This version is usable.
This is the modern implementation of MoRPE. My plan is to keep the C# version in sync with other modern versions.
- More testing
- We currently have minimal testing which just proves basic overall functionality of the training algorithm.
- We need unit tests for all the encapsulated "units" of functionality.
- We need many more scenarios for classifier training.
- Visualization
- The goal is to generate "subspace" plots, similar to figures 7, 8, 9, and 10 from the paper Ing_Morpe.docx.
- The plots would be SVG, but would incorporate a bitmapped graphics file for each 2D cloud.
- Multiple classifier management
- We would expand the
TrainingContext
class to be a container for many classifier variants.- All 1D classifiers (1 per dimension)
- All 2D classifiers (1 per dimension pair)
- All full-dimensionality classifiers.
- Classifiers of various polynomial rank.
- This supports two kinds of operations.
- The cross-validation procedure, used to choose a best-fitting classifier.
- The generation of model output for the "subspace" plots.
- We would expand the
- Model selection via cross-validation
- Code to train multiple classifier variants, to pick the best variant.