Input dimension checking when evaluating surrogates

[archermarx]

Fixes #337. Adds a new utility function _check_input_dimension which is called for all surrogates before evaluating a point. Provides a helpful error message if there's a dimension mismatch.

Example

using Surrogates

lb = 0.0
ub = 4.0
x = [1.0,2.0,3.0]
y = [4.0,5.0,6.0]
my_rad = RadialBasis(x, y, lb, ub, rad = linearRadial())

julia> my_rad(Float64[1.5])
4.5

julia> my_rad([1.0, 2.0])
ERROR: ArgumentError: This surrogate expects 1-dimensional inputs, but the input had dimension 2.
Stacktrace:
 [1] _check_dimension
   @ C:\Users\thoma\Surrogates.jl\src\utils.jl:11 [inlined]
 [2] (::RadialBasis{Surrogates.var"#1#2", Int64, Vector{Float64}, Vector{Float64}, Float64, Float64, Transpose{Float64, Vector{Float64}}, Float64, Bool})(val::Vector{Float64})
   @ Surrogates C:\Users\thoma\Surrogates.jl\src\Radials.jl:159
 [3] top-level scope
   @ REPL[39]:1

julia> my_rad(Float64[])
ERROR: ArgumentError: This surrogate expects 1-dimensional inputs, but the input had dimension 0.
Stacktrace:
 [1] _check_dimension
   @ C:\Users\thoma\Surrogates.jl\src\utils.jl:11 [inlined]
 [2] (::RadialBasis{Surrogates.var"#1#2", Int64, Vector{Float64}, Vector{Float64}, Float64, Float64, Transpose{Float64, Vector{Float64}}, Float64, Bool})(val::Vector{Float64})
   @ Surrogates C:\Users\thoma\Surrogates.jl\src\Radials.jl:159
 [3] top-level scope
   @ REPL[40]:1

Benchmarking reveals this to come at a very small cost:

Before
julia> @benchmark my_rad([1.5])
BenchmarkTools.Trial: 10000 samples with 855 evaluations.
Range (min … max): 143.392 ns … 10.876 μs ┊ GC (min … max): 0.00% … 97.94%
Time (median): 146.784 ns ┊ GC (median): 0.00%
Time (mean ± σ): 175.408 ns ± 332.221 ns ┊ GC (mean ± σ): 7.69% ± 4.01%

▆█▆▅▄▂▁ ▁ ▂▄▄▂ ▁ ▁
████████▆▆▇▅▅▅▆▅▅▅▄▅▆▅▄▄▃▅▅▅▅▅▅▅▅▄▄▅▅▅▅▄▇█████████▇▆██▇▆▅▄▅▄▄ █
143 ns Histogram: log(frequency) by time 247 ns <

Memory estimate: 144 bytes, allocs estimate: 3.

After
julia> @benchmark my_rad([1.5])
BenchmarkTools.Trial: 10000 samples with 825 evaluations.
Range (min … max): 151.394 ns … 10.147 μs ┊ GC (min … max): 0.00% … 97.38%
Time (median): 155.394 ns ┊ GC (median): 0.00%
Time (mean ± σ): 196.150 ns ± 340.000 ns ┊ GC (mean ± σ): 6.83% ± 3.89%

▇█▄▃▂▂ ▁▁▁ ▂▄▅▅▂▁▁▂ ▁
████████▇▆▅▆▇▇▇▆▆▇▆▇▆▆▆▆▆▅▅▅▇██████████████▆▄▅▅▄▅▆▅▅▆▆▅▆▄▃▅▄▄ █
151 ns Histogram: log(frequency) by time 302 ns <

Memory estimate: 144 bytes, allocs estimate: 3.

[archermarx]

I have removed the tests for GEKPLS for now because it expects matrix inputs instead of vectors of tuples like the rest of Surrogates.jl. When #355 is being worked on, I will revisit it

[codecov]

Codecov Report

Merging #376 (bef6e07) into master (5bd2810) will increase coverage by 0.06%.
The diff coverage is 93.54%.

@@            Coverage Diff             @@
##           master     #376      +/-   ##
==========================================
+ Coverage   78.90%   78.96%   +0.06%     
==========================================
  Files          16       16              
  Lines        2280     2306      +26     
==========================================
+ Hits         1799     1821      +22     
- Misses        481      485       +4     

Impacted Files	Coverage Δ
src/GEKPLS.jl	91.20% <ø> (-0.93%)	⬇️
src/PolynomialChaos.jl	0.00% <0.00%> (ø)
src/Earth.jl	87.93% <100.00%> (+0.10%)	⬆️
src/GEK.jl	93.66% <100.00%> (+0.18%)	⬆️
src/InverseDistanceSurrogate.jl	100.00% <100.00%> (ø)
src/Kriging.jl	94.64% <100.00%> (+0.19%)	⬆️
src/LinearSurrogate.jl	100.00% <100.00%> (ø)
src/Lobachevsky.jl	97.24% <100.00%> (+0.05%)	⬆️
src/Radials.jl	88.28% <100.00%> (+0.10%)	⬆️
src/SecondOrderPolynomialSurrogate.jl	100.00% <100.00%> (ø)
... and 3 more

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more