I have to run (possibly) millions of times the function isapprox(A, A_, rtol=1e-4), where A and A_ are two matrices. To improve efficiency, I defined a custom isapprox() function using a more efficient norm: using LoopVectorization function is_approx(A::Array{Float64,2}, A_::Array{Float64,2}, rtol…

Something efficient might look like: function is_approx3(A::Array{Float64,2}, A_::Array{Float64,2}, rtol::Float64) # make sure sizes of A and A_ are the same here (?) s1, s2, s3 = 0.0, 0.0, 0.0 for i in eachindex(A) a, a_ = A[i], A_[i] s1 += (a - a_)^2 s2 += a^2 …

Faster isapprox()

General Usage

jondea July 5, 2023, 10:25am 4

Given that you are comparing norms, can you skip all the sqrts and square your rtol instead?

4 Likes

Topic		Replies	Views
Performance of norm function Performance linearalgebra	21	12486	March 28, 2021
Bug? shuffle() breaks normalisation test Internals & Design bug	8	1230	October 21, 2019
Non-allocating distance function Internals & Design question	6	434	November 3, 2022
Improving function performance - Broadcasting vs Loops Performance	14	2070	August 17, 2021
Summing matrix elements is >1000X slower than summing vector elements General Usage performance	8	1354	April 17, 2017

Faster isapprox()

Related topics