This new release includes a new modular design. The following aspects can now be changed:
- learning rate strategy - neighbourhood function strategy - metric used to compute distance In addition, some parts of the code can benefit from parallelisation when training on large datasets.
1.0
This release has been tested on Ubunutu on various datasets with different SOM sizes. Speed improvement goes from roughly 6-7 to almost 30 depending on the SOM, dataset size and number of epochs.