# This is the "T-C" problem introduced on page 348 of 
# Parallel Distributed Processing, Explorations in the Microstructure of 
# Cognition, Vol. 1, edited by David Rumelhart and James McClelland.
#
# The network must learn to recognize a T or a C at whatever orientation or
# location they appear in the input.
# You may be able to learn the training examples, but will it generalize to
# the held-out testing examples?

addNet T-C 81 60 1
setObj input.numColumns  9
setObj numUpdates 1000
setObj batchSize  50

if [catch {path -s train}] {
  loadExamples t-c.ex -s train -e PERMUTED
  moveExamples train test 0.026
}
autoPlot
viewUnits