LAYSA Custom Radio

talking_animal said:
Well, back in May when Simon tried the BEREP radio over a week, he got 287 artists in 287 plays. So silly me, I assumed the last.fm radio would run through all the material before repeating itself.
Ditchdigger2005 said:
To the music: The radio seems to want to play all or most artists before it repeats them (I've noticed this with smaller radios), but although it seems to have happened in this case I don't think it's clever enough to not repeat an artist over different sessions, our radio is so big I probably just got lucky.His charts for that week are here.
Very interesting TAJ and I commend you for digging this out. However, have a feel for what "radio seems to want", direct comparison with BEREP Group Radio is difficult, since the artist pool is so much bigger. I'm sure there's 10K+ artists on my profile alone and, in May, I think there were about 90 members, obviously not mutually exclusive, artist-wise, but, certainly selected for diversity.
talking_animal said:
Plus, when I've tried the LAYSA custom radio before, it's never repeated itself.
Yep, listened for long periods and I do remember some artist repetition, but not much.
talking_animal said:
Assume that (while informative) the last two of your commandments above:

• frequency of track play is inversely proportional to the global ban-count for the track (I heard this from staff in PowerPlay discussions)
• frequency of track play is proportional to the global love-count for the track (following from the above)

are immaterial to this issue because this is about frequency of artists, not tracks.
Commandments? Oh my :P What's the world coming to? When some of the most astute people on the planet take my "guess"es to be commandments *grovels whilst scribbling note in Must talk to therapist about this book* ;)

We're fishing in the dark here, on how the algorithm works. Your conclusions, seem to, imply that you think it works on a simple reductionist basis. My visitation's more like this:

1. grab large number of tracks, at random, from those requested to play
2. examine recent play cache in removing tracks disallowed by radio licensing constraints
3. score all remaining tracks, on the basis of global ban/love counts and other black magic
4. play track with high score

Thus, those two assumptions would come into play.
talking_animal said:
When I did the math
*tackles: I think he means some math, folks.
it worked out that one should expect that 95% of the time, when one plays 18 tracks off the LAYSA list after five hours, there should be between 0 and 3 duplicates. The number I got was eight (8), which is off the charts. So either we have a whole troupe of artists who have very few streamable tracks, or there is some other factor influencing the selection beyond the ones you've enumerated above.
Finger in the air stuff, but I feel the streamable distribution we have's something like:

• 30% < 6 tracks
• 60% 6 to 14 tracks
• 10% > 15 tracks

If we're on the ball here, or anywhere near, I think, we'd expect most repeats to be from the most streamable of these groupings.

*sighs

All'n'all, our demands on this feature are extraordinary. Last.fm have a moral requirement to make it reasonably good, to satisfy all artists, big or small, get an equal shake type admonishments. However, the level of sophistication we'd like is inevitably expensive on resources and/or design/programing/testing time. Second guessing Last.fm technicalities is good sport, but, at this level, very hard verify and of limited reward.

I'm beginning to think that we should err on the side of caution and provide three, or so, additional cradios, splitting our connections chronologically. Come to think of it the latest's LAYSA Newcomers is worth doing anyhow. The other alternative's to cut the number of artist connections bellow the maximum allowed, or split them across genre/culture/geopoliticaly oriented splinter groups, but this's an area for when the groups reached much higher membership/maturity/artist-throughput.

Thoughts? - Your humble servant, Ian

talking_animal said:
What about dividing alphabetically, however? That way each station would have a mix of old LAYSA chestnuts and newer material.
A chronological divide would be, sufficiently easy to administer to be, manageable manually - controlled by posts, as this one. LAYSA Total (this one) and LAYSA Newbies would be updated every time. LAYSA Chestnuts and LAYSA Middle (needs better name), could be updated less often, without too much disappointment.

An alphabetical split starts to give me a headache, just thinking about their administration. I suggested three divisions 'cause that each gives a size close to where you saw repetition start. However, notwithstanding the use of spreadsheets to control the ball game^*, alphabetical division would require, at least, four divisions, to allow for natural variations in the alphabetical distribution, of connected artists. For most updates, most of these four would be due for change. More work! - Ian

^* - I'm resistant to addicting administration of the our radio(s) to spreadsheets, at this early stage. However, when we do go there (with Spreadsheets.Google.com, I guess), a split on a score basis would be feasible (derived from the virtual group count, calculated upon adding artists).

talking_animal said:
Post journal
This whole more radios, more, I want more, more, more more thing now seems to have a life of it's own. So it's been transplanted -> over here. - Ian

Here is our latest custom radio, straight from the labs:

There are 197 artists on it; see the artist connections for the full list. In addition, I have corrected several errors that crept in somehow so that there are no longer any dud ID numbers in the list.

The problem I had with the two-section list (one for names without punctuation, one for names with) was that the custom radio URL appeared on my system as only numbers. That meant I couldn't easily search it for artists' names. With the full numbered list I (or anyone else) can just copy the list into a new spreadsheet, take out the artists who have flown away or otherwise left the list, add the new arrivals, and republish in both table and custom-radio URL form.

Added (since 28 Sept):
Kimmo Pohjonen, Guy Davis

Flew Away (since 28 Sept):
Tali, Mark Springer & Sarah Sarhandi



Woops! Accidentally edited post instead of quoting it. Sorry.

Flew Away:
Muhal Richard Abrams _6695515
Randy Weston _2118448
Sivuca _1058273
Jah Shaka ₃₃₄
Leila ₁₅₅₃
Bheki Mseleku ₃₀₅
Sukia ₇₆₁
Amjad Ali Khan _9910520
Cherrywine _1240436
Funk Como Le Gusta _1041710
Karrin Allyson _1003648
Sonny Sharrock _1068245

Fell Out:
John Hughes ₂₂₇₉
Electronome ₅₉₇₃
Kazuki Tomokawa _2127989

Added:
Pharoah Sanders, Hamid Drake, Adam Rudolph _6644985
Toto Bona Lokua _7003908
Hermeto Pascoal E Grupo _6710735
Paula Morelenbaum _6737235
Joao Bosco _1242135
Amjad Ali Khan _9910520
Yothu Yindi ₁₄₁₇
U. Srinivas _1063808
Wiseblood ₃₇₄
Pietra Montecorvino _2156333
Mr Hudson & The Library _14251030
Hamid Drake - Joe McPhee _10566705
George Russell _1104254
Paul Winter Consort _1077875
Mustafa Kandirali & Ensemble _6633762