Hello everyone,
I have a problem with adding feature functions in moses_chart : It seems to
work at decoding : it appears in the score breakdown of the ChartHypothesis
and it affects the output.
The problem I have is with mert : my new feature seems to be read but it
does not appear in the n-best list.
The first entry of my n-best list looks like this :
0 ||| main headline inflation in the eurozone are.V. soaring.V. food prices
. ||| lm: -24.0627 tm: -26.3145 (...) w: -5.64583 ||| -13.7065
but my feature (labelled as zp) does not appear in it.
Did someone also have this problem.
Many thanks for helping.
Bests
Fabienne
Here is what I have done :
My feature is very similar to the word penalty and has no model file. So I
have added it to $default_triples in mert-moses.pl. (where the word penalty
also is).
I have also added it into $ABBR_FULL_MAP ("d=weight-d lm=weight-l
tm=weight-t w=weight-w zp=weight-zp g=weight-generation lex=weight-lex
I=weight-i").
I have also tried to pass the lambdas as option to
mert-moses.pl(--lambdas='zp:0,-0.5-0.5 lm:1,0.5-1.5
tm:0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0.14,0.25-0.75;0,-0.5-0.5
w:0,-0.5-0.5')
When I run the decoder, it looks like the zp feature is registered : -zp
0.000000 appears in the decoder config :
(1) run decoder to produce n-best lists
params =
Normalizing lambdas: 0 1 0 0.14 0.14 0.14 0.14 0.14 0.14 0.14 0.14 0.14 0.14
0.14 0.14 0.14 0.14 0
DECODER_CFG = -w %.6f -lm %.6f -zp %.6f -tm %.6f %.6f %.6f %.6f %.6f %.6f
%.6f %.6f %.6f %.6f %.6f %.6f %.6f %.6f %.6f
values = 0 0.337837837837838 0 0.0472972972972973 0.0472972972972973
0.0472972972972973 0.0472972972972973 0.0472972972972973 0.0472972972972973
0.0472972972972973 0.0472972972972973 0.0472972972972973 0.0472972972972973
0.0472972972972973 0.0472972972972973 0.0472972972972973 0.0472972972972973
0
decoder_config = -w 0.000000 -lm 0.337838 -zp 0.000000 -tm 0.047297 0.047297
0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297
0.047297 0.047297 0.047297 0.047297 0.000000
Executing:
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/mosesdecoder/moses-chart-cmd/src/moses_chart
-config
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/tuning/filtered_rule_tables/moses.ini
-inputtype 0 -w 0.000000 -lm 0.337838 -zp 0.000000 -tm 0.047297 0.047297
0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297
0.047297 0.047297 0.047297 0.047297 0.000000 -n-best-list run1.best100.out
100 -input-file
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/devtest2009/tuning.input
> run1.out
Defined parameters (per moses.ini or switch):
config:
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/tuning/filtered_rule_tables/moses.ini
cube-pruning-pop-limit: 1000
glue-rule-type: 0
input-factors: 0
input-file:
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/devtest2009/tuning.input
inputtype: 0
lmodel-file: 0 0 5
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/lm/news.lm
mapping: 0 T 0 1 T 1 2 T 2
max-chart-span: 10 10 1000
n-best-list: run1.best100.out 100
non-terminals: X
search-algorithm: 3
ttable-file: 2 0 0 7
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/tuning/filtered_rule_tables/
rule-table-r1.pt 2 0 0 7
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/tuning/filtered_rule_tables/
rule-table-r2.pt 6 0 0 1
/mount/arbeitsdaten9/projekte/morphosynt/hiero-system-standard-vi-zcount/working-dir/model/glue-grammar
ttable-limit: 20
weight-l: 0.337838
weight-t: 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297
0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297 0.047297
0.000000
weight-zp: 0.000000
weight-w: 0.000000
Loading lexical distortion models...have 0 models
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support
