Wizard to help users write metadata

[JackKelly]

Simple wizard to walk people through the writing of their metadata

[balajikalluri]

Hello guys, I am intending to build NILMTK metadata with my own ICT plug load dataset. I've both individual and aggregate data of single circuit collected in CSV file format and scratching my head in getting started to implement my own metadata schema having followed component details of metadata framework. Any wizard on this?? -KMB

[JackKelly]

Hi @balajikalluri,

That's great that you'd like to use the NILM Metadata schema to describe your dataset!

sorry, I don't think I'll have time to build a wizard any time soon.

Instead, I'd recommend you have a look at the tutorial in the NILM Metadata docs.

Good luck ;)

[nikil511]

Hi Jack, I am in a simular situation, I guess I will study the tutorial. However just a quick question. If I understood correctly, typical datasets used with NILMTK use measurements from a central meter, and submeasurements from known appliances used to train the CombinatorialOptimisation and try to detect those appliances again in the central one. (correct?)

Does the toolkit address how to detect unknown appliances in the measurements of the central meter? (i.e. analyse the central meter data once to detect some appliances, strip out their signatures, train system based on those, and dissagregate) Do you have any tips on how could one solve this problem? or is it near impossible, so its better just to create an ever-growing database of known appliances to train against ?

Cheers, Manolis

[JackKelly]

Hi @nikil511

typical datasets used with NILMTK use measurements from a central meter, and submeasurements from known appliances used to train the CombinatorialOptimisation and try to detect those appliances again in the central one. (correct?)

Yes, that is correct!

Does the toolkit address how to detect unknown appliances in the measurements of the central meter?

Not really, no.

There are 'unsupervised' NILM algorithms which can find repeating patterns in the data (e.g. see Kim et al 2011). But NILMTK doesn't yet have an unsupervised algorithm.

[nikil511]

"unsupervised" is the keyword I was lookin for! :-) Thanks

On Wed, Nov 12, 2014 at 3:55 PM, Jack Kelly notifications@github.com wrote:

Hi @nikil511 https://github.com/nikil511

typical datasets used with NILMTK use measurements from a central meter, and submeasurements from known appliances used to train the CombinatorialOptimisation and try to detect those appliances again in the central one. (correct?)

Yes, that is correct!

Does the toolkit address how to detect unknown appliances in the measurements of the central meter?

Not really, no.

There are 'unsupervised' NILM algorithms which can find repeating patterns in the data (e.g. see Kim et al 2011 http://web.engr.illinois.edu/%7Ehanj/pdf/sdm11_hkim.pdf). But NILMTK doesn't yet have an unsupervised algorithm.

— Reply to this email directly or view it on GitHub https://github.com/nilmtk/nilm_metadata/issues/9#issuecomment-62720451.

[nipunbatra]

@JackKelly : Till the time we have this wizard, I think adding a line to the README pointing to some YAML validation online tool is sure to reduce the number of Github issues :)

@balajikalluri : Docs and existing metadata for other data sets should provide a good starting point.

[nipunbatra]

@nikil511 An unsupervised algorithm does exist in nilmtk, but needs to be worked upon further. BTW, your query might be better suited to nilmtk issue queue as opposed to nilm-metadata issue queue.