examples/PRO/pro.config

### Part 1: parameters similar to Z-MERT
# target sentences file name (in this case, file name prefix)
-r	 ./ref.en
-rps	 4			# references per sentence
-p	 ./params.txt	  	# parameter file

#metric setting:
-m	 BLEU 4 closest
#-m	 TER nocase punc 5 5 joshua/zmert/tercom-0.7.25/tercom.7.25.jar 1
#-m	 TER-BLEU nocase punc 20 50  joshua/zmert/tercom-0.7.25/tercom.7.25.jar 1 4 closest
#-m	 METEOR en norm_yes keepPunc 2  #old meteor interface  #Z-MERT Meteor interface(not working)
#-m	 Meteor en lowercase '0.5 1.0 0.5 0.5' 'exact stem synonym paraphrase' '1.0 0.5 0.5 0.5' #CMU meteor interface

-maxIt	 30	   	     # maximum PRO iterations
-cmd	 ./decoder_command   # file containing commands to run decoder
-decOut	 ./output.nbest      # file prodcued by decoder
-dcfg	 ./joshua.config     # decoder config file
-N	 300                 # size of N-best list
-v	 1                   # verbosity level (0-2; higher value => more verbose)
-seed    12341234            # random number generator seed

### Part2: PRO parameters
#-trainingMode can be 1,2,3,4
#1: train dense feature weights only
#2: train dense & sparse feature weights together
#3: train sparse feature weights only(with dense feature weights fixed) also works)
#4: treat sparse features as one component(summary feature), train dense and summary feature weights together

-trainingMode	2

#-nbestFormat can be "sparse" or "dense"
#for trainingMode 1: either "dense" or "sparse"
#for trainingMode 2-4: use "sparse" format

-nbestFormat	sparse	#dense or sparse

#use one of the classifiers(and the corresponding parameter setting) below:
#1.perceptron paramters
-classifierClass	joshua.pro.ClassifierPerceptron
-classifierParams	'30 0.5 0.0'

#2.MegaM parameters
#-classifierClass	joshua.pro.ClassifierMegaM
#-classifierParams	'./megam_command ./megam_train.data ./megam_weights'

#3.Stanford Max-Ent parameters
#-classifierClass	joshua.pro.ClassifierMaxEnt
#-classifierParams	'./maxent_prop_file'

#4.LibSVM parameters
#-classifierClass	joshua.pro.ClassifierSVM
#-classifierParams	'./libsvm_command ./libsvm_train.data ./libsvm_train.data.model'

-Tau	8000		#num of candidate samples
-Xi	50		#num of top candidates
-interCoef	0.5	#linear interpolation coef. range:[0,1]. 1=using new weights only; 0=using previous weights only
-metricDiff	0.05	#threshold for sample selection