The best thing for starters, I think, is to pick some working code (command-line utility, web app, module, whatever) that deals with something that matters to you, get a clear idea of what it does, and then identify something that it doesn't do yet (or doesn't do well or easily), and enhance/rework/replace it to cover that missing functionality.

It's a great approach because: (a) the goal is a matter of personal interest, (b) you see how someone else has approached things that are related to your goal, (c) you can quickly start to see the results of your efforts (whereas trying to build something new from scratch tends to take longer), and (d) you start by learning one of the most valuable skills in programming: the ability to maintain and use other people's code.