#! /bin/bash
# by Dan Baker, WVU, 2006
# This script creates a range of estimates using the cocomost learner.
# This is meant to be used on a single test record to create
# an actual prediction: i.e. it's not meant for an experiment (no actual, mre's listed)
#
# The input must be all numeric without any comments, like the output of ../baker_lc/convertcsv.awk
#
# The output is in the form #Estimate,a,b,CostDrivers,Seed
# There will N+1 estimates made. There are N estimates using X% of Train, and one estimate that uses
# all of Train and can be noted by a NA value in the Seed column.

#Set up variables
Train=$1
Test=$2
vTrain="/tmp/"$USER"/vtrain.tmp"
echo -n "" > $vTrain
N=10
X=0.9
XNum=$(gawk  -v X=$X 'END{print int((NR*X)+0.5)}' $Train)

#Assumes only one test record
TestCount=$(gawk 'END{print NR}' $Test)
if ((TestCount>1))
then
  echo "Error: This script is meant for only one test record."
  exit
fi

{
#Header Line
echo "#Estimate,a,b,CostDrivers,Seed"

#Run cocomost with all the training data
CostDrivers=$(./cocomostFSS $Train)
Results=$(gawk -f ../baker_lc/lc.awk Pass=1 TargetEMs=$CostDrivers $Train Pass=2 Mode=5 TargetEMs=$CostDrivers $Test)
echo $Results",NA"

for ((I=1;I<=$N;I++))
do
  # Create a random X% of the training data
  Seed=$RANDOM
  gawk -v Seed=$Seed -v TFile=$vTrain -v Xn=$XNum 'BEGIN{CONVFMT="%.20g"}NR==1{srand(Seed);}{Dat[rand()]=$0}END{c=0;for(I in Dat){if (c<Xn){print Dat[I]>>TFile}c++}}' $Train

  CostDrivers=$(./cocomostFSS $vTrain)
  Results=$(gawk -f ../baker_lc/lc.awk Pass=1 TargetEMs=$CostDrivers $vTrain Pass=2 Mode=5 TargetEMs=$CostDrivers $Test)
  echo $Results","$Seed
done
} | sort | malign