Clarification on data format

[cj2001]

I have a quick question on the data format. The docs show:

train_hrt = dataset.train_hrt # numpy ndarray of shape (num_triplets, 3)
i = 0
print(train_hrt[i]) 

The docs say "Each triplet (head, relation, tail) in WikiKG90M-LSC represents an Wikidata claim, where head and tail are the Wikidata items, and relation is the Wikidata predicate." So I took away from that that these values are Wikidata identifiers (example Q42 is Douglas Adams). However, if I look at what the format of these triples are print(train_hrs[0:10]) what I get is

[[      0    1182  650146]
 [      0    1182  650332]
 [      0     876  989978]
 [      0     168 1054521]
 [      0    1182 1632931]
 [      0     876 2437486]
 [      0     324 2705654]
 [      0     324 5885915]
 [      0     427 6972024]
 [      0     427 7156732]]

If I take the head of that first triples, I had assumed this might be something like Q0, however Q0 does not exist in Wikidata.

So is there a way to map these integers to the text in Wikidata?

Thanks in advance!

[hyren]

Hi, the numbers in train_hrt are NOT Wikidata identifiers, but rather the node id/relation id in the KG. For example, the first row [0 1182 650146] represents there is the relation 1182 between node 0 and node 650146. And these idx ranges from [0, num_node) and [0, num_relation) on the KG.

We currently do not provide the raw text, but will probably release them in the future (as in #147 (reply in thread)). Thank you.

[cj2001]

Cool. I replied in that thread throwing in my support for you all eventually (hopefully soon!) putting that raw text data out there. Cheers!