#! /bin/bash

#this file runs competing learners on the given subset of data

Path=$1
Filename=$2

#SeedFile=$Path"/"$Filename".seed"
ConvertedFile=$Path"/data/"$Filename".numeric"
Iterations=30
#Iterations=100
TestSets=1
DoCocomostV2=1
DoCocomostV3=0
Experts=4
#M="0.8"
Mprime="0.2"

#Run the algorithm many times
for ((i=1;i<=$Iterations;i++))
do
	#Randomize Data
	TestFile=$Path"/data/test/"$Filename".numeric."$i
	TrainFile=$Path"/data/train/"$Filename".numeric."$i
	Seed=$RANDOM

	gawk -f randomize.awk -v Seed=$Seed -v TestSets=$TestSets -v TrainFile=$TrainFile -v TestFile=$TestFile $ConvertedFile
	
	# basic Local Calibration
	LogFile=$Path"/log/"$Filename".LC."$i
	echo -n $Filename","$i","$Seed",LC,-1-2-3-4-5-6-7-8-9-10-11-12-13-14-15-,"
	./effort $ConvertedFile $Path "BasicLC" $TestFile $TrainFile "-1-2-3-4-5-6-7-8-9-10-11-12-13-14-15-" $LogFile

	if (($DoCocomostV2==1))
	then
		TargetEMs=$(./lcFssFilter $TrainFile)
		
		LogFile=$Path"/log/"$Filename".cocomostV2."$i
		echo -n $Filename","$i","$Seed",cocomostV2,"$TargetEMs","
		./effort $ConvertedFile $Path "BasicLC" $TestFile $TrainFile $TargetEMs $LogFile
	fi
	
	if (($DoCocomostV3==1))
	then
		# This loop creates the expert's guesses
		ExpertFile=$Path"/log/"$Filename".experts."$i
		for ((j=1;j<=$Experts;j++))
		do
		  # Set Variables
	  	  vTestFile=$Path"/data/v_test/"$Filename".numeric."$i"."$j
		  vTrainFile=$Path"/data/v_train/"$Filename".numeric."$i"."$j
		  vTestSets=$(gawk -v M=$Mprime 'END{print int(NR*M)}' $ConvertedFile)
		  vSeed=$RANDOM
		  LogFile=$vTrainFile".log"
		  
		  # Generate the vTestFile and vTrainFile from the TrainFile
		  gawk -f randomize.awk -v Seed=$vSeed -v TestSets=$vTestSets -v TrainFile=$vTrainFile -v TestFile=$vTestFile $TrainFile

		  # Find the Best Evaluated Subset from the vTrainFile.
		  TargetEMs=$(./lcFssFilter $vTrainFile)

		  # Try local calibration with the virtual dataset in order to evaluate the expert.
		  gawk -f simpleLC.awk Pass=1 TargetEMs=$TargetEMs $vTrainFile Pass=2 TargetEMs=$TargetEMs $vTrainFile > $LogFile
		  #gawk -f simpleLC.awk Pass=1 TargetEMs=$TargetEMs $vTrainFile Pass=2 TargetEMs=$TargetEMs $vTestFile > $LogFile

		  # Find the Pred30 of this Expert		  
		  Stats=$(./calc $LogFile)
		  Pred30=$(echo $Stats | gawk 'BEGIN{FS=","} NR==1{print $3}')
		  
		  # Let the expert make a guess at the actual Test Set
		  TestResult=$(gawk -f simpleLC.awk Pass=1 TargetEMs=$TargetEMs $vTrainFile Pass=2 TargetEMs=$TargetEMs $TrainFile)
		  Estimate=$(echo $TestResult | gawk 'BEGIN{FS=","} NR==1{print $1}')
		  
		  # Record the experts estimate, weight (as pred30), randomization seed, and subset to a file.
		  echo $Estimate","$Pred30","$vSeed","$TargetEMs >> $ExpertFile
		done
		
		# Now generate an estimate from the combined wisdom of the experts
		ExpertEstimate=$(gawk -f experts.awk $ExpertFile)
		Actual=$(gawk 'BEGIN{FS=","} NR=1{print $NF}' $TestFile)
		Mre=$(gawk -v e=$ExpertEstimate -v a=$Actual 'BEGIN{re=(e-a)/a;mre=re<0?-1*re:re;print 100*mre;}')
		echo $Filename","$i","$Seed",cocomostV3,N/A,"$ExpertEstimate","$Actual","$Mre
	fi	
	
	#--------------Old Cocomost--------------
	#if (($DoCocomost==1))
	#then
	#	IdealsFile=$Path"/log/"$Filename".ideals."$i
	#	gawk -f findIdealsByLC.awk Pass=1 TargetEMs="-" $TrainFile Pass=2 TargetEMs="-" $TrainFile > $IdealsFile
	#	TargetEMs=$(./idealsFssFilter $TrainFile $IdealsFile)
		
	#	LogFile=$Path"/log/"$Filename".cocomost."$i
	#	echo -n $Filename","$i","$Seed",cocomost,"$TargetEMs","
	#	./effort $ConvertedFile $Path "BasicLC" $TestFile $TrainFile $TargetEMs $LogFile
	#fi
	#----------------------------------------
		
done