iTunes 5 random shuffle

At 8:29 AM -0700 9/16/05, Edward Reid wrote:
>what is mathematically known as "uniform", or at least something close.

Just to clarify, in this case "uniform" means "even distribution", correct?

--
Andrew Laurence
atlaurenuci.edu

On 9/16/05 10:29, "Edward Reid" <edwardpaleo.org> wrote:

> If you look at a page printed with tens of thousands of randomly placed
> dots, you *will* see patterns. If you see no patterns, this is a sure
> indication that the placement is *not* random but rather is carefully
> calculated to avoid patterns, including uniform density of placement at
> distance scales where the eye would pick up nonuniform density.

Hence the old Dilbert joke about the random number generator troll counting
out "nine, nine, nine, nine..."

"Is that really random?"

"Well, with random numbers, you can never be sure"

--
John C. Welch Writer/Analyst
Bynkii.com Mac and other opinions
jwelchbynkii.com

At 09:02 AM 09/16/2005 -0700, Andrew Laurence wrote:
>>what is mathematically known as "uniform", or at least something close.
>Just to clarify, in this case "uniform" means "even distribution", correct?

Thanks for pointing out that my use of terminology was a bit short of the
mark. Yes, "even distribution" expresses the desired state well in plain
English, though it has no mathematical definition that I'm aware of.

The issue is further complicated by that fact that what I called "random"
is known mathematically as a "uniform probability distribution". The
randomness is expressed by "probability distribution", and the "uniform"
means that all points are assigned equal probabilities. (Even this is a
loose description, but this isn't the time or place to go into the details
of probability distributions. Besides, that's another thing I'd need to
brush up on.)

What I called "uniform" is, as you say, well expressed in English by
"even". It does not and will not have a precise mathematical definition,
because what it really means is "not appearing to the human mind to contain
patterns". Any distribution which is uniform (and that's "uniform", not
"uniform probability") would show patterns to the mind -- for example,
rotating selections among albums would be seen as a pattern, not as
randomness. So again it's a perceptual thing we're looking for, not a
mathematical definition.

Edward
Art Works by Melynda Reid: http://paleo.org

Wired published an article on this topic today:

http://www.wired.com/news/culture/0,1284,68893,00.html?tw=wn_tophead_2

-Jeff

>Thanks for pointing out that my use of terminology was a bit short
>of the mark. Yes, "even distribution" expresses the desired state
>well in plain English, though it has no mathematical definition that
>I'm aware of.

I think what you are trying to get at is that for "random", a
mathematician means that for each choice of song, the computer should
pick one of the N songs, each with equal probability. What a user
typically would expect is closer to a "random permutation" of the
songs.

In the first case, if you have 100 songs, then the computer will
actually play the same song twice in a row, roughly once every 100
songs. No sane DJ would consider that what people wanted from a
"random" play list, and after playing 1000 songs, some songs will
have been played almost twice as often as other songs. (probabilities
range roughly from 75%-125%)

A more reasonable user definition would be "shuffle" rather than
random. A typical user would expect that with a 100 song collection,
after playing 1000 songs, each song would be played roughly 10 times,
and each song would have been played scattered more or less uniformly
across the full time.

A simple solution for this is to create a permutation of all the
songs, play that, and then repeat. You could still occasionally get
a duplicate song, but now for 100 songs, it would happen roughly one
time in 10,000 songs. Even that you could solve, by doing something
like this:

Create a permutation of all songs. When only 20% of the songs
remain, create a new permutation of the played 80%, and append it to
the remaining 20% and continue. Repeat that process.

Now it is impossible for a song to be played again within 20% of the
time, and the probability range is more like 90-110%.

Another method is to remember the time the song was last played, and
weight songs that have not been played recently more highly.

Naturally, all of these would need to be tested by real users to
determine what they actually thought...

Enjoy,
    Peter.

--
<http://www.stairways.com/> <http://download.stairways.com/>

I always use shuffle play in iTunes and find it acceptable, but I've had a somewhat more complex desire for my playback experience: I want iTunes to randomly play songs based on my preferences, using My Song Ratings as the determining factor. For example, there would be a 40 percent chance of a 5-star song playing, 30 pct for 4-stars, 20 pct for 3-stars and 10 pct for 2-stars (all user adjustable settings).

Another way of describing this is I want it to mimic the playing habits of many radio stations--play popular songs more often, but throw in less-popular songs along the way to add some variation to the listening experience. When I first started using iTunes last year, I mistakenly thought it did this sort of thing, and was very disappointed when I found it didn't (then I screwed up my library (very very disappointing) and had to start over again...).

While searching for other options a few months back, I found a posting on another website by someone who created multiple playlists, then created another playlist that played those playlists. He described this as his method to get roughly what I want. I've played with it just a little recently, still have to do a lot more experimentation to find the best mix. I enjoy the added control while playing music. Is it more random? No, quite the opposite, but it fits my desires for "pseudorandomicity".

Ciao, Tedley

At 09:15 -0700 UTC, on 2005/09/20, Peter N Lewis wrote:

[...]

> Another method is to remember the time the song was last played, and
> weight songs that have not been played recently more highly.

Yes, that's how I've been using iTunes 4. I use Party Shuffle and have it use
a Smart Playlist as its source. That Smart Playlist contains the 500 least
often played tracks (with a rating of 3 or more stars and excluding streams
and classical music). This way it feels 'random enough' and is guaranteed to
play each track equally often.

The downside is that over time, when you add more tracks, you oldest tracks
will play less often and newest more often. Changing the number of recently
played tracks of that Smart playlist (upping it to 1000, for example) can
bring that in balance again.


--
Sander Tekelenburg, <http://www.euronet.nl/~tekelenb/>

At 9:15 AM -0700 9/20/05, Peter N Lewis wrote:
>I think what you are trying to get at is that for "random", a
>mathematician means that for each choice of song, the computer should
>pick one of the N songs, each with equal probability. What a user
>typically would expect is closer to a "random permutation" of the
>songs.

For the correct algorithm, think bingo or keno.

- Assign a number to each song
- Put all of the numbers in a hat
- Shake the hat
- Draw the numbers

<http://en.wikipedia.org/wiki/Bingo>
<http://en.wikipedia.org/wiki/Keno>


--KP

I think what you are trying to get at is that for "random", a mathematician means that for each choice of song, the computer should pick one of the N songs, each with equal probability. What a user typically would expect is closer to a "random permutation" of the songs.

No, it's more than that. For example, it can be surprising when two tracks from the same album are played near each other in a random order. It's only surprising because of what the original poster said: we're always looking for order, whether it's there are not.

I can't tell if this is what Smart Shuffle does, but I think that people prefer imposing such restrictions (e.g. after a song has been played from a particular album, don't play another track from that album too soon again) rather than letting the (pseudo)random sequence used by iTunes have free reign.

At 9:15 AM -0700 9/20/05, home444 wrote:

>I always use shuffle play in iTunes and find it acceptable, but I've
>had a somewhat more complex desire for my playback experience: I
>want iTunes to randomly play songs based on my preferences, using My
>Song Ratings as the determining factor. For example, there would be
>a 40 percent chance of a 5-star song playing, 30 pct for 4-stars, 20
>pct for 3-stars and 10 pct for 2-stars (all user adjustable
>settings).
>
>Another way of describing this is I want it to mimic the playing
>habits of many radio stations--play popular songs more often, but
>throw in less-popular songs along the way to add some variation to
>the listening experience. When I first started using iTunes last
>year, I mistakenly thought it did this sort of thing, and was very
>disappointed when I found it didn't (then I screwed up my library
>(very very disappointing) and had to start over again...).
>
>While searching for other options a few months back, I found a
>posting on another website by someone who created multiple
>playlists, then created another playlist that played those
>playlists. He described this as his method to get roughly what I
>want. I've played with it just a little recently, still have to do a
>lot more experimentation to find the best mix. I enjoy the added
>control while playing music. Is it more random? No, quite the
>opposite, but it fits my desires for "pseudorandomicity".

This question of the user being able to control the parameters of how
iTunes plays back a library of songs is very fertile ground for
improvement. I'd like to see an entire preference panel of choices
provided.

I work in a shop where we use iTunes to play back a library of about
3000 songs. Generally, it is put on Party Shuffle with the Library as
source and it plays all day. There are other large Libraries on other
machines in our rather large building that occasionally get tapped as
a source when we get bored with our own stash.

Things I'd like to see.

1. Don't play any song that has been played in the last two days,
seven days, two weeks or month.

2. Only play from music imported in the last month, two months, three
months, etc., or from the entire Library without restriction.

3. (As our music Library is communal and contributed to by people
with wildly different tastes) A way to log who imported which songs.
(And on the days they are absent) don't play any song imported by a
designated user.

4. Do play (or don't play) any songs from a selectable set of all
available genres.

5. (Because our computer runs unattended most of the time) when
playing song from another library over the network, by-pass purchased
music that requires access be given, instead of just hanging until
someone has to manually advance the play (until the next hang).

6. A way to stipulate "never" play two songs by the same artist back
to back, no matter if they are from different albums. This could also
be on an escalating scale: don't play any artist within two, five,
ten songs of the last time played.

These are just a few of the more salient features I'd like to see.
--
Kevin Hopkins <kh2uiuc.edu>

On 22/9/2005 3:31 AM, "Kevin Hopkins" <kh2uiuc.edu> spake thus:

> Things I'd like to see.

Well, you can already do #1, #2 and #4 to varying degrees with Smart
Playlists. iTunes has the metadata to handle all of those: last played, date
added and genre, although doing genre filtering would proably be messy. #5
probably wouldn't be that difficult for Apple to add (given that iTunes can
clearly already identify files that require authorisation). #6 is pretty
much a variation of #2, so it should be technically easy to add. #5 would
require additional metadata, which could be more of a problem (although
doesn't ID3 allow extensible metadata now?).

Of course, if they would just implement Smart Playlists that allow us to
specify general Boolean rules, that solve most of the problem :)

--
Nigel Stanger, Dunedin, NEW ZEALAND.
http://public.xdi.org/=nigel.stanger

At 6:29 AM -0700 9/22/05, Nigel Stanger wrote:
>Of course, if they would just implement Smart Playlists that allow us to
>specify general Boolean rules, that solve most of the problem :)

1. Control of repeat play.

2. Play only the new stuff.

3. Importer (taste) filter.

4. Radio buttons - click "I feel like a jazz day" click - "I want to
rock my socks off." click -"Everything but opera." [Due to constant
additions and updating, play lists become a maintenance pain.]

5. Unattended, network considerations.

6. Enforced randomization of multi-importers' tastes.

As you said #1 could be accomplished by keying off of "last played."
Another less elegant way to assure no repeats for a long time while
in Party Shuffle is to allow "upcoming songs" not to be limited to
the arbitrarily set 100 songs, but to the maximum number of tunes in
the Library. Of course, switching between Sources in Party Shuffle
resets the upcoming selection, so a song just played under one Source
could get reselected from the new Source list and played again, so
keying off of "last played" would be more foolproof.

Number two could be solved by Smart Playlists and using "date added"?
I guess I show my ignorance, as I haven't explored Smart Playlists to
any great degree. I do find the slavish reliance on Playlists in
iTunes a pain.

Number three would require a user to enter their name before
importing something. This could be a Preference. On a single user
machine, constantly having to enter your name if you wanted to enter
something would be tedious, but on a multi-user machine, it could
prevent a lot of arguments. This would require metadata for the
importer's name. A radio button toggled importers listing would give
you a customizable "DJ List."

Number four is a request for flexibility, serving mood on the fly. A
Preference panel of radio buttons would make genre toggling and
filtering a snap. The problem with genre, though, is that it isn't
always part of the information imported via the Net, or correct even
if it is imported. More maintenance required.

Number five is just a situation that the programmers overlooked. A
network user should be able to tell their iTunes to skip over files
requiring authorization. This came up when Party Shuffling off of
someone else's Library as a Source.

I don't see #2 being a variation of #6. #2 could key off of "date
added," but #6 has to do with enforcing a more random play. #6 would
key off of "artist." In the environment in which I use iTunes, most
of the listeners can tolerate the other guy's selection for one song.
Put two of them back to back and you get complaints.
--
Kevin Hopkins <kh2uiuc.edu>

On 22 Sep 2005, at 2.29 pm, Nigel Stanger wrote:

> #5 would require additional metadata, which could be more of a
> problem (although doesn't ID3 allow extensible metadata now?).

(Number 5 in the original list was to skip purchased music.)

You can do this already with a smart playlist that filters on Kind --
set to 'doesn't contain' and 'Protected' -- since all the purchased
music has 'Protected MPEG audio file' as the kind. (Er, I'm guessing
on the basis of previews here, actually, since the iTunes Music Store
isn't available in Australia yet.)

The original situation was when protected music was being played over
the network, though, so this would mean you'd need to have the smart
playlist set up on the remote machine already.

David Shepherdson