Data Lab Sri LankaExplore. Understand. Build.

Learning notes

A few ideas worth
taking with you.

Short, practical reading notes to help you think through the next question. Start with the subject that is most useful to you.

Data science / Original reading notes

Start with the question, not the model

A useful analysis begins before the first algorithm.

Imagine a small dataset of daily travel times. Before predicting anything, ask what each row represents, whether the measurement method is consistent and which days might be missing. A summary is only as meaningful as the observations behind it.

Separate a descriptive question, such as what happened, from a predictive question about an unseen observation. If you evaluate a predictive model, keep evaluation data separate from the data used to fit it. Record the question and the limitations alongside the result.

Back to the reading list ↑

Mathematics / Original reading notes

A dot product, one step at a time

A short calculation can reveal the structure behind a larger idea.

Take two vectors: a = (1, 2, 3) and b = (4, 5, 6). Multiply the matching entries, then add them: a · b = 1 × 4 + 2 × 5 + 3 × 6 = 4 + 10 + 18 = 32. Each pair contributes one part of the final number.

For a second example, the dot product of (1, 0) and (0, 1) is zero. These nonzero vectors are perpendicular in ordinary two-dimensional geometry. Try changing one entry and explaining how the calculation changes before using a software library.

Back to the reading list ↑

IT practice / Original reading notes

Make a small project reproducible

A future reader should not have to reconstruct your decisions.

Give a project a short README that describes its purpose, required inputs and the order of the steps. Keep original data separate from derived files, and give outputs names that explain what they contain.

Document assumptions and avoid putting passwords, API keys or private data in example files. When a project needs software dependencies, record the environment used. The goal is a clear path from an input to a result, not a complicated folder structure.

Back to the reading list ↑