INTERACT FORUM
More => Old Versions => JRiver Media Center 18 for Windows => Topic started by: gvanbrunt on August 17, 2012, 07:31:30 pm
-
I don't mean what it does, but how is the randomization accomplished? The reason I ask is it seems to be somewhat "unrandom". I often see two or three tracks from an album close together together etc. I know randomization can be difficult to accomplish effectively on computers, but is the current implementation optimal?
Or does it work just fine for others and I'm just crazy?
Wait. Don't answer that part...
-
I often see two or three tracks from an album close together together etc.
That's how you know it is random. If you didn't see that (in a large enough sample), you should wonder if it's the result of an algorithm that's intentionally avoiding it. And that, of course, would not be random.
-
I remember a thread from a couple years ago where this was discussed, and Matt confirmed it was truly random, which does result in some 'grouping', as you've described.
However, I think Matt said he would tweak it to try to avoid that, since most people don't really want that when they select random, even though it's 'accurate'.
Perhaps he didn't tweak it, maybe it's been changed again, I can't say, sorry.
I guess it's been 8 years :o
http://yabb.jriver.com/interact/index.php?topic=20600.msg151784#msg151784
-
It isn't random though. I almost never hear tracks together from different albums. I often hear some together from the same album. In a system that is truly random that would not be the case especially when there are more tracks from the artist available than just those from the album.
Also an exported smartlist that I randomize on my phone with other playback software does not do this. I think whatever algorithm is in use is flawed.
-
One other pattern I notice. I often notice some songs it plays much more often than others. On 3 or 4 plays through a smartlist after refreshing it brings up some songs over and over. Others it never plays at all...
There shouldn't be patterns if it is random.
-
This topic is an oldie, but a goodie.
The thread Justin linked to is still relevant today:
http://yabb.jriver.com/interact/index.php?topic=20600.msg151784#msg151784
Random data often has things that look like patterns. That's how random works.
And because users don't really want a true random shuffle, Media Center spreads the same artist around after a shuffle to make it less common to get tracks by the same artist next to each other. This makes the data less random in a mathematical sense, but more random in a common-sense way.
-
I often notice some songs it plays much more often than others.
This is more interesting. I tend to agree in some visceral, psychosomatic, totally non-scientific way.
Case in hand I have a smartlist with my top tracks - 4 and 5 stars, handy at all times. Recently I just further tweaked it putting a year limit and BPM constraints on it. The resultant list had an unexpected high count of tracks with zero no. of plays. Things like these would get anyone startin' wondering - what happen with those tracks before the constrains, why where they never played in the initial, bigger smartlist? But to prove anything is rather difficult. There are many things that could affect the findings; if I ever skipped a track before it was accounted as "played", the whole thing is off its statistical rails.
But I have a simpler question. What gets randomized on a shuffle? One variable (say track name) + added logic to lower probability of same artist OR more variables?
-
And because users don't really want a true random shuffle...
Part of what is being questioned is not the order of the list but the content. If Shuffle isn't random, then it doesn't provide a reliable means for selection by shuffling a selection and then taken the first N items. That's a technique commonly used in smartlists, but now I wonder if internal selection tools (limit, mix, etc.) use the same method. If so, this might explain results that are not only not random, but contrary to any common-sense notion of a 'random' selection. Things like...
One other pattern I notice. I often notice some songs it plays much more often than others. On 3 or 4 plays through a smartlist after refreshing it brings up some songs over and over. Others it never plays at all...
The quasi-random shuffle might be fine for setting the final order of a list, but a truly random function should be used for all other purposes.
-
This is more interesting. I tend to agree in some visceral, psychosomatic, totally non-scientific way.
Case in hand I have a smartlist with my top tracks - 4 and 5 stars, handy at all times. Recently I just further tweaked it putting a year limit and BPM constraints on it. The resultant list had an unexpected high count of tracks with zero no. of plays. Things like these would get anyone startin' wondering - what happen with those tracks before the constrains, why where they never played in the initial, bigger smartlist? But to prove anything is rather difficult. There are many things that could affect the findings; if I ever skipped a track before it was accounted as "played", the whole thing is off its statistical rails.
But I have a simpler question. What gets randomized on a shuffle? One variable (say track name) + added logic to lower probability of same artist OR more variables?
Interesting. The lists I have an issue with are based on 4 and 5 stars as well. I wonder if that has anything to do with the issue.
Matt:
Is randomize applied last, or before the smartlist is processed?