To learn how to create a SageMaker notebook instance click here
Definition: A/B testing is a way to compare two versions of something to find out which performs better. It is one of the simplest forms of a randomized controlled experiment.
How it works: the test works by showing to two groups of users, assigned at random, different versions of a product, drug, site, etc., and determining which version has most successful metrics (previously chosen according to the goals of the experiment).
Technical characteristics:
- Randomized controlled experiment.
- One version is the control and the other is the treatment. The treatment is a new version that we want to figure out if it performs better than the old one (in clinical trials the control can be a placebo).
- One has to estimate the sample size to achieve statistical significance.
- Blocking technique should be used whenever necessary as a means of accounting for certain biases that may be found in any group to maintain randomness in our sampling.
You can find the dataset → here.
- The dataset contains information about almost 300K users that were involved in a A/B test. It is an
unpaired
dataset. - Features:
- user_id: unique identifier for each user.
- timestamp: associated date and time for each visit to the website by a given user.
- group: the category a user was grouped into pre-A/B test (control or treatment groups).
- landing_page: the page that was displayed to a user when they visited the company website (new_page or old_page).
- converted: whether a user converted or not (0 or 1); Note: Users in the control group ought to be displayed the old page, while those in the treatment group ought to see the new page.
- Radical redesigns are best tested using an A/B experiment, while
multivariate
tests indicate how various user interface elements interact with each other and support incremental improvements to a design.