Here is a rough workflow to produce evaluation weights files for Moby Dam

1. Play many matches of Moby Dam against itself and against competitors,
   and save the pdn log files.

2. Convert the collection of pdn files into csv files of positions with their
   pattern indexes, feature values and outcome labels, per game phase:

   rm -f eval-ph?.csv
   train/pdnreader -a eval-ph0.csv -r 38:32 -k *.pdn
   train/pdnreader -a eval-ph1.csv -r 31:24 -k *.pdn
   train/pdnreader -a eval-ph2.csv -r 23:16 -k *.pdn
   train/pdnreader -a eval-ph3.csv -r 15:7 -k *.pdn

3. Per game phase run the Keras/Tensorflow model to create the weights files:

   python train/model.py eval-ph0
   python train/model.py eval-ph1
   python train/model.py eval-ph2
   python train/model.py eval-ph3

4. Optionally, post-process the weights files to make them fully white/black
   symmetric:

   mkdir result
   train/procweights -e -v . -o result
