------------------------------------------------------------------------------
Change Log  (only recent)
------------------------------------------------------------------------------
12/5/12 Version 9.0.4
       Fixed yaap measure (Stephen Robertson).
12/5/12 Version 9.0.3
       Added yaap measure (Stephen Robertson).
2/28/12 Version 9.0.2
       Cleaned up warning messages again, Corrected -c flag implementation
7/14/09 Version 9.0
       Long past time to get this out there, despite my plans
       for more features!  Minor cleanup of warning messages 
       in 64bit and gcc 4.4 compilers
12/31/08
       comments and documentation of Zscore file format corrected.
       trec_eval.c get_zscores.c
3/3/08 Version 9.0alpha.  
       Complete rewrite of entire trec_eval (needed for a long time - 
       core code was written in the 1980's and was ugly then!). 
       Complete separation of individual measure calculations -
       computers are now fast enough so can afford recalculation of lots
       of intermediate values.  
       Parameters for measures (eg, cutoffs for P) can be specified on the
       command line.  
       Choice of measures can be specifed on the command line.
       Should be much easier to add measures, and much easier to add
       new input file formats with associated measures.
       An initial set of preference evaluation measures (with their own
       input rel_info format) have been added.
       Help now gives targeted measure and format descriptions.  Try
          trec_eval -h -m all_prefs -R prefs
       to get info on preference measures and formats, for instance.
       All internal calculations are in double rather than float. Yields
       minor variations in output at rare times; mostly when going from a
       double percentage to a corresponding doc cutoff (eg, in iprec_at_recall).
       All globally known procedure names or variables now begin with 'te_'
       to allow incorporation of procedures in other programs.
       
       Measures added:
       	        ndcg, ndcg_cut, set_F, success, map_avgjg, P_avgjg, runid,
		relstring, ndcg_p
		various preference evaluation measures.
       Measures renamed:
                set_P			was exact_prec
	        set_recall 		was exact_recall
	        set_relative_P		was exact_relative_prec
	        set_recall		was exact_recall
	        set_map			was exact_unranked_avg_prec
		gm_map                  was gm_ap
		11pt_avg                was 11-pt_avg
                Rprec-mult_*            was *R-prec
		P_*                     was P*
                recall_*                was recall*
		relative_P_*            was relative_P*
		iprec_at_recall_*       was ircl_prn.*
       Measures dropped for now:
       		3-pt_avg       	
		avg_doc_prec   	
		avg_relative_prec	
		exact_relative_unranked_avg_prec	
		map_at_R       	
                int_map        	
                exact_int_R_rcl_prec
                int_map_at_R   	
                unranked_avg_prec*
                relative_unranked_avg_prec*
                rcl_at_142_nonrel	
                fallout_recall_*
                int_*R-prec
                micro_prec     	
                micro_recall   	
                micro_bpref    	
		bpref variants
		time base measures.
	Input formats added:
	        prefs - allows expression of preferences
		qrels_prefs - same as standard qrels, except treated as prefs
		qrels_jg - same as standard qrels, except allows judgment sets
                           from multiple users (judgment groups).

Version 8.1, Added infAP, minor bug fixes
7/24/06 Improved infAP comments (implementation verified by Yilmaz).
        trec_eval_help.c: allow longer measure explanations.
6/27/06 get_opt.c Fixed error message
6/22/06 Added measure infAP (Aslam et al) to allow judging only sample 
        of pools.  -1 for rel in qrels file interpreted as pool doc not judged.
6/22/06 trvec_teval.c: fixed bugs in calculation of bpref if multiple
	relevance levels were used and a non-default relevance level
	was given. (Eg. A doc with rel level of 2 was counted as unjudged
	rather than judged nonrel if a relevance level of 3 was needed
	to consider relevant.)
4/5/06  Changed comments in README, trec_eval.c, trec_eval_help.c files 
        which incorrectly claimed queries with no relevant docs are 
        ignored (this was true with very old versions of trec_eval).  Now
        reads that queries with no relevance information are ignored.
        Giorgio Di Nunzio and Nicola Ferro,
------------------------------------------------------------------------------
Version 8.0, full bpref bug fix, see file bpref_bug.  I decided to up the 
        version number since bpref results are incompatible with previous 
        results (though the changes are small).