In what follows, we will first describe the training phase
of the predictor network 
(a strictly layered feed-forward net trained by back-propagation
[10][4]).
The training phase
is based on a set 
of training files.
Then we will describe three
working off-line variants of ``compress'' and
``uncompress'' functions based on .
All methods are guaranteed to encode and decode arbitrary
unknown text files without loss of information.