Stub:
High level over view of how user creates DRMs (which are actually wrappers around underlying bindings data structure)
How Samsara gives R-Like syntax to these DRMs with operations like drmA.t %*% drmA
. How the spirit of this is to let
practitioners quickly develop their own distributed algorithms.
Here we’ll talk a bit how the user can write distributed bindings for any engine they wish, how they must implement a few linear algebra operations on the distributed engine in question.
How in JVM based distributed engines, computations happens at JVM on node, native solvers tell application how to dump out of JVM and calculate natively, then load back into JVM for shipping.
How algos like dssvd dspca dqr, etc make back bone of algos framework.
Mahout’s long legacy as leader in Reccomenders in big data, and what is available today.
How we recognize that not everyone wants to re-invent K-means and linear regression so we are building up a collection of common and essoteric algorithms that will come ‘pre-canned’
How these are legacy but still exist.