Benchmark speed

[dkoslicki]

Goal: determine performance of MinDivLp.py as a function of:

• [x] support size of vector to be reconstructed
• [x] number of rows in sensing matrix
• [x] number of columns in sensing matrix

Approach: using the data (or modifications thereof, or randomly generated "similar looking" sensing matrices), in pseudo-code:

for num_rows in <some range>:
    for num_columns in <some range>:
        form the sensing matrix A
        for support sizes in <some range>:
            for N random x vectors:
                form y
                time MinDivLP.py
                store timing results
average over N
report performance (plot or tabular form): min, max, median, mean, std dev of time

[cmcolbert]

Results from benchmarking speed: results_N=20.xlsx Log-log graphs of results (support size vs. mean time over varying column counts and large_k values):

[dkoslicki]

@cmcolbert is this log base 10 or log base e, or...?

[cmcolbert]

@dkoslicki This is log base e.