🟩 MSE Loss Visualizer
Those squares aren't decoration — their AREA is the squared error, literally.
📋 How To Use This Lab
Drag
Slope
and
Intercept
— watch a real square grow at every point, sized exactly to (error)².
Press
▶ Optimize
— the line rotates toward best fit while every square visibly shrinks.
Flip
MSE ⇄ L1
— squares become simple line segments; switch to the "With Outlier" dataset to see MSE get dragged around while L1 barely budges.
Try different
Datasets
from the dropdown to see how loss behaves on clean vs noisy vs outlier-heavy data.
ŷ = 0.00x + 0.00
MSE = 0.000
Each square's area = (true y − predicted y)²
Slope:
0.00
Intercept:
0.00
Dataset
Clean Linear
With Outlier
High Noise / Scattered
MSE (Squares)
L1 (Lines)
▶ Optimize
↺ Reset
🟩 SQUARY
Drag the sliders and watch real squares appear — their area IS the loss, not just a metaphor.
🧑🏫 PROF. TORCH
MSE = mean((y − ŷ)²) — literally the average area of every one of these squares.