Why B is the answer

Turning raw data into more useful model inputs

Better inputs help traditional models learn the signal.

In detail

• Feature engineering turns raw data into more useful inputs.
• Domain knowledge guides this.

See it in different ways

mindmap
  root((ML basics))
    Labeled data
      Classification
      Regression
    Unlabeled data
      Clustering
    Validation
      Cross validation
      Held-out test set

flowchart LR
  S0["Raw data"] --> S1["New features"] --> S2["Model"]

Why the other options are wrong

A. Writing tests for each function in the model code

That is testing.

C. Adding new product features based on model output

That is product work.

D. Tuning the number of GPUs for each training run

That is compute planning.

Learn more

• Our lesson: How transformers work
• Attention Is All You Need