Data Types for ML
A spreadsheet, a cat photo, a stock chart: three datasets, three shapes. Scramble each one's insides and only the spreadsheet, rows like price=340k, survives while the other two turn to noise, telling you what you hold.
- ▸Shuffle a table's rows and it is the same dataset; scramble a photo and the picture is gone.
- ▸The same house table predicts a price or predicts whether it sold, depending only on which column is the target.
- ▸A table's rows shuffle freely; a stock chart's order is load-bearing, so the next topic splits it by time.
Four datasets land on your desk: a table of house sales, a folder of cat photos, a stack of product reviews, a chart of a stock's daily price. None of them look alike, and before you pick a model, or even decide what you are predicting, one small question sorts them: scramble each one's insides, and does it survive?
Shuffle the rows of the house table and you have the exact same dataset back, because each row stood on its own. Shuffle the pixels of a photo and the cat is gone. Shuffle the words of a review and it is gibberish. Shuffle the days of the stock chart and the trend vanishes. That single test splits the four into one shape whose internal order you can destroy for free, Tabular data, and three where the arrangement was the information: Image data, Text data, and a Time-series.