General Regression Neural Networks: Training, Inference, and File Persistence

Summary

The document outlines a class for building a generalized regression neural network. Its constructor sets input and output vector sizes, while a learning method accepts training patterns arranged in flat input and output arrays. Training stops when it reaches either a specified epoch limit or an allowed error. The class exposes the resulting mean squared error and completed epoch count, and reports a memory error if it cannot allocate enough space.

A calculation method produces a network response for new input data. Save and load methods store or restore the network topology, training error, and weights in a binary file; loading fails when the saved and configured topologies differ. Included examples apply the class to XOR and integer multiplication and addition. These demonstrate implementation and basic function learning, but the document gives no financial forecasting application, benchmark, generalization analysis, or guidance on selecting training parameters.

Key ideas

• The network takes input and output vector sizes when it is created.
• Training uses paired patterns and stops at an epoch limit or an error threshold.
• The class exposes mean squared error and the number of completed training cycles.
• Saved models include topology and weights, and loading requires a matching topology.
• The examples demonstrate basic functions rather than trading applications.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.