Fix index dimension mismatch in TensorNetwork construction

[Todorbsc]

Closes #155

[Todorbsc]

@jofrevalles @mofeing the dimension mismatch seems to be fixed with the verification I implemented in TensorNetwork construction. I "uncommented" the TN dimension mismatch construction test and now it raises the proper exception as well as I tested the Chain example that Jofre commented in the issue, what raises an error now:

julia> mps = Chain(State(), Open(), [rand(2, 3), rand(2, 6, 4), rand(2, 5)])
ERROR: DimensionMismatch: Index D has inconsistent dimension: [3, 6]
Stacktrace:
 [1] TensorNetwork(tensors::Vector{Tensor{Float64, N, A} where {N, A<:AbstractArray{Float64, N}}})
   @ Tenet ~/QuanticGit/Tenet.jl/src/TensorNetwork.jl:58
 [2] Chain(::State, boundary::Open, arrays::Vector{Array{Float64}}; order::Tuple{Symbol, Symbol, Symbol})
   @ Tenet ~/QuanticGit/Tenet.jl/src/Ansatz/Chain.jl:99
 [3] Chain(::State, boundary::Open, arrays::Vector{Array{Float64}})
   @ Tenet ~/QuanticGit/Tenet.jl/src/Ansatz/Chain.jl:63
 [4] top-level scope
   @ REPL[3]:1

Also, let me know if there's something else needed, specially with what you commented in the issue about the unsafe_region.

[jofrevalles]

Okay, nice! Since we already have tests for @unsafe_region, and these are passing, we can merge this!

[mofeing]

Do you mind benchmarking with and without the safe guards / checks for some large TNs? Like 100-1000 tensors

[Todorbsc]

Do you mind benchmarking with and without the safe guards / checks for some large TNs? Like 100-1000 tensors

I benchmarked for rand(TensorNetwork, 1000, 3).

Without the indices safe region:

julia> @benchmark tn = rand(TensorNetwork, 1000, 3)
BenchmarkTools.Trial: 88 samples with 1 evaluation.
 Range (min … max):  17.460 ms … 560.275 ms  ┊ GC (min … max):  0.00% … 12.05%
 Time  (median):     25.241 ms               ┊ GC (median):     2.20%
 Time  (mean ± σ):   56.874 ms ±  82.945 ms  ┊ GC (mean ± σ):  22.67% ± 24.72%

  █▅▁                                                           
  █████▅▁▅▁▁▅▆▆▅▇▅▁▁▅▁▆▁▁▁▁▁▅▁▁▁▁▁▁▁▁▁▅▁▁▁▅▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅ ▁
  17.5 ms       Histogram: log(frequency) by time       407 ms <

 Memory estimate: 25.09 MiB, allocs estimate: 64188.

With the indices safe region:

julia> @benchmark tn = rand(TensorNetwork, 1000, 3)
BenchmarkTools.Trial: 89 samples with 1 evaluation.
Range (min … max):  23.384 ms … 505.261 ms  ┊ GC (min … max):  0.00% … 15.44%
Time  (median):     30.132 ms               ┊ GC (median):     0.00%
Time  (mean ± σ):   61.098 ms ±  85.135 ms  ┊ GC (mean ± σ):  12.13% ± 14.75%

 █▃ ▁                                                          
 ████▇█▁▅▁▇▁▁▁▆▁▁▆▅▁▁▁▁▁▁▁▁▁▁▁▅▅▁▁▁▁▁▁▁▁▁▁▅▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅ ▁
 23.4 ms       Histogram: log(frequency) by time       485 ms <

Memory estimate: 28.02 MiB, allocs estimate: 103174.

Correct me if I am wrong, but as regularity may influence the times, I would just look at the minimum time: Without the checks 17.460 ms vs. With the checks 23.384 ms.

[Todorbsc]

@mofeing Another benchmark for rand(TensorNetwork, 200, 3).

Using checks:

julia> @benchmark tn = rand(TensorNetwork, 200, 3)
BenchmarkTools.Trial: 400 samples with 1 evaluation.
 Range (min … max):   2.218 ms … 253.500 ms  ┊ GC (min … max):  0.00% … 23.87%
 Time  (median):      3.163 ms               ┊ GC (median):     0.00%
 Time  (mean ± σ):   12.499 ms ±  28.164 ms  ┊ GC (mean ± σ):  31.62% ± 21.23%

  █▄▁                                                           
  ███▇▅▆▅▅▅▄▅▅▅▅▄▅▁▅▁▁▄▄▁▁▁▄▁▁▁▄▄▆▁▁▅▆▁▅▁▄▁▄▄▁▄▁▅▁▁▁▁▁▁▁▄▁▄▁▄▄ ▆
  2.22 ms       Histogram: log(frequency) by time       129 ms <

 Memory estimate: 2.80 MiB, allocs estimate: 19763.

Wiithout checks:

julia> @benchmark tn = rand(TensorNetwork, 200, 3)
BenchmarkTools.Trial: 353 samples with 1 evaluation.
Range (min … max):   1.240 ms …   1.031 s  ┊ GC (min … max):  0.00% …  0.32%
Time  (median):      2.028 ms              ┊ GC (median):     0.00%
Time  (mean ± σ):   14.150 ms ± 65.195 ms  ┊ GC (mean ± σ):  30.34% ± 22.77%

 █                                                            
 ██▅▆▄▅▅▁▁▄▁▁▁▁▄▅▅▅▄▆▄▆▁▁▁▄▁▄▁▁▁▁▄▁▁▁▁▁▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄ ▅
 1.24 ms      Histogram: log(frequency) by time       222 ms <

Memory estimate: 2.84 MiB, allocs estimate: 11968.

Min. time: Checks 2.218 ms vs. No checks 1.240 ms.

[mofeing]

Thanks!

Correct me if I am wrong, but as regularity may influence the times, I would just look at the minimum time: Without the checks 17.460 ms vs. With the checks 23.384 ms.

Well, the min time can also be affected by jitter so I would take the mean which means there is a 20% increase in runtime... but it's still 25ms which is ok.