The Five Types of Data You Will Meet

In a CSV file, everything looks like text. But when you load it into Python or a database, each column gets a type. Get the types wrong and your model will behave bizarrely, or refuse to train at all.

The five you must know

Numeric: continuous vs. discrete

Continuous numbers can take any value (height: 170.5 cm). Discrete numbers are whole counts (students in class: 27). Machine learning models treat these differently. A continuous model might predict 27.3 students, which is meaningless.

Categorical data, disguised as numbers

Colors like red, blue, green are categories. Sometimes datasets encode them as 1, 2, 3 for space. But computing an average category (1.8) is nonsense. Models need to know these are categorical, not numeric.

import pandas as pd

df = pd.read_csv('students.csv')

# Force zip code to stay a string
df = pd.read_csv('students.csv', dtype={'zip_code': str})

# Parse a date column properly
df['birthday'] = pd.to_datetime(df['birthday'])

# Mark a column as categorical
df['grade'] = df['grade'].astype('category')

print(df.dtypes)

The big idea: types are the skeleton of a dataset. Get them right at load time, and everything downstream gets easier. Get them wrong, and you will spend hours chasing phantom bugs.