mpdstats is a plugin for beets that collects statistics about your listening
habits from MPD. It collects the following information about tracks:
- play_count: The number of times you fully listened to this track.
- skip_count: The number of times you skipped this track.
- last_played: UNIX timestamp when you last played this track.
- rating: A rating based on play_count and skip_count.
This plugin requires the python-mpd library in order to talk to the MPD server.
Install the library from pip, like so:
$ pip install python-mpd
To use it, enable it in your
config.yaml by putting
mpdstats on your
plugins line. Then, you’ll probably want to configure the specifics of
your MPD server. You can do that using an
mpd: section in your
config.yaml, which looks like this:
mpd: host: localhost port: 6600 password: seekrit
If your MPD library is at another location then the beets library (e.g.,
because one is mounted on a NFS share), you can specify the
music_directory in the config like this:
mpdstats: music_directory: /PATH/TO/YOUR/FILES
If you don’t want the plugin to update the rating, you can disable it with:
mpdstats: rating: False
If you want to change the way the rating is calculated, you can set the
rating_mix option like this:
mpdstats: rating_mix: 1.0
For details, see below.
Now use the
mpdstats command to fire it up:
$ beet mpdstats
A Word on Ratings¶
Ratings are calculated based on the play_count, skip_count and the last action (play or skip). It consists in one part of a stable_rating and in another part on a rolling_rating. The stable_rating is calculated like this:
stable_rating = (play_count + 1.0) / (play_count + skip_count + 2.0)
So if the play_count equals the skip_count, the stable_rating is always 0.5. More play_counts adjust the rating up to 1.0. More skip_counts adjust it down to 0.0. One of the disadvantages of this rating system, is that it doesn’t really cover recent developments. e.g. a song that you loved last year and played over 50 times will keep a high rating even if you skipped it the last 10 times. That’s were the rolling_rating comes in.
If a song has been fully played, the rolling_rating is calculated like this:
rolling_rating = old_rating + (1.0 - old_rating) / 2.0
If a song has been skipped, like this:
rolling_rating = old_rating - old_rating / 2.0
So rolling_rating adapts pretty fast to recent developments. But it’s too fast. Taking the example from above, your old favorite with 50 plays will get a negative rating (<0.5) the first time you skip it. Also not good.
To take the best of both worlds, we mix the ratings together with the
rating_mix factor. A
rating_mix of 0.0 means all
rolling and 1.0 means all stable. We found 0.75 to be a good compromise,
but fell free to play with that.