Keeping up with data — Week 52 reading list
Great visualisations are informative, intriguing, entertaining and action-inspiring. But most of all, they are stories told with data. FiveThirtyEight has made some great ones. The picture above is from their article celebrating the best — and weirdest — charts their visual jurnalists have published in the last 12 months. I enjoy these articles. Firstly, because of the aesthetics and secondly because I often find new angles even to the old well-known stories.
End of the year is traditionally hectic and this one is not any different. Therefore, the list only contains three articles. But what is lacking in quantity is made up in quality.
- Tis the Season…to be Bayesian! When predicting demand for inventory, one often needs to account for effects of seasonality and holidays. Holidays, in particular, are difficult to model as they often occur only once a year and they typically entwine with certain seasons (think Christmas for instance). Stich uses a Bayesian Holiday Model enabling them to robustly capture holiday effects within time series data. The beauty lies in a generic holiday effect function with an ability to describe the shape, length and peak of the holiday effect combined with the Bayesian approach of updating priors. (Stitch Fix)
- The Future of Time Series Forecasting: An opinion piece by Aileen Nielsen about using traditional statistical models and machine learning models for forecasting. Sometimes you want an explainable model with a high accuracy, sometimes you want a good enough forecast applied to thousands of time series. But the main point is that the machine learning and statistical approaches can be combined: either as alternative models assembled together or with one method determining how to set the metaparameters of the other method. Either way, do not forget about the limits and assumptions of forecasts when using them for decision making. (O'Reilly Media @ Medium)
- Stop Burning Money on A/B Tests: A memorable article about how to use A/B tests successfully. Using a very powerful analogy of climbing a mountain — A/B tests can tell you which route to the peak is the fastest. But you need to have comparable conditions for both routes and you need to be climbing the right mountain. A/B tests can’t solve every business problem. But they can help in selecting the best available solution (provided one of them solves the underlying problem). (Daliana (Zhen) Liu @ Entrepreneur’s Handbook)
I’d seen the A/B testing article in my LinkedIn feed. I started reading it but then I got distracted by a car race happening in the Lego corner of our living room. After the race the LinkedIn app refreshed my feed and the article was gone. Sometimes, analogies are helpful not only for easier understanding but they can also play a role of a mnemonic device and a keyword when you need to find an article again without knowing the author or the title.
Merry Christmas, happy holidays and all the best for 2021!