
Introduction
In the real world, data rarely matches textbook definitions and examples. We have to deal with issues such as faulty hardware, uncooperative customers, and disgruntled colleagues. It is difficult to predict what kind of issues you will run into, but it is safe to assume that they will be plentiful and challenging. In this chapter, I will sketch some common approaches to deal with noisy data, which are based more on rules of thumb than strict science. Luckily, the trial and error part of data analysis is limited.
Most of this chapter is about outlier management. Outliers are values that we consider to be abnormal. Of course, this is not the only issue that you will encounter, but it is a sneaky one. A common issue is that of missing or invalid values, so I will briefly mention masked arrays and pandas features such as the dropna()
function, which I have used throughout this book.
I have also written two recipes about using mpmath for arbitrary precision calculations. I don't recommend using mpmath unless you really have to because of the performance penalty you have to pay. Usually we can work around numerical issues, so arbitrary precision libraries are rarely needed.