[Imports] US - HIDFL (DHS) Hospital Import

[William Edmisten]

I would still like to propose an import. Manually going through the data
and each hospital's website is fine to find missing hospitals in OSM, but
it's less feasible for the full 8,000 records.

Based on the feedback in this thread, it sounds like we'll need to better
assess the quality of the data first. I've seen a previous approach of
comparing 100 random records to OSM and measuring the accuracy based on
that.
This would also require a separate wcedmisten_imports account.

I think this would still need to be a semi-automated import - where the
HIFLD data is used as primary source, but a human still verifies every
proposed change.

Addressing Greg's point about a fully automated script:
> It's not formally part of the norms, but I think imports should be done
> with code, basically a program that takes the OSM db as it is and the
> import db and produces a set of proposed editing actions
> (additions/conflations) and things to inspect (object in OSM but not in
> import).  That can then be debugged before any changes are made.  It is
> critical that it be automated so that it can be re-run with little effort.

I'm a bit unclear how this would work in practice. I feel like there are so
many edge cases in OSM tagging styles that it would be quite difficult to
ensure the program handles everything appropriately without checking
everything.
Do you have any prior examples of successful fully automated import
programs that have been run?

This is why I'm only proposing a semi-automated import, just to have human
eyes on things.

Best,

William Edmisten (wcedmisten)

On Fri, May 5, 2023 at 5:37 PM Mike Thompson <miketho16 at gmail.com> wrote:

> Thanks for the additional information William!
>
> On Fri, May 5, 2023 at 1:08 PM William Edmisten <wcedmisten at gmail.com>
> wrote:
>
>> I hope to distinguish two different projects which I may have conflated
>> in my initial Wiki page: 1) using HIFDL Hospitals as reference material,
>> but not primary source, to find missing hospitals in OSM and 2) the actual
>> import of the HIFDL Hospitals dataset, which would rely on HIFDL as the
>> primary source rather than as reference.
>>
> Are you therefore dropping the proposal for an import?  It seems that if
> you are going to do all of this manually, there is no reason for an
> import.
>
>>
>>
>> I do not believe 1) constitutes an import, because it is only using the
>> dataset as reference to find missing POIs, but using objectively verifiable
>> sources (each hospitals' website), similar to the workflow of many
>> "armchair" mappers. I believe this is a widely accepted practice in the OSM
>> community.
>>
> Your original email, and the wiki page that it cited, led me to believe
> that you intended to do an import, and thus the rules for imports would
> apply, and thus my comments.
>
>>
>>
>> To the point on VAL_DATE, I believe checking the hospital website
>> mitigates this issue also.
>>
> You might consider going to each state and getting its current list of
> licensed hospitals, this is what HIFLD does to build/maintain this
> dataset.  This would avoid the issue of old data in the HIFLD dataset.
>
> Mike
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/imports/attachments/20230506/e44f205d/attachment.htm>