Encog 2.5.0 for Java, .Net and Silverlight have been released. Some of the major enhancements in Encog 2.5 are:

  • Support Vector Machines (SVM)
  • Encog Engine and Flat Networks
  • Many performance Improvements
  • OpenCL Training per Device/Concurrent
  • Cross Validation
  • Many Improvements to RBF Networks
  • Additional RBF Functions and SVD Training

The latest version of Encog can be downloaded from the following locations:

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

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

Flat Weight Order in Encog 2.4 vs. 2.5

The order that flat weights are stored in Encog 2.5 has changed slightly from Encog 2.4. Most likely this change will not affect you. If you are simply using NetworkCODEC to keep copies of neural networks in-memory. Or are not even using NetworkCODEC at all, this change will be transparent to you. However, if you are directly manipulating the flat weights, or you are using NetworkCODEC as a means of network persistence, you may encounter some issues.

Structure of Encog Flat Networks

In this article we look at the structure of the Encog Flat Network. The Encog flat network is how Encog represents most neural networks, at the lowest level. This allows Encog to process neural networks with maximum speed. It also keeps the neural networks in a form that they can easily be sent to a GPU.

Categories: 

Source control of choice.

Encog 2.5 Beta for Java & .Net Released

A new beta for Encog 2.5 has been released. This time it is being released for both .Net and Java. This is beta-1 for .Net and beta-2 for Java. Unless something major is found, these will likely be the last betas for 2.5. The official 2.5 will be released sometime around December 1st, 2010. The silverlight version will also be released around that time.

Look for bugs! And report them!

You can download the Java version here:

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

Understanding number formats... How would you write the number one million and one tenth?

Beta Encog OpenCL Benchmark v1.0b1 Released

The first beta version of the Encog OpenCL Benchmark has been released. Please test and let us know how well it performs. This replaces the previous TuneCL application. Using this application you can create a variety of neural network types and see how they would perform on a GPU You can adjust the Encog OpenCL ratios to see what gives the best performance for your size of neural network.

For more information about the application, visit this URL.

http://www.heatonresearch.com/encog/tune1.html

What continent/region are you from?

From 2+2=4 to Advanced Calculus

If you have not seen the Khan Academy yet, you should take a look. They have many videos teaching math. I've been using it to fill some of the holes in my knowledge of Calculus. This site has gotten quite a bit of publicity in the past few months. It is pretty amazing, he literally starts with 2+2=4 and works his way up to advanced mathematics.

Encog 2.5 Beta1 for Java Released

Just released Encog 2.5-beta1 for Java. There have been major changes to this version of Java.

The biggest are:

* Addition of Support Vector Machines (SVM)
* Major rework of OpenCL Support
* Addition of a task scheduler that can train multiple networks at once on CPU and GPU
* Nearly everything uses flat-network structure only, which results in great speed improvements
* Many performance improvements
* Rework of RBF's and addition of the Mexican Hat RBF & other new RBF's
* Common binary format for training data
* Import/Export for Excel

Pages

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer