A/B testing with Ambrosia in a Nutshell¶
Imagine that you want to run your own A/B test, and after the product analysis and gathering ideas into a hypothesis, you usually have to go through several routine calculation steps: from collecting and transforming raw data to measuring the statistical significance of the experiment result and confidence intervals construction.
In order to solve the problem of carrying out a large number of calculations using various techniques, in Ambrosia, we have identified the following stages of experiments and provide tools and automation for them:
Process
Raw data aggregation, outliers removal, metric transformation as well as various methods for experiments acceleration. Storable data processing pipelines that can be reused.
Design
Experiment parameters such as effect uplift, groups size, and experiment statistical power are designed using metrics historical data by a theoretical or empirical approaches.
Split
Group split methods support different strategies and multi-group split, which allows to quickly create control and test groups of interest. Currently, only batch data splitting methods are supported.
Test
Tools for the statistical inference are able to calculate relative and absolute effects, construct corresponding confidence intervals for continious and binary variables. A significant number of statistical tests is supported, such as t-test, non-parametric, bootstrap, and others.