Measure training time (time.perf_counter)

The 'benchmark' part: how long does model.fit() take? Use time.perf_counter() for high-precision timing.

script.py v3.1

import time
import numpy as np
from sklearn.linear_model import LinearRegression, Ridge
from sklearn.ensemble import RandomForestRegressor
from sklearn.datasets import make_regression

# Generate a bigger dataset so timing is meaningful
X, y = make_regression(n_samples=10000, n_features=20, noise=15, random_state=42)
print(f'Dataset: {X.shape[0]} samples, {X.shape[1]} features')

# ── Benchmark function: times the fit() call ──
def benchmark_training(model, X, y, name):
    start = time.perf_counter()
    model.fit(X, y)
    end = time.perf_counter()
    elapsed = end - start
    print(f'  {name:30s}  {elapsed*1000:8.1f} ms')
    return elapsed

print('\nTraining benchmark (3 models):')
print('-' * 50)

t1 = benchmark_training(LinearRegression(), X, y, 'LinearRegression')
t2 = benchmark_training(Ridge(alpha=1.0), X, y, 'Ridge')
t3 = benchmark_training(RandomForestRegressor(n_estimators=50), X, y, 'RandomForest(50 trees)')

print('-' * 50)
print(f'Fastest: LinearRegression ({t1*1000:.1f} ms)')
print(f'Slowest: RandomForest ({t3*1000:.1f} ms)')
print(f'RF is {t3/t1:.1f}x slower than LinearRegression')

# ── Why perf_counter not time()? ──
print('\nWhy perf_counter?')
print(f'  time.time() resolution:        ~1 ms')
print(f'  time.perf_counter() resolution: ~0.001 ms')
print(f'  perf_counter is monotonic (never goes backwards)')

Python output click Run to execute

Click "Run" to execute the code on the left.
The output will appear here.

🧠 How it works

time.perf_counter() is the right tool for measuring durations. Higher resolution than time.time() (which is for wall-clock time). Monotonic — never goes backwards (important on multi-core systems). The benchmark function wraps fit() with start/end timers. LinearRegression is fastest (closed-form solution). RandomForest is slowest (builds 50 trees). This is what 'benchmarking' means — measuring TIME so you can compare models.

✏️ Try these changes

Change n_samples from 10000 to 100000
→ 10x more data → all models take longer
Change n_estimators from 50 to 200
→ RandomForest takes 4x longer
Add from sklearn.svm import SVR; benchmark_training(SVR(), X[:1000], y[:1000], 'SVR')
→ SVMs are slow on large data
Wrap the timing in a loop (run fit 5 times, take average)
→ more stable measurement

← Previous lab

Run predict() on a loaded model

Next lab →

Measure inference time + throughput