Thursday, 4 August 2016

A/B Testing vs Multi-Armed Bandit

Interesting article arguing that the Multi-Armed Bandit approach is better than A/B testing.

One of the comments suggests that A/B testing gets a faster result (for a given sample size), allowing you to turn off the weakest options sooner. So it could still be better for a quick experiment.

http://stevehanov.ca/blog/index.php?id=132

No comments: