23 Mar 2022

[research]

mjxmr ^{has published the benchmark results for his Time Series Quick Simulator (tsqsim) tool:}

In order to limit the strain on this quarter’s budget, I resorted to testing against only the following frameworks: statsmodels^{, darts}

The simulator was written for the Monero Research Lab ^{in order to predict the number of near future transactions.}

Below are the results of the benchmark against the (almost) equivallent native C++ implementation of the AutoRegressive (AR) model with lags of 2.

Results

Benchmarked 'py_statsmodels.py' Time taken = 10.922s Ratio Pred2Base     = 0.705 [===       ] (the lower the better) 
Benchmarked 'py_darts.py'   Time taken = 14.307s Ratio Pred2Base     = 0.711 [===       ] 
Benchmarked 'Native C++'    Time taken = 0.144s  Ratio Pred2Base     = 0.68  [===       ]

Benchmark images and steps to reproduce the experiment are included in the Reddit thread^.

According to fellow researcher Rucknium^{, the results show an almost two orders of magnitude speedup.}

To learn more about mjxmr’s work, consult my previous report^{. You can support his continued work on Monero by donating to his CCS proposal.}