Make categories in TODO

TODO.md

@@ -5,40 +5,52 @@ While most of the features are already there, there is still some work to be
 done, mostly around optimizing, but also in feature summarization, feature
 evaluation, and distance metrics.
 
-Feel free to submit a PR editing this list if you have some wishes, and to
+Feel free to submit a PR editing this list if you have some wishes, or to
 ask questions if you want to tackle an item.
 
 ## Actual TODO
 
-- Add a list of dependencies / installation guide for windows and mac
-- bliss: A way to learn a metric with a "user survey" on their own libraries using code from the thesis
-  (probably reuse the interactive-playlist in blissify?)
-- Maybe add playlist functions for single songs as convenience methods?
-- Regularly update the python bindings with the new code
+### New features
+
+- Find a proper "objective" way to evaluate how good bliss playlists are. Probably
+  no better way than making mini bliss playlists and make local surveys so each
+  users can evaluate these playlists.
+- A way to learn a metric with a "user survey" on their own libraries using code from the thesis:
+  in progress here https://github.com/Polochon-street/bliss-metric-learning/.
+- A waypoint feature: go from song1 to song2, both picked by the users, in n songs, without any repetitions between playlist 1 and playlist 2
+- A direction feature ("I want the tempo to go down or stay the same")
+- Publish the genre clustering / evaluation code, and try to enhance it.
+- A "song group" feature (I want to make a playlist that's in the vibe of these n songs [like 4-5])
+  (can probably recycle the "album" feature)
+- Use genre clustering (cf already existing code) to find an appropriate M matrix, and put it as alternative.
+  Hopefully will make playlists not drift
+- Have some presets like "upbeat music" "melancholic music", etc
+
+### "Under-the hood" features
+
 - Check the chroma feature for anomalies (the last 4 numbers look anomalous in most of my cases -
   compare with https://www.audiolabs-erlangen.de/resources/MIR/FMP/C5/C5S2_ChordRec_Templates.html etc)
-- Freebsd support? (see https://github.com/Polochon-street/bliss-rs/issues/60)
-- Try to trim out the crates (it is a bit too big right now)
-- grep TODO and see what can be fixed
 - Better duplicate finding in the playlist module (sometimes songs have different across albums but should have the same footprint)
-- Split library's test module (the file is too big right now)
 - Optimize / lower RAM consumption
 - Take a look at https://nnethercote.github.io/perf-book/ to see if optimizing is possible
 - Find a way to differenciate classic vs contemporary music (new feature? Or better use of existing features?)
-- Publish the genre clustering / evaluation code, and try to enhance it.
   For instance, Cigarettes after sex / Sweet doesn't seem to give similar enough tracks?
-- A waypoint feature: go from song1 to song2, both picked by the users, in n songs, without any repetitions between playlist 1 and playlist 2
-- A direction feature ("I want the tempo to go down or stay the same")
-- A "song group" feature (I want to make a playlist that's in the vibe of these n songs [like 4-5])
-  (can probably recycle the "album" feature)
-- Use genre clustering (cf already existing code) to find an appropriate M matrix, and put it as alternative
-  Hopefully will make playlists not drift
+
+### Maintenance tasks
+
+- Add a list of dependencies / installation guide for windows and mac
+- Maybe add playlist functions for single songs as convenience methods?
+- Regularly update the python bindings with the new code
+- Freebsd support? (see https://github.com/Polochon-street/bliss-rs/issues/60)
+- Try to trim out the crates (it is a bit too big right now)
+- grep TODO and see what can be fixed
+- Split library's test module (the file is too big right now)
 - Improve bliss-python somehow / use it in a small demo project maybe?
   A blissify in python?
 - Investigate what type SAMPLE_RATE is in Aubio - maybe u16 is enough
-- Have some presets like "upbeat music" "melancholic music", etc
 
 ## Done
+
 - Split out ffmpeg (see https://github.com/Polochon-street/bliss-rs/issues/63 and https://users.rust-lang.org/t/proper-way-to-abstract-a-third-party-provider/107076/8)
   - Make ffmpeg an optional (but default) feature 
   - The library trait must be Decoder-agnostic, and not depend on FFmpeg