I have a somewhat extensive collection of music that is getting closer and closer to being fully rated, which means I would like to be able to use this to generate my play lists more or less automatically.

Unfortunately, I'm apparently too stupid to figure out how to accomplish this using any of Amarok's several methods for generating play lists. My only defense is that I'm also apparently unable to locate anything resembling documentation of Amarok.

When I select Help -> Amarok Handbook, there is virtually no useful information, and what scraps I find on http://amarok.kde.org/wiki/ falls very short of enlightening.

What I would like to accomplish is to have play lists generated that consists of:

a) Mostly highly rated (>3 stars) tracks not recently played (say, not in the last 3 weeks).

b) A sprinkling of lower rated (=3 stars) tracks not played in a fairly long while (say two months).

c) A couple tracks not yet rated/played (=0 stars and 0 play count).

If I try to set this up using the dynamic playlists (which would be my #1 preference, since that would mean I could just set it up and it would play indefinitely), I fail miserably. I can't even get close.

Even if I compromise with the time constraint, and set:

Proportional bias 100% last played older than 21 days.
Proportional bias 90% rating between 3.5 and 5 stars.
Proportional bias 5% rating equals 3 stars.
Proportional bias 5% rating equals 0 stars.

The result is not what I would expect. Tracks played within the last hour will pop up, and the ratios between differently rated tracks are nothing like what I'd expect, and the result even includes a bunch of tracks with ratings between 0.5 and 2.5 stars.


Ok, so switch to plan B, the new automated playlist generator. Since it generates a fixed length play list, it's not quite what I'd want, but I'm willing to give it a try.

My first attempt:

Match all
- Prevent duplicates
- Playlist length: equals 100
- Match any
- Match all
- Playlist length: equals 90
- Match tag: rating greater than 3 stars
- match tag: not last played within 21 days
- Match all
- Playlist length: equals 5
- Match tag: rating equals 3 stars
- match tag: not last played within 60 days
- Match all
- Playlist length: equals 5
- Match tag: rating equal to 0 stars
- match tag: play count equals 0

this gives me a play list that looks nothing like what I'd expect, including being >200 tracks long and having tons of 0 and 3 star tracks, and also takes a very long time doing it (like, 10+ minutes).

Experimentation suggests that it's the playlist length constraints that causes the excessive running time. Removing them, I'd expect to get a play list which is roughly evenly divided between the three categories, but instead the distribution appears to mirror the relative sizes of matching tracks in my collection.

It appears that the only way I can accomplish anything even remotely like what I hoped to accomplish is by:

a) Using the collection browser, search for "rating:>3 last:>21d OR playcount:0"
b) "Replace playlist" with the root object.
c) Select shuffle as the sorting.
d) Manually select all tracks except the 90 first tracks, and remove them.
e) Save this playlist.

f) Using the collection browser, search for "rating:3 last:>60d OR playcount:0"
g) "Replace playlist" with the root object.
h) Select shuffle as the sorting.
i) Manually select all tracks except the 5 first, and remove.
j) Save this playlist.

k) Using the collection browser, search for "rating:0 playcount:0"
l) "Replace playlist" with the root object.
m) Select shuffle as the sorting.
n) Manually select all tracks except the 5 first, and remove.

Go into the saved playlists -> amarok database
p) Add the playlist saved in step e)
q) Add the playlist saved in step j)
r) Select shuffle as the sorting.

And then, since I like to be able to inject specific tracks into my random playlist at points of my choosing, I also need to:

s) Save the resulting playlist.
t) Reset the playlist sorting
u) Load the playlist saved in step s)

So, now I wonder - isn't there a less... cumbersome way to accomplish this? The documentation I've found has been no help.

I even considered trying to modify some script to directly poke into the database, but alas, even this was foiled by the fact that the links to the database schema on http://amarok.kde.org/wiki/Development/Database_Schema yields only a 404 Not Found.

Amarok's documentation is here:
http://userbase.kde.org/Amarok/Manual
It's newly written and not entirely complete, so it's not been included in a release of the application yet.

I'm not sure what version of Amarok you're using, but if it's not the latest (2.4.0), changes have been made to dynamic/automated playlists. They're certainly not perfect, however, and they are being worked on.

I'm using 2.4.0, and from browsing through the manual you linked, it's certainly an improvement, but I'm still uncertain how to accomplish what I would like. I don't believe I found this manual while using the resources available through the Amarok home page.


About dynamic playlists it says:

Don't be confused, though -- a dynamic playlist containing the following two biases: 50% Artist: Michael Jackson, 50% Genre:Rock can contain tracks that are from neither Michael Jackson nor have a Rock genre. This is because while both biases are used, half the time a track will be chosen which is Jackson, and half the time not, and the same for Rock. So if you want 50/50 Jackson and Rock, set the biases to 100% for each.

So, I take it that me specifying

Proportional bias 90% rating between 3.5 and 5 stars.
Proportional bias 5% rating equals 3 stars.
Proportional bias 5% rating equals 0 stars.

Really means:

10% * 95% * 95% ~= 9% rating between 0.5 and 2.5 stars

I have to say, this is not quite what I'd expect.


The documentation for the automated playlist generator doesn't really give me any information I hadn't already deduced, and specifically, it does not deal at all with things like playlist length and constraint groups, nor with the issue of how having "any" constraint groups will affect the number of items generated by subitems of such a group.

Since I'm able to add playlist constraints to subgroups, it seems logical that these constraints would be local to the constraint subgroup, but this is clearly not the case, and then I wonder why I'm even able to add them to any group but the top?

It is confusing right now, and we know this. The developers are in discussions right now about a better, simpler user interface.

I'm the handbook writer, and since we are not quite done with it, it's not yet included in the Help menu. I'm incorporating the changes for 2.4.1 right now, and hope to finished within a few days. Whether or not that will allow it to be included in 2.4.1, I do not know. All I can do is write and illustrate until we are done.

valoriez wrote:It is confusing right now, and we know this. The developers are in discussions right now about a better, simpler user interface.

I'm the handbook writer, and since we are not quite done with it, it's not yet included in the Help menu. I'm incorporating the changes for 2.4.1 right now, and hope to finished within a few days. Whether or not that will allow it to be included in 2.4.1, I do not know. All I can do is write and illustrate until we are done.

Perhaps it is just confusion, but I honestly can't see or figure out any way, using any of Amarok's many ways of generating playlists, to accomplish my goal (see first post) short of the manual approach I outline.

Am I correct in concluding that this simply is not possible to accomplish with Amarok?

(I took Banshee for a quick spin, and there I quickly found a way to do a very good approximation of what I want. Not cherishing the thought of migrating my metadata, however. Plus, Amarok has some very neat features I'd hate to lose.)