README

Text-or-data detector

Neural Networks first experiments by Sergey Vasilyev. July 1997.

The neural network is trained based on the share of symbols in a text (0..100%). Probably, only ASCII files are supported, as it was developed in the assumption that all letters are encoded as 0..255 (back then in 1997).

Build

On Mac OS X, Free Pascal is needed:

brew install fpc

Then build both files:

fpc -Mtp -WM10.9 istext.pas
fpc -Mtp -WM10.9 train.pas

Check that they work:

$ ./train
Usage: Train.exe <text> <weights> <Y|N>

$ ./istext 
Usage: Test.exe <text> <weights>

Train

Feed the text/data files into the network:

./train SOMEFILE.txt expr1 Y
./train SOMEFILE.bin expr1 N

Repeat as much as needed.

Detect

./istext ANOTHERFILE.txt expr1
./istext ANOTHERFILE.bin expr1

There is a pretrained neural network in a file called garry — it can be used instead.

./istext ANOTHERFILE.txt garry
./istext ANOTHERFILE.bin garry