Machine learning visualisations

Visualisations About

All visualisations/Unsupervised learning

How does PCA compress data without labels?

Difficulty: Approachable
Estimated time: 5 min

Loading visualisation

Step 1 of 4Choose a bad projection

Inspect the long dashed reconstruction distances on the nearly perpendicular axis.

Collapsing data onto this line discards the direction where the points vary most.

Points projected onto a rotatable one-dimensional PCA axis

Axis angle -54 degrees. Variance 0.21. Reconstruction error 6.90.

Projection angle-54°