I feel like the first mistake someone makes is to try to practice "machine learning". If you're going in with that as you're goal you're likely to fail.
Instead, tell a story. Motivate what you're doing with real questions and real data and you'll be driven to do all 5 of these lessons (and many more smart things).
Every time someone comes in with a pet algorithm I cringe a bit. There's certainly an air of everything being "just marbles" and thus having application of every algorithm to every problem, but the real question is rarely about the algorithm—it's about the set up, the cleaning, the story. Even when it's about the algorithm you're actually just trying to tell a better story.
So focus on that.
Figure out what you want to "do ML" before you get too excited about what ML is. It's often really painful and annoying with bug fixing turnaround clocking in the hours or days. It's also some of the prettiest math around and a collection of neat hacks for getting great answers to nigh unanswerable questions.
But it's always about answering a question. Start there.
Instead, tell a story. Motivate what you're doing with real questions and real data and you'll be driven to do all 5 of these lessons (and many more smart things).
Every time someone comes in with a pet algorithm I cringe a bit. There's certainly an air of everything being "just marbles" and thus having application of every algorithm to every problem, but the real question is rarely about the algorithm—it's about the set up, the cleaning, the story. Even when it's about the algorithm you're actually just trying to tell a better story.
So focus on that.
Figure out what you want to "do ML" before you get too excited about what ML is. It's often really painful and annoying with bug fixing turnaround clocking in the hours or days. It's also some of the prettiest math around and a collection of neat hacks for getting great answers to nigh unanswerable questions.
But it's always about answering a question. Start there.