2
$\begingroup$

I'm writing a battery of tests (in Python) for the purpose of measuring the speed of my company's different computational instances. The goal is to see how fast different AWS EC2 instances are at running different ML models or common data science tasks.

I'm reaching out to ask if anyone knows of any time-consuming (but realistic) models that can be built using only the standard Anaconda packages and either the built in datasets, or a popular dataset known in the industry?

The goal would be to give end users a sense of how much computational power they need, and I think using some popular data sets or models to compare relative time to completion would be the best way for them to choose the right amount of power for their needs.

$\endgroup$
2
  • $\begingroup$ I run my LSTM models on a bunch of 50 20-page health documents [discharge summaries] via multiprocessing. Pretty much tests the limits of the machine [all cores run at 99+% CPU utilization for atleast half an hr to one hour]. Machine: m4.xlarge. So, probably download a regular text doc, and run an LSTM for a niche problem statement :) $\endgroup$
    – Dawny33
    Aug 18, 2017 at 6:02
  • $\begingroup$ A Very common benchmark for machine performance is to calculate the Fast Fourier Transform however it's not a "Learning" task. $\endgroup$ Aug 18, 2017 at 14:49

1 Answer 1

0
$\begingroup$

MLPerf is a set of benchmarks designed just for that purpose. It tests a variety of common machine learning tasks across a variety of possible hardware.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.