Encog Version History
Past Encog Versions
- Version 0.5: (July 10, 2008) Initial version, code exactly the same as the code provided in the two books.
- Version 1.0: (Sept 15, 2008) Unified neural network classes so that Hopfield, SOM and feedforward are all intermixable as layer types.
- Version 1.1: (Dec 12,2008) Enhancements necessary to support a GUI workbench.
- Version 2.0: (July 2009) Layers will be able to have more than one child, enhancements to Workbench to support this. Adding recurrent layers and RPROP.
- Version 2.1: (Aug 2009) Support for many additional network types, such as BAM, CPN, ART1, Boltzmann and more.
- Version 2.2: (Nov 2009) Focus on performance and enhancing multicore support. Adding normalization.
- Version 2.3: (Feb 2010) Core: MPROP/RPROP save progress. SOM enhanced, SCG Training. Resume propagation training with no loss of work. Performance improvements.
- Version 2.4: (June 2010) Levenberg Marquardt Training, NEAT & GPU Acceleration (OpenCL)
- Version 2.5: (Nov 2010) Support Vector Machines (SVM), Encog Engine and Flat Networks, Many performance improvements, Cross Validation, Improved RBF.
- Version 3.0: (July 2011) Much greater focus towards additional Machine Learning methods. Focus on prediction, markets and indicators. Some NinjaTrader integration. PNN/GRNN. Encog analyst.
- Version 3.1: (May 2012) Bayesian Networks, Hidden Markov Machines, Nelder Mead Training
- Version 3.2: (Jan 2014) Genetic Programming, much improved NEAT/HyperNEAT
Current Encog Version
- Version 3.3: I do introduce EncogModel. This will eventually replace the current Encog Analyst classes.
Next Encog Version
- Version 3.4: Under development, some deel learning elements: ReLU, Stochastic Gradient Descent, L1&L2 Regularization, Dropout