Live demos.

Six small interactive tools, each making a concept I work with feel less abstract. All runs in your browser; nothing is sent anywhere. Touch-friendly on mobile.

Linear regression

Click to add points. The line follows.

The pink line is the least-squares best fit. R² shows how well it explains the variance.

slope
intercept
n0
Multi-armed bandit

Pull an arm. Thompson sampling will keep up.

Each arm has a hidden reward rate. You play; the bot uses Thompson sampling. Watch posteriors sharpen as evidence accumulates.

You0 reward in 0 pulls
Thompson sampler0 reward in 0 pulls
K-Means clustering

Drop points. Watch them cluster.

Click anywhere to add a point. Press Step to advance one iteration of Lloyd's algorithm.

K3
iter0
n0
inertia
Confusion matrix · threshold

Drag the threshold. Precision vs recall trade.

A simulated binary classifier scored 300 examples. Move the slider to set the decision threshold and see the confusion matrix update.

0.50
TP0
FP0
FN0
TN0
precision
recall
F1
acc
A/B test calculator

Type two conversion rates. See if it's real.

Two-proportion z-test. Lift, z-score, p-value, and a plain-English call on whether the difference is statistically significant at α=0.05.

Visitors
Conversions
Rate: 5.00%
Visitors
Conversions
Rate: 7.50%
lift
z-score
p-value
significance
Bias–variance tradeoff

Slide the model complexity. Watch the lesson.

A polynomial of degree d fits a noisy sine. Low d underfits. High d memorises the noise. The sweet spot is usually simpler than you'd think.

degree3
train MSE
test MSE
gap