-
Notifications
You must be signed in to change notification settings - Fork 48
/
CHANGELOG
100 lines (98 loc) · 4.68 KB
/
CHANGELOG
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
------------------------------------------------------------------------------
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).