Step 1: Understanding Regression

Regression vs Classification

Machine learning problems split into two types depending on the output:

Rule of thumb: if you can put your output on a number line and care about how far off you are, use regression. If you only care about which bucket, use classification.

The Dataset: Server Response Time

You're monitoring a web server. As traffic increases, response time goes up. Can you predict response time from requests-per-second?

500 synthetic measurements, approximately linear relationship, with Gaussian noise.

Key Code Pattern

One thing you'll see in nearly every ML script: np.random.seed(42). The number 42 is arbitrary — any integer works. Setting a seed locks NumPy's random number generator so every run produces the exact same "random" data. This is called reproducibility: it lets a teammate re-run your code and get identical results, which is essential for debugging, peer review, and audit trails. You'll see random_state=42 on sklearn objects later for the same reason.

import numpy as np
import pandas as pd

# Generate synthetic data with a known relationship
np.random.seed(42)   # lock the random number generator so every run produces identical data
n_samples = 500
requests_per_second = np.random.uniform(10, 200, n_samples)
response_time_ms = 1.8 * requests_per_second + 30 + np.random.normal(0, 15, n_samples)

df = pd.DataFrame({
    'requests_per_second': requests_per_second,
    'response_time_ms': response_time_ms
})

print(df.shape)          # (500, 2)
print(df.describe())     # summary statistics

Exploratory Data Analysis (EDA)

Before fitting any model, always inspect your data:

1. Shape — how many rows and columns?
2. Types — are columns numeric or categorical?
3. Summary stats — mean, min, max, standard deviation
4. Missing values — any NaN entries?
5. Scatter plot — does it suggest linearity?

These steps catch problems early and tell you whether a linear model is appropriate.