Understanding the KGI example

53 views
Skip to first unread message

Shashi

unread,
Sep 30, 2019, 2:48:16 AM9/30/19
to PSL Users
I understand the working of PSL rules (Via the provided examples). But, while dealing with the KGI example, I dont understand what those id map to. i.e.

CAT_truth.txt:
A               B
19583 19256 0.0
9076 226231 0.9903381642512079
381387 60758 0.0
284492 277371 1.0
284287 277371 1.0
238106 277371 1.0
633582 220842 0.0
287238 277371 0.9903381642512076

I understand, these data corresponds to NELL database. What I dont understand is how do we map the value A (e.g. 19583) or Category B (e.g. 19256) to its actual label (text label) ?.

Is there an encoding and decoding function for these values?  

Eriq Augustine

unread,
Sep 30, 2019, 3:21:09 AM9/30/19
to Shashi, PSL Users
Hey,

For something like this, you should go back to the original paper this dataset was used in (you can find it in the dataset's README):

The repository listed in the paper should have the full dataset:

-eriq

--
You received this message because you are subscribed to the Google Groups "PSL Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psl-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/psl-users/91947023-6359-460f-aec2-78859c9f9823%40googlegroups.com.

Shashi

unread,
Sep 30, 2019, 5:44:41 AM9/30/19
to PSL Users
Unfortunately, even the paper does not talk about the dataset in specific. From what I read from different sources, NELL data set contains text labels. This some what looks like node ids/numbers. But, how do we confirm this ?.

Eriq Augustine

unread,
Sep 30, 2019, 11:15:22 AM9/30/19
to Shashi, PSL Users
Take a closer look at the repository for this paper.
There is a file in the data directory called "names,txt" that maps ids to names.

-eriq

--
You received this message because you are subscribed to the Google Groups "PSL Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psl-users+...@googlegroups.com.

Shashi

unread,
Oct 1, 2019, 1:36:52 AM10/1/19
to PSL Users
Thanks Eriq. I found it here https://github.com/linqs/kgi_musicbrainz/blob/master/data/1000_dataset_er/names.txt. (Might be useful for others having the same issue.)


On Monday, September 30, 2019 at 12:18:16 PM UTC+5:30, Shashi wrote:
Reply all
Reply to author
Forward
0 new messages