The Core Engine

The core engine is the fulcrum around which are built all the other modules and applications based on Joone. It exposes a Java API providing the developer with the following features:

• Components to create any neural net architecture (feed forward or recurrent)
• Several supervised training algorithms:
• On-line backprop
• Batch backprop
• Resilient Backprop (RPROP)
• Components to build unsupervised neural networks (Kohonen SOMs and Principal Component Analysis)
• Components to build modular neural networks
• A serialization mechanism to save and restore a nnet to/from a file system or to send/receive a nnet on remote locations (via HTTP, RMI, etc.)
• I/O components to read patterns from:
• Ascii comma delimited files from file system or HTTP
• Excel files
• Image (gif & jpeg) files
• RDBMS using JDBC
• Stock price time series from Yahoo Finance
• Components to implement both the supervised and unsupervised learning control
• Components to control the behavior of the neural net (start/stop, recall/learn) and its parameters (learning rate, momentum, etc.)
• Plug-ins to preprocess the input data (Normalization, Moving Average, etc.) and to control dynamically the training parameters (Annealing)
• A complete event notification mechanism, along with a scripting capability to control the behaviour of a neural network at the happening of some events

The engine is composed by a set of components subdivided into the following packages:

• org.joone.engine

  All the classes of the core engine wich represent the bricks to build the neural nets

• org.joone.engine.learning

  All the components to train the neural net

• org.joone.io

  All the I/O components to read and write patterns from/to external sources

• org.joone.net

  The shell components of the neural net. They are useful to manage a neural net as a indivisible entity

• org.joone.util

  Some utility classes to perform several tasks (input normalization, input scaling, scripting, etc.)

• org.joone.log

  The classes to write the output messages to a log file (it can use either Log4J or an internal logger)

• org.joone.script

  The classes to define and execute BeanShell scripts (called also Macro)

• org.joone.helpers

  Helper classes to make it very simple to build, train and test a neural network

• The Joone Complete Guide (pdf)
• The Java API cookbook
• The Netbeans Homepage (We use it to develop Joone)
• The JOONE GUI Editor
• A sample GUI, that learns the XOR problem
• A selection of the best Neural Networks books