fix tree construction for large number of nodes

[GiggleLiu]

fix #63

julia> using TensorOperations: BinaryTreeNode
julia> const order = (7, 14, 13, 5, 56,
           25, 26, 58, 29, 57,
           4, 44, 36, 35, 43, 3, 21, 33, 34, 22,
           50, 38, 37, 49, 6, 31, 45, 46, 32, 8,
           47, 54, 55, 48, 12, 16, 40, 39, 15, 10,
           1, 41, 19, 17, 18, 20, 42, 2, 60, 28, 52, 30, 51, 27, 59,
           9, 23, 24, 11, 53)

julia> order2tree(order) = length(order) == 1 ? order[1] : BinaryTreeNode(order2tree(order[1:end-1]), order[end])
order2tree (generic function with 1 method)

julia> tree = order2tree(order)
Contraction Tree: (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((7 ↔ 14) ↔ 13) ↔ 5) ↔ 56) ↔ 25) ↔ 26) ↔ 58) ↔ 29) ↔ 57) ↔ 4) ↔ 44) ↔ 36) ↔ 35) ↔ 43) ↔ 3) ↔ 21) ↔ 33) ↔ 34) ↔ 22) ↔ 50) ↔ 38) ↔ 37) ↔ 49) ↔ 6) ↔ 31) ↔ 45) ↔ 46) ↔ 32) ↔ 8) ↔ 47) ↔ 54) ↔ 55) ↔ 48) ↔ 12) ↔ 16) ↔ 40) ↔ 39) ↔ 15) ↔ 10) ↔ 1) ↔ 41) ↔ 19) ↔ 17) ↔ 18) ↔ 20) ↔ 42) ↔ 2) ↔ 60) ↔ 28) ↔ 52) ↔ 30) ↔ 51) ↔ 27) ↔ 59) ↔ 9) ↔ 23) ↔ 24) ↔ 11) ↔ 53)

[GiggleLiu]

I change the Tuple to BinaryTreeNode to make the type tree simpler.

According to my benchmark, the runtime performance is not decreased, while the compling stage performance is increased significantly. Could you please have a double check? @Jutho

[coveralls]

Coverage decreased (-0.5%) to 76.288% when pulling 26e5282 on GiggleLiu:fixtree into 5283953 on Jutho:master.

[coveralls]

Coverage decreased (-0.5%) to 76.288% when pulling 26e5282 on GiggleLiu:fixtree into 5283953 on Jutho:master.

[coveralls]

Coverage decreased (-0.5%) to 76.288% when pulling 26e5282 on GiggleLiu:fixtree into 5283953 on Jutho:master.

[coveralls]

Coverage decreased (-0.5%) to 76.288% when pulling 26e5282 on GiggleLiu:fixtree into 5283953 on Jutho:master.

[coveralls]

Coverage decreased (-0.4%) to 76.389% when pulling 0013403 on GiggleLiu:fixtree into 5283953 on Jutho:master.

[codecov]

Codecov Report

Merging #65 into master will decrease coverage by 0.61%.
The diff coverage is 21.42%.

@@            Coverage Diff             @@
##           master      #65      +/-   ##
==========================================
- Coverage   79.78%   79.17%   -0.62%     
==========================================
  Files          13       14       +1     
  Lines        1390     1402      +12     
==========================================
+ Hits         1109     1110       +1     
- Misses        281      292      +11

Impacted Files	Coverage Δ
src/indexnotation/optimaltree.jl	93.82% <66.66%> (ø)	⬆️
src/indexnotation/binarytree.jl	9.09% <9.09%> (ø)
src/implementation/lrucache.jl	79.06% <0%> (-0.62%)	⬇️

---

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 5283953...26e5282. Read the comment docs.

[codecov]

Codecov Report

Merging #65 into master will decrease coverage by 0.52%.
The diff coverage is 33.33%.

@@            Coverage Diff             @@
##           master      #65      +/-   ##
==========================================
- Coverage   79.78%   79.25%   -0.53%     
==========================================
  Files          13       14       +1     
  Lines        1390     1403      +13     
==========================================
+ Hits         1109     1112       +3     
- Misses        281      291      +10

Impacted Files	Coverage Δ
src/indexnotation/binarytree.jl	25% <25%> (ø)
src/indexnotation/optimaltree.jl	93.82% <66.66%> (ø)	⬆️
src/implementation/lrucache.jl	79.06% <0%> (-0.62%)	⬇️

---

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 5283953...0013403. Read the comment docs.

[Jutho]

Thanks; I'll take a look asap.

[GiggleLiu]

The test coverage is decreased, since I wrote some untested printing functions for BinaryTreeNode. Should I add some test to show the printing does not throw an error?

[GiggleLiu]

Here is a using case for testing

julia> using TensorOperations

julia> using Random, BenchmarkTools

julia> Random.seed!(2);

julia> net = ((11, 17), (2, 20), (15, 28, 38), (15, 21, 34), (12, 21, 40), (10, 30), (16, 22, 32),
           (13, 22, 38), (12, 20), (2, 27, 33), (5, 16, 37), (3, 26, 40), (6, 24), (14, 24, 36),
            (9, 18, 37), (1, 19, 31), (8, 26), (11, 23), (6, 17, 39), (14, 19, 36), (13, 30),
             (4, 25, 33), (7, 23, 39), (4, 28), (5, 27, 32), (9, 25, 35), (8, 29), (10, 18, 31),
              (7, 29, 35), (1, 3, 34));

julia> unique_tokens = union(net...);

julia> c = Dict(token=>TensorOperations.Power{:χ}(1,1) for token in unique_tokens)
Dict{Int64,TensorOperations.Power{:χ,Int64}} with 40 entries:
  33 => χ
  18 => χ
  1  => χ
  32 => χ
  2  => χ
  40 => χ
  16 => χ
  11 => χ
  ⋮  => ⋮

julia> @benchmark TensorOperations.optimaltree(net, c)
BenchmarkTools.Trial: 
  memory estimate:  227.23 KiB
  allocs estimate:  2944
  --------------
  minimum time:     210.654 μs (0.00% GC)
  median time:      217.766 μs (0.00% GC)
  mean time:        279.853 μs (14.19% GC)
  maximum time:     62.270 ms (99.48% GC)
  --------------
  samples:          10000
  evals/sample:     1

[Jutho]

This looks good; thanks. I'll merge it and tag a new release.

[Jutho]

New version 1.1.0 is up!