Skip to content

Latest commit

 

History

History
22 lines (16 loc) · 1.8 KB

README.md

File metadata and controls

22 lines (16 loc) · 1.8 KB
Raw

Example graphical outputs

Filter graph

Configuration

  • Training data: NeedleGenerator
  • Number of filter coefficients: 2

Description

The graphical output in this folder is produced using Example03, with input from Example02. These plots should give an indication of the wavenet training that the package facilitates.

  • ./exampleInputs/: Folder containing the 10 training examples used in generating the cost map shown in FilterGraph.pdf.
  • CostGraph.pdf: Graphs of the combined (regularisation and sparsity) cost as functions of the number of training updates, one for each random initialisation.
  • FilterGraph.pdf: The evolution of the filter coefficient configurations, one for each random initialisation, overlayed on the cost map in two-dimensional filter coefficient space. Red markers are initial configurations (generated on the unit circle); blue markers are final configurations. Contours indicate regions of similar cost, computed for the inputs in ./exampleInputs/.
  • FilterGraph.png: Same as FilterGraph.pdf, but in different format.
  • Orthonormality.pdf: Distributions of inner products for all basis functions in the best, final basis1. Should be peaked around 0 and 1.
  • bestBasis_1D.mov: Movie showing the evolution of the best basis1 in one dimension.
  • bestBasis_2D.mov: Movie showing the evolution of the best basis1 in two dimensions.

1 By "best basis" we mean the basis, or initialisation i.e. evolution of bases, which has the lowest cost at the end of the training.