Training
- We do a 10-fold test, training on 90% of the data and testing
on 10%.
- We train for 20000 cycles or 266.67 epochs. There are 1335 training
items, so not all are presented.
- We then turn learning off, run all 1335 items through the
system and record the SOM subnets firing behaviour.
- We then present the test set (without the categorisation feature),
and record its firing behaviour.
- There are 1000 neurons in the SOM subnet, so for each
item we have a 1000 dimensional vector.
- For each test item, we do a Pearson's Product Moment Correlation
with all the training items.
- The test item is categorised as the nearest via Pearson.
