djc_local

数据挖掘中的一算法 ines算法 c下实现的。适合初学习数据挖掘者借鉴

#!/bin/sh
#-----------------------------------------------------------------------
# File    : djc_local
# Contents: probabilistic network induction with local structure
#           on artificial Danish Jersey Cattle data
# Author  : Christian Borgelt
# History : 17.12.1999 file created
#           17.07.2002 all induction loops moved into functions
#           23.07.2002 bdeu handling moved to function induce
#-----------------------------------------------------------------------

function generate ()
{                               # --- generate random databases
  for (( i = 0; i < 10; i++ )); do
    gendb -s$(( $1*(i+1) )) djc.net train$i.tab 2> /dev/null
    gendb -s$(( $2*(i+1) )) djc.net test$i.tab  2> /dev/null
  done
}  # generate()

#-----------------------------------------------------------------------

function collect ()
{                               # --- collect evaluation results
gawk '
/evaluation of/         { network = $3  }
/number of attributes/  { attcnt  = $NF }
/number of conditions/  { concnt  = $NF }
/number of parameters/  { parcnt  = $NF }
/number of tuples/      { tplcnt  = $NF }
/impossible tuples/     { imptpl  = $4  }
/minimum/               { minimum = $NF }
/average/               { average = $NF }
/maximum/               { maximum = $NF }
/additional conditions/ { addcnt  = $NF }
/missing    conditions/ { miscnt  = $NF }
END {
  printf("%-12s", network);
  printf(" %3d %3d %3d", concnt, addcnt, miscnt);
  printf(" %5d %10g", parcnt, average);
}'
}  # collect()

#-----------------------------------------------------------------------

function average ()
{                               # --- average evaluation results
gawk '
function output() {
  if (NR > 0) {
    printf("%-10s", network);
    printf(" %6.1f %6.1f %6.1f", concnt/n, addcnt/n, miscnt/n);
    printf(" %7.1f %10.1f %10.1f\n", parcnt/n, train/n, test/n);
  }
}
BEGIN { network = ""; }
($1 == network) {
  concnt += $2; addcnt += $3; miscnt += $4;
  parcnt += $5; train  += $6; test   += $7; n++;
}
($1 != network) {
  if (n > 0) output();
  network = $1; n = 1;
  concnt  = $2; addcnt  = $3; miscnt  = $4;
  parcnt  = $5; train   = $6; test    = $7;
}
END { if (n > 0) output(); }' local.tmp
}  # average()

#-----------------------------------------------------------------------

function evaluate ()
{                               # --- evaluate a given network
  neval -L1 -c djc.net $1 train$2.tab 2> /dev/null | \
    collect >> local.tmp
  neval -L1            $1 test$2.tab  2> /dev/null | \
    gawk '/average/ { printf(" %10.1f\n", $NF); }' >> local.tmp
  rm -f $1
}  # evaluate()

#-----------------------------------------------------------------------

function induce ()
{                               # --- induce and evaluate networks
  rm -f local.tmp
  if [[ $2 == bdeu ]]; then m="bdm" x="-p-20";
                       else m=$2    x=""; fi
  for (( i = 0; i < 10; i++ )); do
    ines -s$1 -e$m djc.dom train$i.tab $2 $x $3 2> /dev/null
    evaluate $2 $i
  done
  average | tee -a local.res
}  # induce()

#-----------------------------------------------------------------------

function indlocal ()
{                               # --- induce and evaluate networks
  rm -f local.tmp
  if [[ $2 == bdeu ]]; then m="bdm" x="-p-20";
                       else m=$2    x=""; fi
  for (( i = 0; i < 10; i++ )); do
    ines -c2 -s$1 -e$m djc.dom train$i.tab $2 $x $3 2> /dev/null
    ines -c0 -s$1 -e$m -g-1e-12 $2 train$i.tab $2 $x $3 2> /dev/null
    evaluate $2 $i
  done
  average | tee -a local.res
}  # indlocal()

#-----------------------------------------------------------------------

function fixed ()
{                               # --- evaluate empty/original network
  if [[ $1 == indep ]]; then in="djc.dom";
                        else in="djc.net"; fi
  rm -f local.tmp
  for (( i = 0; i < 10; i++ )); do
    ines $in train$i.tab $1 2> /dev/null
    evaluate $1 $i
  done
  average | tee -a local.res
}  # fixed()

#-----------------------------------------------------------------------
 
function topord ()
{                               # --- greedy condition selection
  echo "---topord----------------------------------------------------" \
       | tee -a local.res
  for m in infgain infgr infsgr1 gini chi2 bdm bdeu rdlrel; do
    induce topord $m
  done
}  # topord()

#-----------------------------------------------------------------------
 
function local ()
{                               # --- local structure learning
  echo "---local-----------------------------------------------------" \
       | tee -a local.res
  for m in infgain infgr infsgr1 gini chi2 bdm bdeu rdlrel; do
    induce topord $m -g-1e-12
  done
}  # local()

#-----------------------------------------------------------------------
 
function lwise ()
{                               # --- levelwise local structure learning
  echo "---lwise-----------------------------------------------------" \
       | tee -a local.res
  for m in infgain infgr infsgr1 gini chi2 bdm bdeu rdlrel; do
    induce topord $m -G-1e-12
  done
}  # lwise()

#-----------------------------------------------------------------------
 
function post ()
{                               # --- local structure learning
  echo "---post------------------------------------------------------" \
       | tee -a local.res
  for m in infgain infgr infsgr1 gini chi2 bdm bdeu rdlrel; do
    indlocal topord $m
  done
}  # post()

#-----------------------------------------------------------------------

function cleanup ()
{                               # --- clean up temporary files
  rm -f local.tmp
  rm -f train[0-9].tab
  rm -f test[0-9].tab
}  # cleanup()

#-----------------------------------------------------------------------

echo "network      cond    add   miss  params      train       test" \
     | tee local.res
echo "-------------------------------------------------------------" \
     | tee -a local.res

generate 13 17          # generate random databases
fixed indep             # evaluate empty    network
fixed orig              # evaluate original network
topord                  # condition selection on topological order
local                   # local structure learning (unrestricted)
#lwise                   # local structure learning (levelwise)
#post                    # local structure learning (preserving global)
cleanup                 # clean up temporary files