On Thu, 27 Jun 2019, Allin Cottrell wrote:
On Wed, 26 Jun 2019, Allin Cottrell wrote:
> On Wed, 26 Jun 2019, Sven Schreiber wrote:
>
>> I got this error when trying to fetch something from the Belgian central
>> bank via the dbnomics add-on:
>>
>> Datentypen nicht passend bei Operation [datatypes not matched]
>> *** error within loop in function fix_dimensions_bundle
>>> b.value_label = dvl[key][value]
>> called by function process_series_bundle
>> *** error in function process_series_bundle, line 38
>>> fix_dimensions_bundle(json, &b)
>> called by function dbnomics_get_series
>>
>> This is gretl 2019b and dbnomics add-on 0.33.
>
> I've just committed a work-around to git (updating the version to
> dbnomics 0.34). This involves replacing unreadable "value labels"
> with "<unknown value-label>".
>
> The problem arose from the fact that in dbnomics JSON certain
> aspects of series "dimensions" are represented as nested arrays,
> and gretl doesn't have nested arrays. That sounds like a
> limitation of gretl -- but IMO the problematic things should not
> be arrays in JSON: they should be objects [...]
Ah, on closer inspection the "dimension_value_labels" that are
given in array form in dbnomics JSON for the National Bank of
Belgium (NBB) are usually given (for other data providers, e.g.
the IMF) as JSON "objects" (which convert nicely into gretl
bundles).
So our trip-up with NBB data was actually due to an inconsistency
at dbnomics; I've posted a report on their forum.
I got a response on the dbnomics forum, and it turns out that what I
labeled an inconsistency ("dimension_value_labels" can take the form
of either JSON objects or JSON arrays, depending on something or
other) is actually intended, a "feature". I don't understand the
reasoning behind this, but anyway our dbnomics client now handles
the arrays version fairly gracefully; that is, I've found a better
work-around.
Allin