Would you consider 'number of listens' & maybe skip count as an extra weighting ?
Well I've been listening to music for about 38 years but only using MC for 5 years(?). My playing tends to be weighted toward my newer albums which would unfairly skew the results in their favour if I counted plays. Also, I listen to a lot of tracks which often have noodly endings and I hit skip forward at that point but can still love the tracks. Downgrading for skips would therefore penalise tracks I actually like.
But it cuts both ways, listen to a track too many times and it ends up worn out but there are those exceptional ones that always sound good, given a long enough period between listens.
This is why I weight for favourite artists and favourite albums. There are indeed tracks which *I'm* worn out on but that I acknowledge as classics in their genre, for their time, for their influence etc. I can therefore downgrade the track I'm bored with but the weightings will bring it back up slightly in acknowledgement of their importance overall.
Also how stable is an album rating wrt to number of listens, if an album's rating did not change too much after several listens is its rating more accurate than one with fewer listens.
I do see this. The more I listen to an album the better the rating gets. New albums tend to fluctuate more than albums I've know for years, decades etc.
What formula do you use to weight for track durations ?
My weighting is quite aggressive for duration. I power multiply the duration of every track by 1.75
Let's take an example. An EP has 2 tracks, duration 2mins and 10mins. I rate the 2min track at 4/10 and the 10min track at 8/10
If I just use averages I get (4+8)/2 = 6/10
But common sense to me suggests that the short track is bringing down the true rating of the EP whose majority of content is rated 8.
If we just weight by duration, we get:
((2mins*4)+(10mins*
)/12mins = 7.33
This is already a much better indication of the overall rating IMO.
However, in my current shceme I use a power rating of 1.75
Which gives us a rating for the EP of:
(2mins**1.75)*4 + (10mins**1.75)*8 =
(3.36mins*4)+(56.23mins*
/ 59.59 = 7.77
So this actually not that different from a simple duration weighting but it continues to pull the average towards the average of the majority of the album and away from the average of the minority.
Currently i can't easily account for track duration as a share of the pie as i do my ratings manually
Yup. My scripts handle all the work for me so it's easy for my to adjust my constants and recalculate. I'd *LOVE* to see MC include more comprehensive math work so we could do this internally.
What about the case where track durations are similar between two albums but the only difference was how many there were ?
Then the power function makes little difference (because all track durations are inflated proportionally) and you end up with result closer to a normal average.
Whether in this case averaging alone gives albums with fewer tracks, a higher score than larger albums with just as many good tracks ?
They will come out very similar, or indeed the same. An album with 2 long tracks, rated 5/10 will have the same average as an album with 25 short tracks all rated 5/10.
Noticed earlier that you use 2 digits after the decimal point.
So when you compare albums that have a rating of 4.7x where x is [0-9], would you say they were all similar or .79 > .78 > .77 etc ?
MC seems to internally use two decimal points for decimal values. In my calculations I use many more, but yes an album with 4.88/10 or 4.87/10 is essentially identical. The decimal points just aid in sorting the results.
I used the latter approach upto the point MC offered album ratings in the library browser. a few versions back and noticed they only use one digit after the decimal point. This got me thinking whether it would simplify things by doing the same and rounding up or down.
eg. 4.7[0-4] rounds down to 4.7 and 4.7[5-9] rounds up to 4.8
For sorting, the more decimal points, the better, IMO. Again, it's no extra work for me as the scripts handle the maths.
Album ratings are great as I now have smartlists that not only display all my albums sorted by rating, but I can now also pull random albums to play based on their ratings, last played etc. And in these smartlists I give more weighting again to higher rated albums so that out of my collection the higher rating an album has the more likely it is to be played. Means I always get a good album to listen to when randomly selecting.
MC rocks! But it could rock a little harder, as we are discussing here
Mark