A dataset is a structured collection of related records or values used to answer business questions or train models. Choosing the right data is only the beginning: cleaning, parsing, normalization, and verification determine whether the resulting analysis deserves trust. section 4.0
Treat datasets as engineering inputs rather than unquestioned facts:
- Select data that actually represents the business question, such as identifying which manufacturer dominates a market. section 4.0
- Clean, parse, and normalize the data before performing statistical analysis. section 4.0
- Verify generated analyses carefully because flawed data logic can produce plausible numbers and charts instead of obvious failures. section 4.0
- For supervised learning, separate feature values into X and the predicted label into y, preserving row-by-row correspondence. section 4.5.1
Predicting a drug price
A pharmacy dataset can place dosage, multi-salt status, salt name, manufacturer, and dosage unit in X, while price becomes the label in y. Each feature row must match the price at the same index. section 4.5.1
Table 4.3 Difference between X and Y datasets
|
Term
|
Traditional Name
|
Math Symbol
|
Shape
|
Description
|
|
Features
|
Independent Variables
|
X
|
Matrix (N rows * M columns)
|
The "clues" (Dosage, Salt Name, etc.)
|
|
Label
|
Dependent Variable
|
y
|
Vector (N rows * 1 column)
|
The "answer" you want (Price)
|
Compare X and y by role, mathematical symbol, shape, and the distinction between predictive clues and the desired answer. section 4.5.1





























