What you’ll learn

• how to turn a business request into a concrete ML problem statement
• supervised vs. unsupervised, regression vs. classification, batch vs. online
• why you pick a performance measure before you train anything
• RMSE vs. MAE, and when to prefer one over the other
• loading a dataset with pandas and taking a first look at its structure

The scenario

Imagine you’ve just joined a real-estate company. Your manager wants a model that predicts a district’s median house value from census data — population, median income, number of rooms, and so on — so a downstream investment system can decide where to buy.

Before writing a single line of model code, you need to answer some questions:

• What’s the business objective? The prediction feeds another ML system, so getting the number right (not just a category) matters.
• What’s the current solution? Experts estimate prices manually today, often off by more than 20%. That’s your baseline to beat.
• Is there a continuous flow of data? No — census data is static, so plain batch learning is fine; no need for online learning.

Answering these turns “predict housing prices” into something precise: this is a supervised, multiple regression, univariate, batch-learning problem — supervised because each row has a known target (the price), regression because we predict a number, multiple because we use several features, and univariate because we only predict one value per district.

  flowchart LR
  A["Business goal:
estimate district prices"] --> B["Supervised?
Yes - labeled data"]
  B --> C["Regression or classification?
Regression - predicting a number"]
  C --> D["Batch or online?
Batch - data is static"]
  D --> E["Pick a performance measure
(RMSE / MAE)"]

Selecting a performance measure

For regression problems, the typical measure is the Root Mean Square Error (RMSE):

RMSE(X, h) = sqrt( (1/m) * Σ (h(x_i) - y_i)² )

RMSE(X, h) = sqrt( (1/m) * Σ (h(x_i) - y_i)² )

• mm is the number of instances you’re evaluating on
• h(x_i)h(x_i) is your model’s prediction for instance ii
• y_iy_i is the true label for instance ii

RMSE corresponds to the Euclidean (ℓ2) norm — it squares errors before averaging, so it punishes large errors more heavily. If your dataset has a lot of outlier districts, the Mean Absolute Error (MAE) — the ℓ1 or “Manhattan” norm — is often a better choice, since it treats every unit of error equally:

MAE(X, h) = (1/m) * Σ |h(x_i) - y_i|

MAE(X, h) = (1/m) * Σ |h(x_i) - y_i|

The higher the norm index, the more a metric focuses on large errors and ignores small ones — which is exactly why RMSE is more sensitive to outliers than MAE.

Loading the data

The dataset is the classic California Housing Prices dataset — one row per census “district,” with columns like longitudelongitude, latitudelatitude, housing_median_agehousing_median_age, total_roomstotal_rooms, total_bedroomstotal_bedrooms, populationpopulation, householdshouseholds, median_incomemedian_income, the target median_house_valuemedian_house_value, and a categorical ocean_proximityocean_proximity column.

import os
import pandas as pd
 
HOUSING_PATH = os.path.join("datasets", "housing")
 
def load_housing_data(housing_path=HOUSING_PATH):
    csv_path = os.path.join(housing_path, "housing.csv")
    return pd.read_csv(csv_path)
 
housing = load_housing_data()
print(housing.head())
print(housing.info())

import os
import pandas as pd
 
HOUSING_PATH = os.path.join("datasets", "housing")
 
def load_housing_data(housing_path=HOUSING_PATH):
    csv_path = os.path.join(housing_path, "housing.csv")
    return pd.read_csv(csv_path)
 
housing = load_housing_data()
print(housing.head())
print(housing.info())

head()head() shows the first few rows, and info()info() gives you row counts, dtypes, and how many non-null values each column has — a fast way to spot missing data. Here, total_bedroomstotal_bedrooms has fewer non-null values than the rest, which flags it for cleanup later.

For the one text column, value_counts()value_counts() shows which categories exist and how common each is:

print(housing["ocean_proximity"].value_counts())

print(housing["ocean_proximity"].value_counts())

<1H OCEAN    9136
INLAND       6551
NEAR OCEAN   2658
NEAR BAY     2290
ISLAND          5
Name: ocean_proximity, dtype: int64

<1H OCEAN    9136
INLAND       6551
NEAR OCEAN   2658
NEAR BAY     2290
ISLAND          5
Name: ocean_proximity, dtype: int64

And describe()describe() summarizes every numerical column (count, mean, std, min, quartiles, max) in one call — a quick way to spot capped or oddly-scaled attributes before you go any further.

🧪 Try It Yourself

Exercise 1 – Load a CSV with pandas

Exercise 2 – Count categories with value_counts

Exercise 3 – Summarize numerical columns

Next

Continue to Creating a Test Set (Avoiding Data Snooping) — before you explore this data any further, you need to set part of it aside and never look at it again until the very end.