Re: EasierRDF

On Tue, Feb 15, 2022 at 8:46 PM Chris Mungall <cjmungall@lbl.gov> wrote:

>
>
> On Tue, Feb 15, 2022 at 1:34 PM Martynas Jusevičius <
> martynas@atomgraph.com> wrote:
>
>> When most developers only know decades old software development methods
>> such as object-oriental and procedural programming languages and RDMBSs,
>> then they use them as hammers and expect everything else to be dumbed down
>> to become familiar nails.
>>
>
> Sorry that has been your experience, that hasn't been mine, many
> developers I know are keen (sometimes too keen) to apply new tools and
> frameworks, but also exhibit a good practical sense of what tool is most
> appropriate for the task at hand
>
>
>> Imagine if this discussion happened not in the Knowledge Graph but in the
>> AI community (EasierML?): "In our experience, developers do not like the
>> machine learning stack (PyTorch, Tensorflow etc.), they vastly prefer JSON,
>> TSVs, RDBMSs...". Said no one ever, because there would be no ML if this
>> was the case.
>>
>
> I don't follow the analogy here, I don't think JSON is an alternative to
> Tensorflow. If anything this exemplifies my point, since modern ML tools
> embrace the data stacks that developers were already familiar with (e.g
> pandas) rather than replace them.
>

My point is that JSON is no more an alternative to Tensorflow than it is an
alternative to RDF graphs. Both ML and KGs are rather complex technologies
solving complex problems, with application spaces that are larger and/or
different from what is usually done with JSON. So this EasierRDF debate is
comparing apples to oranges from the start.


>
>
>> You cannot build something truly new by only sticking to old (and
>> arguably simpler) tools, just like you can't have your cake and eat it.
>>
>
> I applaud your efforts for doing something new - linkeddatahub is a really
> cool application. For me, evolution works better than revolution, but your
> mileage and use cases may vary!
>
>
>>
>> On Tue, Feb 15, 2022 at 6:59 PM Chris Mungall <cjmungall@lbl.gov> wrote:
>>
>>>
>>>
>>> On Fri, Feb 11, 2022 at 9:50 AM Orie Steele <orie@transmute.industries>
>>> wrote:
>>>
>>> [snip]
>>>
>>>>
>>>> Any comments on https://linkml.io/ ?
>>>>
>>>
>>> I hesitate to jump into the fray here as these discussions frequently go
>>> nowhere..
>>>
>>> but as one of the instigators of LinkML, I can relate to the OP's
>>> points. I can't speak for all other industries, but in life sciences we
>>> have the combination of good terminological-style OWL ontologies and poor
>>> uptake of the RDF stack. There are awesome resources such as the uniprot
>>> sparql endpoint, but in my experience developers vastly prefer JSON, TSVs,
>>> RDBMSs, (LPG) graph databases, Mongo, etc. I am not talking solely about
>>> web developers here (which seems to be the focus of these discussions), but
>>> about the people in charge of building the major resources and data
>>> scientists who use them, as well as the non-technical data modelers and
>>> curators.
>>>
>>> The approach I have found to work is to give these developers what they
>>> want - JSON schemas, SQL DDL, python data models, spreadsheet-based data
>>> modeling -- and to make it easy to bind semantics/IRIs to these in an easy
>>> fashion on an as-needed basis. The idea with LinkML is to provide a simple
>>> over-arching metamodel that can be compiled to these traditional frameworks
>>> - but also to json-ld contexts, shacl, owl, etc.
>>>
>>> YMMV, if you are happy working purely within the RDF stack and you have
>>> a team of taleneted developers who are also happy here, more power to you!
>>>
>>> I'm also cc-ing Matt Lange who works in the food security semantics
>>> domain and may be interested in this discussion.
>>>
>>>
>>>>
>>>> Regards,
>>>>
>>>> OS
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> ᐧ
>>>>
>>>> On Fri, Feb 11, 2022 at 8:12 AM Mike Prorock <mprorock@mesur.io> wrote:
>>>>
>>>>> Dan,
>>>>>
>>>>> By technical resources - do you mean hiring experts? Or software,
>>>>>> cloud hosting etc?
>>>>>
>>>>> More on the software side of things, and especially onboarding
>>>>> programmers who may have good data science experience or normal
>>>>> systems/full stack development experience, but who are new to dealing with
>>>>> JSON-LD.
>>>>>
>>>>> We treat JSON-LD as critical for VC use cases, especially in our use
>>>>> of the Trace Vocab / Interop items at CCG..
>>>>>
>>>>> Perhaps you could start a thread walking this community through some
>>>>>> real life use case? Eg
>>>>>> https://mesur.io/food-security looks fascinating… (ok I once worked
>>>>>> for UN FAO so I may have a distorted sense of “fascinating“).
>>>>>
>>>>> Ha! yes, we are big fan of solving problems in a few of those
>>>>> "fascinating" areas.  I probably should give a bit of a separate topic
>>>>> thread on the practical applications.
>>>>>
>>>>> Mike Prorock
>>>>> CTO, Founder
>>>>> https://mesur.io/
>>>>>
>>>>>>
>>>>>>>>
>>>>
>>>> --
>>>> *ORIE STEELE*
>>>> Chief Technical Officer
>>>> www.transmute.industries
>>>>
>>>> <https://www.transmute.industries>
>>>>
>>>