readme

C-package of "Long Short-Term Memory" for Protein classification

To run the complete SCOP 1.53 benchmark please unpack the datasetsSCOP1.53.tar.bz2
archive in this directory so that a directory 'datasetsSCOP1.53' exists with the subfolders
'original-dataset' and 'positive-training-extended'. See also the 'runtrain.sh' script
where the location of the datasets are configured.
If not happened before the LSTM binary should be compiled by running 'make' in the
src directory. Then the experiment is started by running

sh runtrain.sh

The LSTM gives out the performance on every epoch and stops after 100 epochs which
is repeated for the 54 datasets.
The performance metrics are Mean Square Error (mse), false negatives (fn) and
false postives (fp) in percent. The progress is also written in to the file 'out.txt' in
the lstmrun directory. This file ist automatically stored with the name of the test familiy
in the directory 'out'. The weight matrices can be found in the directory wms.