A greatly streamlined rating learning system, as well as maintenance scripts for both OS X and Windows are the new features of this release.
I have reworked the entire system from the ground up, putting it on a much firmer mathematical footing (before I had a lot of arbitrarily chosen constants). It should learn ratings more swiftly, works better with some of iTunes's limitations, and also requires much less effort to install. The documentation has been rewritten accordingly. Please let me know if I missed anything.
The theory section has been removed as it is now obsolete. If there are requests, it will be rewritten.
I reversed engineered the iTunes smart playlist format just enough to automate the creation of the 29 playlists used in my music system. To make a major change previously took several hours to hand edit all the playlist in iTunes, export them, and splice them together into the three large xml files I need. Now this process takes less than a minute, with a greatly reduced chance of errors. To speed it up even more, the program can also interpret the results of my rating estimate simulation directly, incorporating them into the playlists.
Writing up the theory sections I spotted a few minor errors in rating error detection. It had the effect of being slightly too conservative in detecting flaws in some cases. Took the opportunity to raise the size of the table to 300 so that manual clearing of play and skip counts can be three times less frequent.
Beta testers are now being accepted! Any bugs you spot or suggestions for improvements in any part of the system or documentation are most appreciated!
If you'd like to participate, visit the forum.
1-5 of 5