[NDTensors] UnallocatedArrays and UnspecifiedTypes

[kmp5VT]

Description

In this PR I add a FieldType folders where I implement a number of abstract, lazy and unspecified immutable fields which will later be implemented into NDTensors.

Checklist:

• Implement UnallocatedArrays and UnspecifiedTypes
• Create a SetParameter system for the UnallocatedArrays types
• Create rudimentary unittests for these new types

As a recap of our conversation 11/20. I will be working on the following

• Remove AbstractUnallocatedArray
• Try to make functions instance off AbstractFill and keep UnallocatedZeros a AbstractZero.
• Write the discussed constructor UnallocatedX(f::X, alloc::Type{<:AbstractArray}) where X is fill or zero. And have set_alloctype call this function.
• Create arithmetic that preserves the Unallocated type
  • addition and subtraction preserve types
• create full unittest suite for these types

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (4638f7c) 84.03% compared to head (7924683) 53.93%.

❗ Current head 7924683 differs from pull request most recent head f8ed013. Consider uploading reports for the commit f8ed013 to get more accurate results

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1213       +/-   ##
===========================================
- Coverage   84.03%   53.93%   -30.10%     
===========================================
  Files         100       99        -1     
  Lines        8544     8491       -53     
===========================================
- Hits         7180     4580     -2600     
- Misses       1364     3911     +2547     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mtfishman]

I think we could split this into two modules, say UnallocatedArrays and UnspecifiedTypes.

Also note that JuliaArrays/FillArrays.jl#299 is merged so we can use AbstractZeros and AbstractOnes.

[kmp5VT]

@mtfishman It looks like theres an issue on Github where it's having trouble using the internet to download/check the deps. This causes tests to prematurely crash/fail. I have been restarting them here and on Main

[mtfishman]

Looks good to me, thanks!

[kmp5VT]

@mtfishman It looks like Jenkins is failing for ITensorGPU with Julia v 1.6 but I cannot replicate the error. When I look at the error thrown by Jenkins, i.e.

failed: [-0.0009661297423744132, 0.0005560180954419236, -0.0010679817640976697, 0.0005560181043973156, -0.0009661297481295442] ≈ [-0.0009661297423742309, 0.000556018095441646, -0.0010679817589570595, 0.0005560180924686908, -0.0009661297199891616]

The biggest difference in this vector is ~2e-11 which i guess is above the rtol for doubles. Do you have any suggestions for this issue?

[mtfishman]

Increase rtol a bit? Floating point results can be different on different systems.

[kmp5VT]

@mtfishman It doesn't look like it had any issue on main. All jenkins ITensorGPU tests have already passed.

[mtfishman]

Ok, if you saw it happen it may be worth doing anyway since it may mean rtol is just slightly too small to be robust to roundoff.