Lets say I want to create a radio station that plays Hindustani classical music (wiki) but selects the right raaga based on local time and the season. The notes of the raagas are by definition correlated to natural moods. So basically it needs a library of notes labeled by time etc; the model then needs to listen to the music library and select tracks that match certain criteria.
Is this a machine learning problem?
For supervised learning, I understand that the criteria is that the answer needs to be known.
I have read @jeremy 's article about the drive train approach for data products and totally agree with looking for the ‘objective’; is it useful to create a more structured process flow for problem abstraction for the datascientist rather than the user’s point of view?
Did some more reading on trying to come up with the problem statement.
Found a rather long post in a google group that discusses the fundamental differences between hindustani classical notes and western classical notes; the scale in Indian classical music is relative to the frequency of the Sa used by the performer, vocal or instrumental. Typically it is tuned to the frequency of the tanpura used and as we know every tanpura or harmonium is tuned differently, but according to a gharana. This leads to the second realization, that the concept of pitch, which is a psychoacoustical attribute of sound, is what distinguishes what we like and don’t like in Indian music, e.g., we commonly have preferences for one artist over another even when they are playing the same raaga. https://groups.google.com/forum/#!topic/rec.music.indian.classical/kTc9ToSH3kA
So it seems like ultimately the model will need to recognize and distinguish between compositions and performances based on pitch (wiki). There are online tools that can be used to generate sounds of particular frequencies and then alter their pitch.