[wdmmg-dev] Proposed model overhaul

[Friedrich Lindenberg]

Hi Martin,

On Tue, Oct 4, 2011 at 12:36 PM, Martin Keegan <martin.keegan at okfn.org> wrote:
>>> a payment in respect of which we want to record a value in more than
>>> one currency?
>>
>> This, otoh, feels like its two ways of looking at the same payment
>> (i.e. a split between currencies would duplicate the amount in the
>> DB). One way to type for currency would be to have a flag on the
>> measure to specify its currency on a per-measure basis, rather than
>> per-dataset. I actually like this a lot, since it emphasises the fact
>> that we deal with spending. We now can decide on how specific we want
>> to be there (i.e. "USD" or "2010 USD" or "31.12.2009 USD").
>
> So, the main case I am thinking of is:
>
> you have a NGO which spends money in several currencies, and someone
> has dumped their raw bank statements onto you (this is not
> unrealistic), and these are in more than one currency (this is very
> realistic indeed)

I think we are talking about two separate issues here: one is a case
of having the right data (e.g. spending in multiple currencies) and
wanting to represent it - which is clearly within scope of the model
language.

The other issue is having incomplete data and wanting to infer the
remaining data from what is there (e.g. filling up a measure with
currency conversion). This, IMO, is very much outside the scope of the
model language.

If we are to include it, I propose we have a special section in the
model (e.g. "pre-load-transform") and allow users to specify
JSON-wrapped snippets of JavaScript that can will be executed by etl
before import. This could then have a library function to perform
currency conversion. The task specification language used could be
that of Google Refine (I still want someone to re-implement the Refine
service in Python for us and ScraperWiki to include).

I just think that including more and more quasi-procedural options
into the model format will lead us to doom (or re-building Windows
NT).

- Friedrich

For reference, a Refine snippet:

[
  {
    "op": "core/column-addition",
    "description": "Create column name_munged at index 2 based on
column name using expression grel:value.strip().toLowercase()",
    "engineConfig": {
      "facets": [],
      "mode": "row-based"
    },
    "newColumnName": "name_munged",
    "columnInsertIndex": 2,
    "baseColumnName": "name",
    "expression": "grel:value.strip().toLowercase()",
    "onError": "set-to-blank"
  }
]