#At class on Tuesday, the following technology was
#identified as required for Project2.

#1) running jrip
# see jrip and jrip10 in minerc

#2) 1Rdiscretization

#3) c45rules
# 	I can't find c45rules in weka.
# 	Use "PART" instead
# 	weka.classifiers.rules.PART -M 2 -C 0.25 -Q 1
#   For more on "PART", see witten&frank p409

#4) script for the wrapper
#	see the demo below

###################################


# to use this demo
# copy this file to $HOME/opt/our/minerc
# bash our minerrc
# . fssdemo
# worker1001

###################################
#The following code takes some inputarff file $1 and re-orders the attributes, left to
#right, most interesting to most dull.

#This code calls infoGain:

rankViaInfoGain() {
 blab "<"
 $Weka  weka.filters.supervised.attribute.AttributeSelection \
 -S  "weka.attributeSelection.Ranker -T -2.7976931348623157E308 -N -1"  \
 -E  "weka.attributeSelection.InfoGainAttributeEval"                    \
 -i  $1 -o tmp.arff
 cat tmp.arff
}

#This code calls wrapper:

rankViaWrapper() { # warning: slow!
 blab "<"
 $Weka  weka.filters.supervised.attribute.AttributeSelection \
 -E  "weka.attributeSelection.WrapperSubsetEval -B weka.classifiers.bayes.NaiveBayes -F 5 -T 0.01 -R 1" \
 -S  "weka.attributeSelection.BestFirst -D 1 -N 5" \
 -i  $1 -o tmp.arff
 cat tmp.arff
}


logNumbers() { 
		gawk '
	BEGIN 						                {Min=0.00001; OFS=","; 
	                                             IGNORECASE=1}
	/@attribute/ 				                {Attr++}
	/@attribute/ && $3 ~ /numeric|real|integer/ {Num[Attr]=Attr; }

	/@data/      {In=1; FS=","}
	/@/          {print; next; }
	In && NF > 2 {
	for(I=1;I<=Attr;I++) 
		if (I in Num) 
			if ($I !~ /\?/)  {
				if ( ($I +0) < Min) {Bad=1} else {Bad=0}
				if (Bad) $I= Min;
				$I=log($I)	
		}
	print $0
	}
	' -
}
#### some workers
worker1001() {
	cd $Tmp
	local bins=2
	local repeats=2
	local learners="nb"
	local data="$Here/minerc.lib/arffs/uci/discrete/diabetes.arff"
	for one in $data; do
		cp $one raw.arff
		stem=`basename $one`
		stem=${stem/.*/}
		cat raw.arff | logNumbers                    > logged.arff
		discretizeViaFayyadIrani raw.arff            > discrete.arff
		discretizeViaFayyadIrani logged.arff         > loggedDiscrete.arff
		for x in raw discrete; do # logged loggedDiscrete; do
			rankViaInfoGain $x.arff > ranked.arff
			#rankViaWrapper $x.arff > ranked.arff
			for Attrs in 6  4 2 ; do
				removeAttributes $Attrs 8 ranked.arff > ranked${Attrs}.arff # removes all attributes from $Attrs to 8
				blab "$stem $x $Attrs "
				echo "#file,x,attrs,bin,learner,goal,a,b,c,d,acc,pd,pf,prec,g"
				for((R=1;R<=$repeats;R++)); do
					Seed=$RANDOM
					for((bin=1; bin<=$bins ;  bin++)); do
						blab "$bin"
						cat ranked${Attrs}.arff | makeTrainTest $Seed 10 1 
						goals=`cat test.arff | classes --brief`
						for Learner in $Learners; do
							$Learner train.arff test.arff | gotwant > results.dat
							for goal in $goals; do 
								cat results.dat |
							    abcd --goal "$goal" \
									 --prefix "`basename $one`,$x,$Attrs,$bin,$Learner,$goal"\
									 --decimals 1
							done
						done   
					done
				done	
				blabln
			done
		done  
	done  | tee $Safe/worker1001.log
	cd $Here
}