[Evergreen-reports] [Sales] Publication year average

[Mike Rylander]

On Mon, Feb 19, 2024 at 12:18 PM Blake Graham-Henderson <
blake at mobiusconsortium.org> wrote:

> Mike,
>
> A new column is what I was thinking. I figured that I'd break something by
> converting the string column into an integer. Though, I didn't think it
> would have necessitated a new metabib field definition, because the results
> of the existing definition could be converted to numeric closer to the
> "Simple Record Extracts" end of the chain? Perhaps introduce the new column
> in one of the views up the DB view chain somewhere? Still referencing the
> original extraction def?
>
>
+1 to the new column, we're definitely on the same page there.

As for whether we should reuse a metabib field or create a new one, where
we can we should be using the right tool for the job and the tool designed
for this is the record attribute infrastructure, not the metabib field
infrastructure.  Whether we should change an existing record attribute
definition (date1 or pubdate), or create a new one that uses the MODS 3.3
transform to pull the mods:originInfo/mods:dateIssued value is a decision
that we can make.  The record attr pubdate is really close, and if we are
OK with changing the sort data so that decade-granular date1 fields (like,
201u becoming 2010 when it's some time in the 2010s but the cataloger
wasn't sure) sortable rather than showing up as NULL (and therefore sorting
to the end of the list).  I think that would be a net improvement, but I
don't want to assume.

I don't think we should be creating new special-purpose /logic/ code when
the existing functionality can be combined and applied to get where we want
to go.  That's why the record attribute infrastructure was built in a
generic and reusable way.  Put another way, Evergreen already knows how to
extract a single value for use by other business processes in the form of
non-multi record attribute definitions (the "other business process" being
reporting in this case) and to normalize arbitrary values in just the way
we want using the index normalizer map (even though we're not "indexing"
this, per se), so we should use that.

Any reporting-specific "view" (materialized or otherwise) is fair game for
growing new columns, so we can definitely mess with things close to the end
of the chain.  We should make sure that that's just about pulling in an
expected value and /not/ normalizing or otherwise cleaning up the data that
other code can already do for us.

Hopefully that all makes sense...

New column name: "Publication Year (numeric normalized)"
> pubyear_int
>
>
+1, IMO that's a good capsule definition for the goal.

Thanks,
--Mike


> -Blake-
> Conducting Magic
> Will consume any data format
> MOBIUS
>
>
> On 2/19/2024 11:06 AM, Mike Rylander wrote:
>
> Hrm... I traced back the date1 record attribute defintion, actually,
> rather than the pubdate metabib field.  It's important to note that record
> attributes and metabib fields have /very/ different use cases, ingest
> performance profiles, and configuration shapes.  What's most important here
> is that metabib fields are primarily meant to support search, and record
> attributes are primarily meant to support discrete value display and
> sorting.  We should try to use a single value (multi=false in the config
> table) record attribute here, rather than a metabib field.
>
> The drawback with Date1 (as in, the data coming from the 008) is that if
> you have really thin records the 008 may not exist. However I don't think
> the risk is really high there -- the record attribute version of pubdate
> comes from the 008 as well, and that is what we use as the data for the
> publication date sort axis.  Oh! And, looking closer, the pubdate attribute
> uses the "Number or NULL Normalize" index normalizer (id=18), which is the
> second half of what I described before -- I'd just forgotten it existed.
> Adding index normalizer 19 in a position before number-or-null, and then
> setting up the view stack to use that record attribute, could be all that's
> needed.
>
> So, I think the record attribute version of pubdate is actually the best
> data source.
>
> One thing to consider is existing uses of whatever extant field we end up
> wanting to make use of.  So, the Real Plan, IMO should do all that -^ as a
> /new/ record attribute rather than hijacking an existing one, nor should it
> use an existing metabib field (recall, those are about searching rather
> exposing data for other things to use), and have it land in a completely
> new column on the Simple Record Extracts materialized view.  Then there's
> no chance of breaking existing reports with a column datatype change.
>
> Thoughts on that?
>
> --
> Mike Rylander
> Research and Development Manager
> Equinox Open Library Initiative
> 1-877-OPEN-ILS (673-6457)
> miker at equinoxOLI.org
> https://equinoxOLI.org
>
>
> On Fri, Feb 16, 2024 at 2:23 PM Blake Graham-Henderson <
> blake at mobiusconsortium.org> wrote:
>
>> All,
>>
>> Thanks for your considerate responses. What Mike said is the conclusion
>> I had come to, and I was wondering if anyone else needs the publication
>> year to be an actual number so that the reporter can do things like
>> average,min,max,etc. From the sounds of it, no one is currently using
>> the Evergreen reporter to produce such a thing (I don't see how you
>> could). I suppose no one is using an external program to make it happen
>> (to meet collection reporting needs from the higher-ups)?
>>
>> I agree with Mike, in that the best place to get the publication year
>> (right now) is the Simple Record Extracts, because it hunts it down from
>> several places in the bib record. Walking it backwards:
>>
>> reporter.materialized_simple_record -> reporter.old_super_simple_record
>> -> metabib.wide_display_entry -> metabib.compressed_display_entry ->
>> metabib.flat_display_entry -> metabib.display_entry
>>
>> Which is a trigger-created-table based upon the index definition found
>> in config.metabib_field
>>
>> one of those views is hardcoded to expect "pubdate" to exist in the
>> metabib_field definitions. Which exists with stock Evergreen
>> definitions. Which is:
>>
>>
>> "//mods33:mods/mods33:originInfo//mods33:dateIssued[@encoding="marc"]|//mods33:mods/mods33:originInfo//mods33:dateIssued[1]"
>>
>> Decoding that is fun. Suffice it to say: the pubyear can come from
>> several places in the record, and I like that better than only looking
>> in one place.
>>
>> So, in conclusion, if a patch were written, I think it would be smart to
>> piggy back on this logic. It might be fairly straightforward to get the
>> first occurrence from the JSON string and cast it to an integer
>> (stripping out non-numeric characters first). That's where my thoughts
>> are right now. I don't think we're going to be writing the patch anytime
>> soon, just thinking through it with everyone.
>>
>> If everyone agrees that this is something that Evergreen should have,
>> and we agree on the method, I might champion the bug and patch for
>> future meetings and releases!
>>
>> -Blake-
>> Conducting Magic
>> Will consume any data format
>> MOBIUS
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.evergreen-ils.org/pipermail/evergreen-reports/attachments/20240219/64e86c7a/attachment.htm>