W3C home > Mailing lists > Public > public-gld-comments@w3.org > December 2013

[QB] Implementation report

From: Sarven Capadisli <sarven.capadisli@deri.org>
Date: Wed, 04 Dec 2013 12:00:03 +0100
Message-ID: <529F0B33.5050306@deri.org>
To: public-gld-comments@w3.org
I would like report a Data Cube implementation.

http://www.w3.org/2011/gld/validator/qb/qb-test?upload=upload-2013-12-04T09-16-44-56

Failed 4


An sample from an IMF dataset is used: http://imf.270a.info/dataset/DM 
with accompanying metadata.

IC4 (i.e., every Dimension must have a declared range) works as 
expected. For what its worth, a bit about the implementation and why 
that particular test fails, and why I'm letting it fail for the time being:

The implementation is based on SDMX-ML to RDF/XML transformation ( 
Source: https://github.com/csarven/linked-sdmx , Documentation: 
http://csarven.ca/linked-sdmx-data ). The transformation effort keeps 
its assumptions minimal about the data and metadata. One particular 
example from the failure is that the source TimeDimension from the 
SDMX-ML KeyFamily for IMF DM doesn't mention a code list that it is 
using. When the code list information is available, it is used towards 
writing triples about qb:codeList and rdfs:range. Having said that, the 
the time dimension value from the SDMX-ML DataSet is "sniffed" to 
determine whether to use a suitable URI (e.g., British reference 
periods) or not in the data. At this time, this information is only used 
in the DataSet transformation, and not the DSD.

The range information can be derived from TimeDimension's concept 
reference. However, such concepts don't particularly help in the end as 
most of the time (based on my own observations) they only consist a 
label. I think a good-enough solution would be to use that concept for 
the range any way, and then later have a mapping from that concept to a 
vocabulary which the data ends up using.


I realize that most of this information is probably irrelevant for 
testing the spec, but I wanted to type it out any way. It essentially 
points out that we need to have more heuristics built into the 
transformations [at least for Linked SDMX] in order to have a 
well-behaving data in QB.

:)

-Sarven
http://csarven.ca/#i
Received on Wednesday, 4 December 2013 11:00:33 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:51:37 UTC