Progress towards a version 1 #135
Conversation
…}` (#1) * Add scalartype parameter to `AbstractTensorMap` * Add scalartype in definition of TensorMap * Add scalartype to AdjointTensorMap * Add scalartype to BraidingTensor * Implement changes for indexmanipulations * Implement scalartype for linalg * implement changes for tensoroperations * implement changes for planaroperations * implement changes for ChainRules * small fixes * Fix typo * Fix missing change * Formatter * Apply scalartype changes to tests * Fix some ambiguities * De-specialize `trace_permute` * Despecialize `_contract!` * Remove unused type-parameter * Despecialize `planaradd!` and `planartrace!` * Small fixes * Change braidingtensor syntax to resolve ambiguities * despecialize some linalg methods * formatter * Add changelog entry * Fix ambiguity on Julia1.6
This changes the behaviour of copy as an instantiator for creating a TensorMap from a BraidingTensor. The rationale is that while this does sometimes happen in Julia Base, this is always in the context of lazy wrapper types, for which it makes sense to not copy the parent and then create a new wrapper. BraidingTensor does not wrap anything, so this definition makes less sense.
* Split Sectors module * convert submodule into its own module * Organize tests * Add precompile statements * Add non-unicode alternative `fusionproduct` for `⊗` * Add `directproduct` for `×` * split "trivial.jl" * Formatter * Change unicode alternatives to `otimes` and `times` Incorporate changes from #aada0b2 * Update some testset descriptions * Resolve method ambiguities * Change some formatting * Import local version of `TensorKitSectors` * Enable TensorKitSectors github action * Switch to local include for CI * Remove `NewSU2Sector` tests from main package * Refactor sectors to group Fsymbol and Rsymbol operations * Update CI Sectors to only trigger when files are changed in that path
* copy entire template and stop using copy(b.V1), copy(b.V2) as they do not work * formatting
- Refactored `TensorMap` constructors to be in line with `Array` counterparts:
```julia
TensorMap{E}(undef, codomain, domain) # + some variants
```
- Added support for `zeros`, `ones`, `rand`, and `randn`.
- Streamlined the interface for these functions, along with `isomorphism`, `unitary` and `isometry` into the form:
```julia
function([T], codomain, domain) # + some variants
```
lkdvos
force-pushed
the
v1
branch
2 times, most recently
from
e47ce79
to
61ddbe8
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #135 +/- ##
==========================================
- Coverage 82.59% 79.94% -2.65%
==========================================
Files 48 42 -6
Lines 5706 4962 -744
==========================================
- Hits 4713 3967 -746
- Misses 993 995 +2 ☔ View full report in Codecov by Sentry. |
* Update TensorOperations v5 compatibility * Manually add unreleased TensorOperations version * Update CI.yml * Revert tensoroperations v5 attempts * Changes to incorporate v5 * Remove manual installation of TensorOperations in docs CI [no ci]
| @assert conjA == :(:N) "conj flag should be `:N` ($conjA)" | ||
| return Expr(ex.head, GlobalRef(TensorKit, Symbol(:_planartrace!)), | ||
| @assert !conjA "conj flag should be disabled" | ||
| return Expr(ex.head, GlobalRef(TensorKit, Symbol(:planartrace!)), | ||
| map(_insert_planar_operations, ex.args[2:end])...) | ||
| elseif ex.args[1] in TensorOperations.tensoroperationsfunctions |
There was a problem hiding this comment.
I am confused by this. Why do tensoradd! & friends are already of the GlobalRef form (i.e. we check ex.args[1] == GlobalRef(TensorOperations, :tensoradd!) but would the other functions in tensoroperationsfunction not yet have the GlobalRef asssignment, and need to have it inserted by this postprocessor? It's old code, so it's probably a question to myself 😄 .
There was a problem hiding this comment.
The other functions in tensoroperations do already have this, except for the ones that get added by different postprocessors (eg free_temporaries). In principle, the postprocessor that adds the GlobalRefs should just come last, but it is already part of the TensorParser() call so it's a bit inelegant 🙃
There was a problem hiding this comment.
Oh right, I see, postprocessor steps that the planar parser adds after the addtensoroperations postprocessor that is by default present in TensorParser.
| else | ||
| return ex | ||
| end | ||
| """ |
There was a problem hiding this comment.
These are not yet used currently? This is in preparation of further changes to the planar parser?
| function TensorMap(dataorf, T::Type{<:Number}, P::TensorMapSpace{S}) where {S<:IndexSpace} | ||
| return TensorMap(dataorf, T, codomain(P), domain(P)) | ||
| end | ||
| for randfun in (:rand, :randn, :randexp) |
There was a problem hiding this comment.
randhaar/randisometry is lost in this process, but that one is anyway slightly different, since it needs to be interpreted at the level of whole blocks rather than elementswise, and as such is also harder to be applied in place.
No description provided.