First Prototype of this package

[zhanglw0521]

This PR aims to create a first prototype (a lux chain actually) that maps a configuration ${x_i}_i$ to the corresponding $\mathcal{B}$ basis $[B^0, B^1, ..., B^L]$.

Currently, the tests pass, but the structure is still a little bit off compared with what we want. Still working on it...

[zhanglw0521]

The test passed well locally, but doesn't work here. This first error looks a little bit weird to me, which seems to indicate that RepLie has not yet been registered? Or is it just something wrong with my Project.toml?

[zhanglw0521]

All local tests passed.

This PR seems to have (almost) all functionalities that were shown on the very right part of the whiteboard yesterday. But it still cannot pass the tests on git(cf. the above comments).

[CheukHinHoJerry]

Seems that we don’t have a registry in CI test that contains RepLieGroups.jl ?

[cortner]

Don't worry about CI yet. I'll review asap.

[cortner]

Jerry - we might have posted at the same time. You are right we need to add the correct registry.

[cortner]

Would you Kind adding the fix?

[CheukHinHoJerry]

CI test for the code runs now but it seems that incorrect version of P4ML is used?

[ Info: Tesing L = 0 O(3) equivariance
Equivariance: Error During Test at /home/runner/work/EquivariantModels.jl/EquivariantModels.jl/test/test_equivariance.jl:27
  Test threw exception
  Expression: F(X) ≈ F(QX)
  MethodError: no method matching _valtype(::Polynomials4ML.MonoBasis, ::Type{Float64})
  
  Closest candidates are:
    _valtype(::Polynomials4ML.AtomicOrbitalsRadials, ::Type{<:Real})
     @ Polynomials4ML ~/.julia/packages/Polynomials4ML/1YAOV/src/atomicorbitalsradials/atomicorbitalsradials.jl:27
    _valtype(::Polynomials4ML.OrthPolyBasis1D3T{T1}, ::Type{T2}) where {T1, T2}
     @ Polynomials4ML ~/.julia/packages/Polynomials4ML/1YAOV/src/orthopolybasis.jl:41
    _valtype(::Polynomials4ML.STO_NG, ::Type{<:Real})
     @ Polynomials4ML ~/.julia/packages/Polynomials4ML/1YAOV/src/atomicorbitalsradials/sto_ng.jl:11
    ...

Also, do we want to add tests for HyperDuals too?

[zhanglw0521]

CI test for the code runs now but it seems that incorrect version of P4ML is used?

Thank you! I used [email protected]. But maybe I should adapt to the latest rather than make the code adapt to my environment.

[CheukHinHoJerry]

Thank you. And also I suggest to run the tests with julia >= 1.9.2 locally since there were some issue with version less than that before.

[zhanglw0521]

I am trying to adapt this PR to the latest version of P4ML. But I am not sure whether or not there is an issue with P4ML @0.2.2:

When I try

using Polynomials4ML: _valtype
totdeg = 4
Rn = legendre_basis(totdeg)
_valtype(Rn,Float64)

it works well and returns Float64, but when I try

Rn = MonoBasis(totdeg)
_valtype(Rn,Float64)

_valtype has no definition of this kind of operation.

Maybe adding

_valtype(b::MonoBasis, T::Type) = T

to P4ML would resolve this?

[CheukHinHoJerry]

Yes I think this is a bug in P4ML. I will create a PR on that side to fix this.

[cortner]

Testing with duals and hyperduals is useful but not urgent. Maybe an issue?

[CheukHinHoJerry]

bump