Refactor tests to test types and transform functionality separately

[glennmoy]

Note: I wrote the individual testsets for each data type before going around and removing them from each transform. In some cases this removed all the code so I left behind some unit tests specific to the transform.

In a follow up PR or commit I'd like to refactor the transform tests, but that is out of scope for this PR.

[codecov]

Codecov Report

Merging #84 (135e533) into main (5da6ca5) will decrease coverage by 0.00%.
The diff coverage is n/a.

❗ Current head 135e533 differs from pull request most recent head 00376b7. Consider uploading reports for the commit 00376b7 to get more accurate results

@@            Coverage Diff             @@
##             main      #84      +/-   ##
==========================================
- Coverage   99.25%   99.25%   -0.01%     
==========================================
  Files          12       12              
  Lines         135      134       -1     
==========================================
- Hits          134      133       -1     
  Misses          1        1              

Impacted Files	Coverage Δ
src/test_utils.jl	100.00% <0.00%> (ø)
src/one_hot_encoding.jl	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5da6ca5...00376b7. Read the comment docs.

[glennmoy]

Not sure why tests are failing considering #93 (which is branched from this) are passing ...

[bencottier]

LGTM. Nice job. -1000+ lines 🤤

[bencottier]

Does a ManyToOne transform work without dims as well?

[glennmoy]

nope, _preformat expects it not to be Colon()

julia> T = FakeManyToOneTransform()
FakeManyToOneTransform()

julia> A  = ones(4, 3);

julia> T(A)
ERROR: MethodError: no method matching length(::Colon)

[bencottier]

These transforms don't preserve type?

[glennmoy]

it does in general, and reminder that

julia> KA = KeyedArray(ones(3, 5), features=[:a, :b, :c], time=1:5);

julia> KA == ones(3, 5)

but as usual AxisArrays breaks the mould, and any operation on them doesn't return an AxisArray anyway

julia> A
2-dimensional AxisArray{Int64,2,...} with axes:
    :foo, ["a", "b"]
    :bar, [:x, :y, :z]
And data, a 2×3 Matrix{Int64}:
 1  2  3
 4  5  6

julia> A .^ 3
2×3 Matrix{Int64}:
  1    8   27
 64  125  216

julia> A .+ 1
2×3 Matrix{Int64}:
 2  3  4
 5  6  7

another reason to drop it

[bencottier]

I forget, are there issues with AxisArray here? Or are you just not bothering?

[glennmoy]

The dims syntax for AxisArrays expects an Axis type not a Symbol so it's incompatible with our interface unless we call axisdim(A, Axis{:foo}) or something to extract the dimension index.

Something we could do maybe to support this without depending on AxisArrays is:

if hasfield(A, axes)
    dims = only(findall(a -> a == dims || typeof(a) <: dims, A.axes))
end

but I'd prefer to implement that at a later time if we actually need it (and I'd almost prefer to drop AxisArrays than do that).

[glennmoy]

I'm gonna force merge this even though tests aren't passing - I'm not sure what's going on but they are passing over in #93 so they should at least be fixed with those.