REQUIREMENTS
------------
- Torch3 (http://www.torch.ch)

INSTALL
------
1) decompress the archive in the Torch3 directory
- cd Torch3;tar zxvf epc.tar.gz
2) add the epc package in the torch packages
- edit Makefile_options_Linux, add epc to packages = 
3) compile again
- make depend;make
4) go to epc examples directory
- cd examples/epc
5) compile the programs
- make measure_verif
- make measure_tc

USAGE
-----
#
# usage: ./Linux_OPT_FLOAT/measure_verif [options] <training score file> <test score file>
#

Arguments:
  <training score file> -> ascii training matrix file (<string>)
  <test score file>     -> ascii test matrix file (<string>)

Options:
  -c <int>              -> column where the score is [0]
  -s <real>             -> step between points for the curves [0.001]

Curves:
  -det_file <string>    -> compute FAR/FRR DET curves []
  -roc                  -> ROC instead of DET 
  -epc_file <string>    -> use the FAR/FRR DCF criterion []
  -criterion <string>   -> frr, far, dcf values [dcf]
  -bound <real>         -> bound for the criterion values [1]

Points:
  -eer_file <string>    -> compute EER points []
  -hter_file <string>   -> compute HTER points []

Stats:
  -no_stats             -> compute far/frr statistics 



  The output files (roc, det, epc) for verification is a simple ascii file of
  six columns: 
    <train-far> <train-frr> <train-hter> <test-far> <test-frr> <test-hter> <criterion>

  Install  the examples package and run try_it scripts.


#
# usage: ./Linux_OPT_FLOAT/measure_tc [options] <training score file> <test score file>
#

Arguments:
  <training score file> -> ascii training matrix file (<string>)
  <test score file>     -> ascii test matrix file (<string>)

Options:
  -c <int>              -> column where the score is [0]
  -s <real>             -> step between points for the curves [0.001]

Curves:
  -F1                   -> compute the F1 measure instead of (P+R)/2 
  -roc_file <string>    -> compute PRECISION/RECALL ROC curves []
  -epc_file <string>    -> use the PRECISION/RECALL DCF criterion []
  -criterion <string>   -> recall, precision, dcf values [dcf]
  -bound <real>         -> bound for the criterion values [1]

Points:
  -bep_file <string>    -> compute BEP points []

Stats:
  -no_stats             -> compute precision/recall statistics 

  
  The output files (roc, epc) for text categorization is a simple ascii file of
  six columns: 
    <train-precision> <train-recall> <1/2(train-precision + train-recall)> 
    <test-precision> <test-recall> <1/2(test-precision + test-recall)> <criterion>

	with "-F1" option:
    <train-precision> <train-recall> <2(train-precision * train-recall)/(train-precision+train-recall> 
    <test-precision> <test-recall> <2(test-precision * test-recall)/(test-precision+test-recall> 

  Install the examples package and run try_it scripts.