First thing pick, according to the hardware and libs you have on your system,
one of the Docker Torch RNN images
by writing one of the corresponding stirngs crisbal/torch-rnn:base,
crisbal/torch-rnn:cuda6.5, or crisbal/torch-rnn:cuda7.5 in the file named
image.conf in this directory.
Then, put an ASCII STL model in the data directory; assume it is named
pot.stl first run the preprocessing as
./preprocessing.sh data/pot.stl
at the end, you should find a pot.h5 and pot.json in the data dir; now run
the training as
./train data/pot.stl 1000
where 1000 is the number of iterations among checkpoints; while the
computation runs, you should see some file named pot-checkpoint_NNNN.t7 in the
data dir. As soon as you have one of such chekpoint you and sample the
genrator as
./sample.sh data/pot.stl 3000 20000
where 3000 is the checkpoint that you want to use (meaning that
pot-checkpoint_3000.t7 must be present in data dir) and 20000 is the
length of the file you want to generate.
If you have a RAW model, instead, just replace pot.raw in place of pot.stl
in all the above commands.