search

tbreloff / OnlineAI.jl

Machine learning for sequential/streaming data
Other
34 stars 10 forks source link

Explanation of how OnlineAI works #5

Open hpoit opened 8 years ago

hpoit commented 8 years ago

Hi Tom,

Let's walk through a sample problem to see if I understand how you solve it.

First, I will assume these are the schools of thought to solve it and would like to know if you agree image

From each of these schools of thought, 3 common parts can be found image

So the sample problem is code generation right? Let's make this code generation more specific, of an important but missing or incomplete package in Julia from R. How would a human approach this problem?

OnlineAI would autonomously decide how to best represent, evaluate, and optimize this problem. The user's only job would be to define the objective function (the problem).

This is my take. Tell me if you agree with it. Exciting times!!!

hpoit commented 8 years ago

Perhaps it would be best to divide OnlineAI into three parts:

Representation.jl Evaluation.jl Optimization.jl

or any memorable fantasy names for them in case they already exist.

tbreloff commented 8 years ago

Just so you're aware... OnlineAI is an experimental testbed... Not meant for public consumption. You should see LearnBase.jl, MLDataUtils, and the JuliaML org for the public interfaces we're working on. Any code specific to automatic code generation should be in a new repo.

On Friday, June 3, 2016, Henry Poitras notifications@github.com wrote:

Perhaps it would be best to divide OnlineAI into three parts:

Representation.jl Evaluation.jl Optimization.jl

or any memorable fantasy names for them in case they already exist.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tbreloff/OnlineAI.jl/issues/5#issuecomment-223662282, or mute the thread https://github.com/notifications/unsubscribe/AA492picQgjXfK9Hq11mrYJ8NLHVJ6toks5qIHdzgaJpZM4ItyoM .

hpoit commented 8 years ago

Hey Tom. Thanks for helping me out. I'm not really worried about the interface, but the problem solver itself, consolidated, on one repo. I am unaware of a repo with a clear documentation of distributed efforts to build a universal learner like the one Domingos proposed, apparently also sponsored by Microsoft Research and Sebastian Seung. I'm surprised this hasn't been addressed by the Julia community.