what is "True random shuffle "

[per01]

Hello all. Let me start by saying I really prefer Rockbox over my OF (Sansa Clip+ 8gb + 16gb card 3771 tracks Rockbox version 3.7). All of you are incredibly brilliant and I have a hard time understanding most of your posts as I am not an MIT educated engineer nor computer scientist (I am a lowly anally retentive tax accountant).

I do however have a knack for picking up patterns and although the shuffle may be a true random shuffle by definition, and is way better than Sansa's shuffle,  I do notice a certain pattern during playback which can not be attributable to observation bias. I have neither graphed nor documented my conclusion so scientifically my theory can't be proven, you'll just have to take my word for it; or ignore this post completely.  I just want to add my voice to those who have sensed that the shuffle function isn't as completely as random as it is defined to be.

Specifically, it appears to me that  tracks off of a certain album or 2 (or 3)  are somehow chosen to be interspersed with other apparently random chosen tracks.  After every 7-10 or so random songs, another track off of one album pops up.  In my most recent observation from this past week,  even though I have many Grateful Dead songs on my Sansa,  songs from "Postcards of the Hanging" are being played at least once every 1/2 hour.  Two weeks ago is when I first noticed the pattern and I waited another week before I  reached my conclusion and decided to publish my thoughts on your forum.

I am not imagining this phenomenon nor am I complaining about your programing skills (as I've said, you all are incredibly brilliant), I just wanted to share my experience.

[Llorean]

Specifically, it appears to me that  tracks off of a certain album or 2 (or 3)  are somehow chosen to be interspersed with other apparently random chosen tracks.  After every 7-10 or so random songs, another track off of one album pops up.  In my most recent observation from this past week,  even though I have many Grateful Dead songs on my Sansa,  songs from "Postcards of the Hanging" are being played at least once every 1/2 hour.

What would shuffle be other than songs being interspersed with other random songs?

As soon as you hit shuffle, or choose "insert shuffled" or however you start your playback, you can immediately pause playback and view your playlist. If there's a significant distribution problem it should be relatively easy to check by creating even as few as a half dozen new playback situations.

I know you believe that your knack for pattern recognition means your position is unassailable, but whether this is true or not if this isn't a problem that can be demonstrated in the actual playlist files and not solely in the ears of the beholder it makes it more or less impossible to solve as there's nothing visible to begin tracking down.

[per01]

I appreciate your devotion to Rockbox and your attempt to parse my post. That you even bothered to respond shows how much the Rockbox community cares about their product.

I never meant to imply that my "position is unassailable", my postion is very sailable  .  In fact,  I'm really not even taking a position one way or the other.  I just wanted to advise others of my experience and add my voice to those (if there are any others) Rockbox users that have a gut feeling that the shuffle feature may be playing one artist or album in less than random order.

I know it's not proper to read emotions into these types of things,but I sense a little anger in your reply.  Please don't be angry at my original post, that's the last thing I wanted to do.

[AlexP]

Gut feelings count for nothing.

Shuffle a playlist repeatedly, look at the distribution, do some statistics   This is what soap did a few posts back, and the distribution was random.  Until someone can prove otherwise, it is all just imagination.

[Llorean]

No anger. At most skepticism.

You offered no information that suggested it wasn't selection bias. In fact you specifically said "I am not imagining this phenomenon" which is what gave me the impression you felt your position was solid. Proof that you aren't imagining it would be easy to come by - you could generate a number of playlists under identical conditions and look at how the files are distributed. As well you said it "can not be attributable to observation bias" but what you've described (I noticed a phenomenon and then when I looked for it, I found it) is almost a textbook way of describing observation bias.

Shuffle shuffles all the songs you've selected. It's entirely reasonable to expect to occasionally have 2 or 3 hours periods where many songs from one album show up. Especially if you have lots of music. Some albums will be distributed exactly evenly across the entire playlist, other ones will be "clustered" around points of time because random distribution means that it also won't force them not to be near each other.

As AlexP, Soap and I have said, it's basically trivial to test this in an objective way (rather than introducing the subjectivity of "what I remember hearing" and so on). You could do as little as generate a shuffled playlist of all music (describing in your post what method you used to do this) then go through the first 50 songs and note down the number of songs from each album. Do this about 20 times (re-generate a new shuffled playlist, count). Then calculate what percent of your music each album represents (number of songs in album divided by total number of songs in the playlist you generated) then what percent of those thousand tracks you counted were from each album (number of counted instances divided by 1000). This isn't particularly thorough but if there's a strong flaw in the shuffling algorithm to cause preference for certain folders it's fairly likely to show if it's something as easily observable as described above.

[per01]

Your cluster theory makes sense to me and I accept that as the answer.

I will now crawl back under my stone (to paraphrase  Richard Thompson) and continue to lurk and learn from the Rockbox community.

Have you noticed that this thread has the most page views after the top 3? This subject must be on the minds of many Rockbox users. Anyway Happy Holidays to everyone reading this and to those people responsible for this great firmware.

[saratoga]

Have you noticed that this thread has the most page views after the top 3? This subject must be on the minds of many Rockbox users.

It has a lot of page views because its 4.5 years old, and old threads have more views then new threads.  I think if this mattered to people they'd have used the last 4 years to figure out if the shuffle was random enough.

[Llorean]

Actually, if we ignore the three stickied posts and sort all of the posts in "General Discussion" by page views, it's ranked 215 or so (it's on page 11 at 20 per page, 2 from the bottom, minus the three stickies).

This is more or less what we're talking about - a quick glance at information can lead to one conclusion but when you take the time to actually investigate it may turn out that that conclusion isn't soundly supported by the all of the data.

It'd be great if someone who's experiencing "non-random shuffle" would produce some data since that would clear things up once and for all if it is non-random, and would settle your own mind if it turns out to really be random.

[[Saint]]

It seems to me that the people bickering about "true random shuffle" don't actually *want* "true random shuffle" but instead merely think they do based on what they believe it to be...

When I read this I see a large amount of "I played a shuffled playlist and I heard more than one track from the same arbitrary group in the same arbitrary period....this can't *possibly* be "true random shuffle".

When what these people don't seem to realise is that yes, it is "random"..."truly random", and what they fail to realise is that because of that fact there is absolutely nothing stopping the tracks from not being grouped into similar artists/albums/etc out of sheer coincidence...because there's nothing that makes sure that this can't happen...it's random.

The people arguing for "true shuffle" seem to really want a precisely shuffled list whereby they will hear no two tracks from the same artist/album/etc in the same arbitrary period...ever.
As for some reason this is deemed to be "not random enough".

The point I feel has been missed here is that because of the fact that it is entirely random, it's absolutely possible that you could "shuffle" a playlist into the exact same order you began with...so there would be NO perceived difference at all when it is actually a different playlist and randomness simply presented you with the same order because it has no reason to NOT do that.



[St.]

[Llorean]

I think we covered that point back with

If you reshuffle each time, luck could present you with all of his songs, in order, as the first things you hear every single time. Or you could never hear them. That's why it's called "random."

though not quite in as much detail.

I'd also like to make it clear that nobody here is denying the possibility of a bug that causes non-random playback or stacked playlists. Just that evidence so far indicates that playlists are randomly shuffled and that some hard data about exactly what's happening, and how these not-entirely-random playlists are created would be necessary.

For example, a bug may not present itself when a playlist is created then shuffled, but may when a collection is recursively inserted with "insert shuffled." And even then it may only occur if the library is above or below a certain size, or is organized a certain way. This of course *should* never happen, or isn't expected to, but because of the number of permutations involved in library composition and playlist creation it's more or less impossible for someone to test everything, which is why we keep saying "if you're experiencing this, please provide data so that it can be investigated."

Saying "I think it happens but am only chiming in to support this view" doesn't really accomplish anything, and not yet providing data (for whatever reason - time available, a fear it's only asked for to try to prove you wrong, or whatever) also means the problem will remain unsolved if it exists.

You don't even necessarily have to analyze it yourself. Create a shuffled playlist, save it, and repeat. Document how you do it, and do it a few dozen times (an effort of five minutes or less) and then .zip them up and upload them somewhere for someone else to do the analysis on. At the very least it's a start.

And, if you do look into it and find that it's actually random, please post it here so that we're not left with a lingering doubt that there may be some obscure bug out there messing up random playback in some use case or other.

[AlexP]

I'd also like to make it clear that nobody here is denying the possibility of a bug that causes non-random playback or stacked playlists. Just that evidence so far indicates that playlists are randomly shuffled and that some hard data about exactly what's happening, and how these not-entirely-random playlists are created would be necessary.

Exactly so.  Of course there could be a bug, but if there is and it is to be investigated and then fixed, we need actual proof rather than feelings and impressions.