Tansman music practice scheduler tool

For this project, I will build an LP-based optimal scheduling tool for music practice using Pulp.

My models will be documented in Models.ipynb.

The corresponding code (such as it is) is in models.py.

Problem and Motivation

Every musician, professional and amateur alike, has faced the challenge of making efficient use of practice time. In the course of learning a long, complex piece of music, we can easily fall into patterns where we know that we're not making the the most efficient use of practice time. One way in which we can make better use of our time is of course planning up front. This project is about building a system for generating a schedule automatically given some information about a set of practice items, the time available and parameters that feed a simple personalized utility function.

The work is motivated by trying to address several types of traps encountered when trying to establish a practice regime:

The "over-commitment" trap. It's easy when confronted with a large piece not to appreciate the amount of calendar time required, especially given the need to take breaks.
The "ineffective repetition" trap. A particular difficult passage often requires a large number of short sessions. Spending too much time at one sitting quickly leads to diminishing returns. It can also lead to the reinforcement of bad habits. A particular engaging and/or easy section can lead to a similar pattern.
The "insufficient reinforcement" trap. In this case, you make progress on a particular section, and move onto another, but fail to revisit the original piece within a sufficient period of time. This can lead to having to re-learn a passage.

The theoretical model motivating this is the "spaced repetition" learning model that underlies the approach taken by tools like DuoLingo. In this project, I won't be implementing spaced repetition training system, but the goal is to draw from some of the same principles.

Data

I'll build a set of models that are used to evaluate the features of the formulation, as well as a small model of an actual piece that I'm learning. This will involve some estimation of the practice time required based on past experience, and subjective assessments of what constitutes a good practice experience. The inputs will include:

Time available per session to practice, and number of sessions each day
An estimate of time needed to master each practice item (e.g., phrase or passage)
A way to related items it's beneficial to practice together in the same session
Overall time available for a set of related items (e.g., a section of music)
Characterization of the intensity, fun, etc. associated with items
Ideal amount of time to wait between practicing each item (for memory consolidation)
Maximum time to wait between sessions practicing each item (forgetting)

Deliverables

In addition to the report and presentation, deliverables will include:

A Python module implementing the underlying model
A Jupyter notebook with examples

Practical application

In imagine that the use of the scheduler will be the context of a work flow that looks something like this:

User breaks up the tasks to be completed into logical groupings. Each grouping represent a set of tasks that the user wants to bring to the next level.
Goal is to reach "next level" of competence for each of the items. For example, this might be memorizing a particular passage, working out an interpretation of a passage, or addressing a particular techncial challenge.
User specifies time available on particular days, descriptions of items, and estimates of time requirements and other features. T
The user generates and schedule and iterates on above.

I'd also want to update the software to allow names to be assigned to items, and to map slots to actual dates and times. This is straightforward. I'd also want to tweak the model to account for different days having different time available, the notion of having days off (maybe a day a week), etc.

References

"Unbounded Human Learning: Optimal Scheduling for Spaced Repetition", Reddy, et al. This is a pretty sophisticated way of doing things. Can I do something using LP or mixed IP that approximates this?
Nice little write up on spaced repetition. http://blog.pickcrew.com/the-science-of-learning-new-languages/
The wikipedia page on spaced repetition. https://en.wikipedia.org/wiki/Spaced_repetition
http://pianopracticeassistant.com/spaced-repetition/
"Practice Planner: A Journal of Goals and Progress". Harvey R Snitkin
"The Art of Practicing". Madeline Bruser.
"On Practicing: A manual for students of guitar performance". Ricardo Iznaola. Not really much more than a pamphlet but a pretty good summary of challenges and approaches. Defines distribution of practice time: Building time, interpretive time, performing time, allocation of time during sessions. Recommends 2-3 hours per day for college-level professional students, and has other useful distinctions in a section on "Time-Allocation of Materials".

kompanek / tansman