self-interacting features regardless of their weight

[trufanov-nok]

Resolves #698
One thing that bothers me is unexpected decrease of total number of features for test 65:
test/train-sets/ref/search_dep_parser.stderr
from 696025 to 695754

[trufanov-nok]

I believe that num_features is calculated wrongly in search_dep_parser as it calculated as in old permutations mode. I've tried to fix it and the total number of features in test 65 is dropped to 593481.
Still i'm not sure if this is enough and if this can explain previous number changes.

I think Hal should know this piece of VW?
@hal3 , Could you please take a look and confirm that this total number of features is has sense?

[hal3]

+cc kaiwei and He He who know more about this code

-- (brief: thumb-typed on a tiny keyboard)
Hal Daumé III me@hal3.name @haldaume3 http://hal3.name
"Arrest this man, he talks in maths." http://nlpers.blogspot.com

On Jun 25, 2015, at 6:47 AM, Alexander Trufanov notifications@github.com wrote:

I believe that num_features is calculated wrongly in search_dep_parser as it calculated as in old permutations mode. I've tried to fix it and the total number of features in test 65 is dropped to 593481.
Still i'm not sure if this is enough and if this can explain previous number changes.

I think Hal should know this piece of VW?
@hal3 , Could you please take a look and confirm that this total number of features is has sense?

—
Reply to this email directly or view it on GitHub.

[JohnLangford]

I'm going to merge this now, because I want to drop a version and it seems that getting self-interactions right is more important than accounting. Nevertheless, it's good to get the accounting right when we have a moment.