srasher wrote: 
> Then lmsmusly.py works from the command line. Great!
> 
> Next step will be to include it into LMS, will try this tomorrow.
> 
> Cheers,
> Seb

Well, I got mixed results for that:

1. Installation on the ODroid:
Getting it to run the playlist generation on the Odroid wasn't possible
for me with medium effort. python3-requests could only be installed on
Ubuntu 14.04 (max2play's latest available version base for the U3) in a
compatible version (> 2.4.2) by using PIP. That was somehow OK. After
several minor issues, I reached a dead end due to some other command
used inside lmsmusly.py only working with >= Python 3.5. Trying to
install that beside the "default" 3.4 version on the Odroid lead to
dependency issues in apt, so at some point I gave up.

2. Generated playlist contents:
My next try was to let lmsmusly.py run on the x86_64 machine. Not a
viable solution for everyday use but OK for trials in the beginning or
for the "Party only" use case. Well, it doesn't show exactly the results
that I had hoped for. First it took rather minutes than seconds to
generate a 20 song static playlist addition based on the first song of
the current playlist (one of the examples on the LMSmusly web page). It
generates them, but the mix is rather strange. Within those 20 songs
(and I tried 5 or 6 times with always different seed tracks) it never
had less than 5 songs from the same album as the seed track. The
additional songs then most of the time had a majority of songs from the
same artist. And then there always where one or two songs that were
totally "off". Example: when starting a playlist with Alice Cooper, I
got 14 Alice Cooper songs, 7 from the same album, one from The Pogues,
one from AC/DC, one from Deep Purple, one from Aerosmith (that's all
fine), one song from Cradle of Filth (Black Metal, totally off
genre-wise) and one from Cindy Lauper (well, discussable ;-)). Similar
mix results with other seed tracks.

While the Cradle of Filth song might be something like a glitch, maybe
caused by the 30 seconds analysis time window or something, the general
distribution of the tracks was not what I had expected. Statistics: I
have analyzed ~25000 songs from 2200 albums, wide variety of genres from
Pop to Death Metal. All of the seed tracks "main genre's" have at least
100 albums within the collection, mostly from not less than 20 artists.

I know that the generation isn't based on metadata at all, still the
genre and artist distribution is something I would look at to judge the
"quality" of the mix ;-)

Maybe something went strange in my analysis of the tracks (left
everything at default in the related config.py).

I think it is a very interesting tool, especially given the hassle that
MusicIP usage nowadays generates. Will do more trials soon.

Thanks for your efforts anyway, Roland! Much appreciated.

Cheers,
Seb



"The only word I know is 'Grunt' - and I can't spell it" (R.I.P. D.A.)
------------------------------------------------------------------------
srasher's Profile: http://forums.slimdevices.com/member.php?userid=6209
View this thread: http://forums.slimdevices.com/showthread.php?t=108495

_______________________________________________
plugins mailing list
[email protected]
http://lists.slimdevices.com/mailman/listinfo/plugins

Reply via email to