Replies: 9 comments 2 replies
|
It depends what data you want to add to a device in the future. This is just the ZHA ones, add other integrations and you have a huge set of permutations to identify aliases for. |
|
I would expect this sort of data refinement/merging to be something to be done once you have an edit system set up, so editors can review the data and decide whether to merge (or keep separate) the various devices, and even do things like assign correct brand names to the TUYA devices. If you do too much cleanup to the collected data before editors see it, then there might be loss of useful information. |
|
Thanks for your feedback @andrew-codechimp @kepstin I generally agree with both views.
Generally, we would not remove data but try to assign through aliases where it makes sense. These aliases could then still be reviewed at a later point in time in case they are not correct - if that is something we identify to be required. For 2), we would aim to merge obvious duplicates, i.e. having same attribute values but e.g. different manufacturer - like the Tuya case, to become aliases. The idea behind that would be to generally reduce manual overhead at a later stage to review those cases - while there for sure will be many still regardless. |
|
As per public announcement to be the 'ultimate db on devices' I would think there needs to be both aggregation but also clarity. E.g. it would be beneficial for someone deciding between physical device A and B in terms of requirements, to see that A has x amount of users (aggregated total) with y different vendors or B with a single vendor. That doesn't necessarily state anything by itself, but depending on the vendors reputation one might decide for themselves how to interpret it. Like y shabby vendors vs risking a single company potentially stopping support overnight. |
|
To keep everyone aligned on this one: As mentioned here https://discord.com/channels/330944238910963714/1222192972519706798/1476301493161426974 Of course, if you happen to have further ideas on this topic let us know here :) |
|
Just a short update on this one - we're still working on that and refined the list of integrations and potential action points summarized here https://docs.google.com/spreadsheets/d/1S7zWacE1wSBOgsgj5JOE6rcyXXbVtlyGTgjeOwKA8Sg/edit?gid=942816726#gid=942816726 While we still plan to proactively create tickets to follow up upon per integration feel free to engage with the existing list and the feedback therein. |
|
Regarding the Tuya / Whitelabel discussion there's a proposal here |
|
Wanted to share this discussion here as well: home-assistant/architecture#1226 CC: @arturpragacz |

Uh oh!
There was an error while loading. Please reload this page.
As we have seen with white label brands like Tuya but also generally with how integrations are reporting devices regarding the fields manufacturer and model we want to focus more on reducing the "noise" and thus device duplicates upon receiving submissions.
Apart from having the threshold limitations for accepting submissions (>5 independent submissions for a device and >7day back-off before submitter is accepted (not yet implemented)) we need to further filter submissions to retain a healthy database going forward.
One solution we could think of at the moment to retain certain device duplicates for reference without removing them entirely is to proceed in a direction of introducing "aliases" for a device and trying to assign them automatically upon submission. That's something we could also use for future use cases when adding new devices through the Web UI and/or having aliases be acknowledged / approved by the community.
A concrete example for an alias could be the following - taken from the recent Tuya discussion on Discord:
Here you can see that zha reports multiple different manufacturer/model versions of TS011F
Instead, we could take just 1 TS011F entry and reference consecutive as aliases to it, while then counting only parent-entries.
Can you think of other use cases to further reduce noise?
Final note: For now, we would still keep the set of integration/manufacturer/model/model_id as the preferred way to uniquely identify devices. Which would mean that those need to be set in a pristine condition to serve as basis for further evaluation and contribution. In addition, we also aim to introduce an integration-independent metric of devices to have an effective list of unique devices.
All reactions