Goals and collaboration on tasks

[gbohner]

In order to facilitate working with multiple project members we need to first:

1. Create a design document, that outlines APIs between the various parts of the package.
2. Create a working prototype (complete with installation and examples)

Then individual tasks could be assigned as:

1. Integration into julia package manager ( make it work with Pkg.Clone() for now, then think about optional dependencies, when requesting model implementation packages and adding them on the fly ).
2. Extension API and guidelines (incorporating new libraries as sub-models)
3. Model composition / tuning implementation

[gbohner]

Design decisions to discuss:

1. Data container (agnostic to specifics)
2. Task descriptions / Composition
3. Meta-learning (hyperparameter setting)
4. Model composition
5. Evaluation and diagnostics (simple set of available metrics, benchmarks, diagnostics plots)

[gbohner]

See the poster for a bit more details: https://github.com/alan-turing-institute/mlj/blob/master/material/MLJ-JuliaCon2018-poster.pdf

[dominusmi]

Design decisions to discuss:

1. Data container (agnostic to specifics)
2. Task descriptions / Composition
3. Meta-learning (hyperparameter setting)
4. Model composition
5. Evaluation and diagnostics (simple set of available metrics, benchmarks, diagnostics plots)

I think each of these deserves an issue otherwise it's going to be a bit of a mess on a single thread. We can also better set priority being set as independent

[ablaom]

Agenda for Machine Learning in Julia Kickoff meeting

• introductions; expressions of special interests within the project

• clarify the status of the mlj repo: what bits of code are known to be broken, etc

• to further that end, make a plan for getting some basic test
  code into "runtests.jl", and setting up Travis

• clarification of protocols and responsibility for managing the repo

• determine if there are any known obstacles to moving to Julia 0.7

• field feedback on Anthony's proposal for the package interface spec.

• draw up a list of other immediate priorities and tasks and,
  determine who will take responsibility for what.

• time permitting, a discussion of some intermediate-level design aspects:

  • lazy loading versus automatic loading of packages/interfaces

  • learning networks (aka pipelines, composite learners)

  • agnostic data containers

  Please take a look at this idea for conceptualization learning networks as "dynamic data" and this suggestion for supporting multiple data containers. Both suggestions
  are implemented in this proof-of-concept repo.

Anyone want to add something?