features for beta, not to be started till after the alpha

[cartazio]

These are just some near term goals for after the initial alpha release

Array tooling pieces

• More fancy formats
  • Symmetric (padded and packed)
  • Hermitian (padded and packed)
  • Banded (all the various formulation)
  • Triangular (Padded and Packed)
  • Blocked Formats (and associated blocked reads and writes and custom simd as applicable)
• Larger than ram Off heap vectors, both MemMapped to disk and ephemeral
• Quadrant decomposition machinery (needed for Morton order, Symmetric formats)
• Dense Rectilinear projection / slicing machinery (needed to express general rectilinear slicing on arbitrary rank dense row/column major arrays, plus the rank varying static projections)
• index queries for valid starting locations (some of this is sneaking into the alpha, but not much )

Generalization pieces

• generalized Array/layout classes to support sparse/dense time series indexed arrays

numerical computation tooling

• Linear solvers
  • starting with textbook / blas standards
  • sparse friendly direct/iterative solvers
• Least Square solvers
• Eigenvalue / Eigenvector solvers
• Preconditioners

sparse numerical tooling

• see above, but trickier and sparse friendly, though some prelim support snuck into the alpha

good test coverage

• should aim for 100% coverage wrt quickcheck/unit tests. For precision sensitive routines also be sure to test on inputs known to trigger problems with incorrect routines

[cartazio]

blocker: need to have randomv2 out before doing the right approach on some general algorithms, shouldn't be too bad

[ocramz]

I know full well that nested IntMaps are not efficient, however they served well as a scaffolding for writing algorithms. My plan for the future releases is to rely on typeclasses completely and substitute the low-level data structures for more efficient ones. This said, it'd be great if you wrote use-cases and documentation for `numerical`, so that I could see how to build on top of it

…

On Tue, Mar 14, 2017 at 6:31 PM, Carter Tazio Schonwald < ***@***.***> wrote: blocker: need to have randomv2 out before doing the right approach on some general algorithms, shouldn't be too bad — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFoRqL7TD9bUueEEMcMrhzge2iilb7QRks5rls7VgaJpZM4B8BoA> .