The Data Science Workflow
From question to decision.
Many data projects fail for the same reason. Someone dived into the data, without knowing what decision it was meant to support. They produced beautiful charts that nobody used. A simple workflow prevents this.
- Projects fail the same way
- Beautiful charts, no decision
- A simple workflow fixes it
The workflow
Start with a question, tied to a real decision. Get the data. Clean it, which often takes most of the time. Explore it, with charts and summaries. Analyse or model it. Then communicate your result. And finally, watch what happens, which usually raises a new question. Data science is a loop, not a straight line.
Good questions
A vague question is, how are sales? A good one is, which three cafes lost the most profit last quarter, and why? A good question names a metric, a time, and a decision. Before you begin, ask, what would we do differently, depending on the answer? If the answer is nothing, then skip the analysis.
Your tools
You'll use a small set of tools. Python and pandas for data. SQL, the language of databases. Matplotlib and seaborn for charts. Scipy and statsmodels for statistics. And scikit learn for models. They're all free. A notebook keeps your code and charts together.
Summary
Start with a decision, follow the loop, and use a few great tools. That's your map. Next, we go deep into the tool that you'll use every single day. Pandas.
Key takeaways- Tie questions to decisions
- Clean, explore, analyse, communicate
- It's a loop