Array of custom types/structs?

[datnamer]

Lets say I define a custom type with Cython or Numba's upcoming JIT class capability. Can I fill a vector or array with these types and operate on them quickly, if they support the right operations? (not an object array, but having the type system recognize them).

[izaid]

At some point (hopefully soon), yes, and DyND itself has it's own (currently limited) custom type support.

[datnamer]

Great, thanks! I'm very much looking forward to that capability to use in my research.

[izaid]

What do you work on? I'd be keen to know how DyND (and other Continuum projects) would help you.

[datnamer]

Using individual based models to explore causal mechanisms, in concert with traditional statistical/ml tools.

The other tool that would help is robust custom type support in numba.

By the way, have you seen Julia's recent tribulations and solution to type instability introduced by certain representations of missingness?

Might help inform Dynd design decisions.But this may require a total rewrite of missing data handling, unless Dynd's static typing can help?

[datnamer]

@izaid @insertinterestingnamehere I have some questions for you, when you get a chance please:

In Julia, Heterogenous arrays provide for very slow operations because of type instability. I think they are typed as " Array{Any}".

I noticed that DyND (or datashape) has means of expressing more type information when all array elements are not the same.

What benefits this feature of dynd s provide other than a productivity boost to the programmer form knowing more metadata/type info about the data structure? Can the compiler use it to provide more efficient code over complex heterogenous datastructures, or do we need to convert to singly typed arrays to get fast code? How about dispatch?

Edit:

One more Q: does Dynd have dependent typing?

Thanks!