I've been a disciple of Code Monkey's "Managing Your iPod With Smartlists" (http://filebox.vt.edu/users/channum/files/smartlist_management_v5.1.pdf) tutorial since Summer 2007. Many thanks to Mr. Monkey for putting that together. It really started me down a whole new path of listening to my tunes.

So...this may be a bit too wordy but here's what I've learned only recently:

Having used a multi-platformed SPL approach to generate "random" playback for about 60GB of music, I went back into some of the base lists, thinking I would clear them out and reset the data base. I wanted to do this because as I rely on the Dynamic Pool for several SPL's and want to keep it as broadly sampled as possible.

Unlike a playlist that is based on a "random" selection, SPL's that are based on a least played rule ("least often" or "least recently") cannot be highlighted and cleared (iTunes won't allow it). So, when I sorted this apparently random list of least played selections by Artist, I realized that I had no artists between in the letters A-K (more or less, depending on the SPL), the lists for 4-star Least and Old; 5-star Least and Old; Old and Least (using CM's SPL names). Playing around with the "least" SPL's, I found that the exclusion of certain letters is dependent on the number of tracks you allow to be in the list; i.e. the smaller the sample size the less letters of artists are allowed.

This reveals that for SPLs using a "least" rule, iTunes is using a criteria other than randomness. I guess that's obvious, I'm just curious as to how it works correlated to the number of allowed tracks. It also has a negative impact on the balance of a final SPL if it is excluding artists.

Anybody else notice this on Code Monkey's or another set of playlists using least played rules? Am I missing something? I'm interested to hear others thoughts on this, and perhaps design around the limitation without just having to use much larger sample sizes.
=============
Interesting.

I, too, follow the Code Monkey document. I just checked one of my base SPs that uses the "limit items selected by least often played" option and discovered the same thing you did except my list was limited to artists R-Z only.

Obviously that isn't cool.

I tweaked the SP by adding a criteria of "Last Played is not in the last 48 months" and then changed the selected by to "random."

The net effect is the same as the original intention (songs you haven't heard in a long time) but now the SP contains artists from A-Z and numbers.

Thanks for the post as I probably wouldn't have noticed otherwise.
=============
Although I don't use Code Monkey's tutorial I can confirm your findings in regards to the least played rule. If you have a large number of songs tied at "least played", iTunes will not select them randomly. But iTunes doesn't say it will either. It just says "select by least played", nowhere does it say "select by least played at random". ;)

What I have found is that even when configured to select randomly iTunes doesn't do a good job. For example, I just made a SP limited to 10 items and out of the 13000 possible items it chose 3 pairs out of the same album! That's 30% which is not random at all.

Still, on a large library it would be very hard to perceive this, specially if you also have shuffle play enabled.
=============
i also use a customized version of code monkey's approach. it's revolutionized my listening and i've recently "upgraded" from carrying an 80GB 5.5G to carrying an 8GB nano 3G.

Code Monkey mentions, "Whenever two or more songs match the same smartlist limitation criterion, e.g. selected by least often played, iTunes simply picks them in the reverse order of their position in the library (sorted by Artist). As such, smartlist limiting criteria that produce lots of ties will result in some clumping of artists."

john
=============
What I have found is that even when configured to select randomly iTunes doesn't do a good job. For example, I just made a SP limited to 10 items and out of the 13000 possible items it chose 3 pairs out of the same album! That's 30% which is not random at all.

I (mildly) disagree. Even chosing all 10 songs from the same album would a valid statistically random sample - and no more "unlikely" than all 10 from different albums.

statistially random /= psychologically unalike to the human mind
=============
This is addressed somewhere in the document: Any time there are a large number of ties for the selection criteria all iTunes does is, get this, walk backwards through the library alphabetically by artist and then start again at the end once it finishes. For example, right now my library is around the letter 'P'. It does this for all selection criteria, it's just the "Least Played" is the most likely one to generate large numbers of ties.

It's a kludge to be sure, but it's the way iTunes works. The only way to get around it is to introduce even more tiers, e.g. select 30% of your library "at random" and then select by least played from that - more trouble than even my anal retentive self can find justification for :)
=============
Thx all for the responses (i see a few more even before I've finished writing this!). As I think a bit more about it, a "Least Often" rule is kinda useless unless the library is small or the play time is high. Having many unplayed songs in the library doesn't really allow the criteria to come into play, and even when they do get played the same limitations will pinch off certain artists because of the way they were chosen alphabetically in the first place (at least I think that's what will happen).

ScoobZ, your idea to change the criteria to a random selection from songs not played in the last 48 months works, but alongside the other random SPLs it seems at first thought a bit overlapping.

I'm playing around with creating some other SPLs that are customized to the age of my library and how it can automatically generate a decent pool of the songs that have been "least played". I've created a group of 5 star songs that haven't been played in the last 6 months and not part of the other 5 star SPLs. For me, this generated about 250 songs in total, so I cut it off at 100 and select from the pool at random. In another month, assuming similar listening time, I will regenerate 75 5 star songs into this group. I've created a similar 100 song 4 star SPL, but instead use 12 months as the cutoff, given a much higher population to choose from and seeing what to expect as a recycling count.

I've also created a 4 and a 5 star SPL using a random sampling and cutoff tag of "least recently added". These shouldn't be susceptible to the problems discussed above, and will create another set of my oldest songs from which to cycle into the mix. I'm excluding my new SPL and the Random SPLs as well for these.

Funny you mention your "upgrade" urbanlegend, I almost did the same thing to replace a dead iPod about three weeks ago and chickened out with another 80GB :)
=============
Even chosing all 10 songs from the same album would a valid statistically random sample - and no more "unlikely" than all 10 from different albums.

statistially random /= psychologically unalike to the human mind
That's not correct. Pardon me if I bore you with some math and I'm going to go slightly offtopic here....

Let's say my library were 10 albums with each having 10 songs each and I made a playlist selected at random limited to 10 songs. What is the probability that they are all from the same album? There's only 10 possibilities out of a combination of 10 out of 100. Mathematically that's 100! / [(100-10)! * 10!], or 10 out of 17310309456440 different possibilities. That results in a very small probability of 0,00000000006%.

And to another question: what is the probability that 10 randomly picked songs are all from different albums? There are 36288000000000000 possible ways to choose exactly 10 different songs without repeating the same album. Since the total combination is 17310309456440 that's a 0,05776904235% possibility (significantly higher than the above).

In other words...

The likelihood of a random selection picking all 10 songs from the same album out of 10 albums with 10 songs each is 0,00000000006%.

The likelihood of a random selection picking all 10 songs without repeating the same album out of 10 albums with 10 songs each is 0,05776904235%.

Those are not psychologically different perceptions, those are mathematical facts! :D

EDIT: Something is not right with my second calculation, I'll look into it some time later and edit. But I hope you got the idea...
=============
As I think a bit more about it, a "Least Often" rule is kinda useless unless the library is small or the play time is high.Not quite. It's intended for medium to large libraries where you consistently add music. By having least played as one of the selection criteria the newer music gets more rotations through the playlists and catches up with the older music. In the absense of having such a rule, your playcounts per track, assuming comparing music of equal ratings, will always be biased towards music added earlier to your library. Least Played smooths out play counts.

Although it produces a large amount of ties and, consequently, clumping of artists, this is only around 22% of all tracks selected which will be lost in subsequent filters. Sure, you might notice this week you heard a bit more Motley Crue and Marvin Ga*ye (and next week you might notice a bit more Kelly Clarkson and Kiss) but it's a very small price for the smoothing out of playcounts.

Regardless, do what works for you, that was the whole point of my system: to inspire you to come up with something that works for your own needs. If the small amount of clumping least played results in bothers you more than not hearing newer music more often until it catches up, you figured out something in your favor :)
=============
That makes sense CM. Thx. Your encouragement in the guide to modify the approach per user taste has not gone unnoticed.
=============
electrozoid - right you are (damn! hate to say that :)). Conditional probability always trips me up - guess I was thinking about shuffling the whole library.....

.....I'll have to puzzle over your calculations.....thanks for the correction.
=============
BTW, I'll always want a SPL that will play more Kelly Clarkson than a totally random SPL would ;)

I wish there were more posts around here concentrated on SPL logic. One big sticky at the top doesn't give me enough to work with. I'll try to post more often on the topic.
=============
Not quite. It's intended for medium to large libraries where you consistently add music. By having least played as one of the selection criteria the newer music gets more rotations through the playlists and catches up with the older music. In the absense of having such a rule, your playcounts per track, assuming comparing music of equal ratings, will always be biased towards music added earlier to your library. Least Played smooths out play counts.

Although it produces a large amount of ties and, consequently, clumping of artists, this is only around 22% of all tracks selected which will be lost in subsequent filters. Sure, you might notice this week you heard a bit more Motley Crue and Marvin Ga*ye (and next week you might notice a bit more Kelly Clarkson and Kiss) but it's a very small price for the smoothing out of playcounts. Thanks for that. You may have mentioned it in the doc but seeing it clarified here is nice and helpful.
=============
Just to keep this thread going... I love to hear smart playlist ideas also.

I really learned a lot from CodeMonkey's document. I want to thank him again for that effort. My system, adopted from learning about CodeMonkey's, is much simpler and a little more work, but it works exactly the way that I want it... and that's the goal.

I have one playlist called "elite". It contains my 50 favorite tracks at the moment. I adjust this manually when I feel like it.

I have another playlist called "excellent". It contains a lot of songs rated 5, but not all. The ones that I did not include fall primarily into three categories: really long songs, need to be in the mood for it songs, or need to hear in combo with the song in front/behind it on the album, i.e. they flow together without a gap like Money/Us & Them. This is also a manual playlist. All of the elite songs also appear on this playlist so I don't lose them if I drop them off elite and forget to add them back here.

SPL Rack1 picks 25 of the 50 elite at random, not played in last 7 days.

SPL Rack2 picks 25 from excellent at random, not on elite, not played in last 90 days.

The final playlist that I listen to combines Rack1 and Rack2 randomly. All three of the SPL's are set to live update.

This method would be horrible for use as my primary music picker, but that's not how I do it. I only use this playlist for listening to a few songs when I don't feel like an entire album. So it always comes up with a song that I like and usually one I am in the mood to hear. I often come up with an album idea 3-4 songs into this playlist, and then I am off to play that album.
=============