Manipulating Dates

In most datasets you will be working on, there will be one or more columns containing date information. Usually, you will not feed that type of information directly as input to a machine learning algorithm. The reason is you don't want it to learn extremely specific patterns, such as customer A bought product X on August 3, 2012, at 08:11 a.m. The model would be overfitting in that case and wouldn't be able to generalize to future data.

What you really want is the model to learn patterns, such as customers with young kids tending to buy unicorn toys in December, for instance. Rather than providing the raw dates, you want to extract some cyclical characteristics such as the month of the year, the day of the week, and so on. We will see in this section how easy it is to get this kind of information using the pandas package.

Note

There is an exception to this rule of thumb. If you are performing a time-series analysis, this kind of algorithm requires...