Jack Kelly

Some very quick thoughts (I'm on my phone and just grabbing a minute of time to write...): The common metadata specifies on_power_threshold for all appliance types (this can be overridden...

If the end goal is just plotting then do we need to apply a threshold? Could we just rescale each channel to the range 0 to 1 and plot that?...

@nipunreddevil wrote: > I implemented exact FHMM Awesome! Well done! that's a great addition to nilmtk! @oliparson wrote: > I expect you'll need to break long sequences into smaller blocks...

Thanks for doing these tests! Are those numbers in seconds? I presume so.

Hi, Just a quick thought: did you consider using Matt Johnson's [pyhsmm-factorial code](https://github.com/mattjj/pyhsmm-factorial)?

Hi @nuclearbit, welcome to nilmtk ;) > Any idea if the factorial code of Matt J an optimized one (using approximate inference) ? I'm afraid I don't know for sure...

Welcome @joostbr ;) And thanks for the information!

I've started a new issue to discuss approximate inference in F(S)HMMs: #78

Sounds like a great idea. It might be worth keeping the existing FHMM implementation and adding a new one (perhaps with a slightly different name) unless we're absolutely certain that...

Go for it ;) On 1 Feb 2015 21:11, "Nipun Batra" [email protected] wrote: > Just to ensure that the user knows that they may get down sampled data! > >...