Despite using AUROCs for years, I could never develop an intuition for what they actually meant. So I made
a cute visualization where you can generate a population and visually see what various AUROCs mean.

This lets you explore AUROCs by playing with a population of little blocks that come in two types: (positive) and (negative).

You can generate a population of these little blocks, and see the AUROC, treating the height of each block as the score and the type as the label (with as positive and as negative). The AUROC is calculated via the frequency that a type has a higher value/height compared to a type. Each population is sampled from gaussians with a type-specific mean and a shared standard deviation.

Generate new population:

You can sort the blocks by three styles:

Population AUROC: Undefined%

This lets you explore AUROCs by playing with a population of little blocks that come in two types: (positive) and (negative).

You can generate a population of these little blocks, and see the AUROC, treating the height of each block as the score and the type as the label (with as positive and as negative). The AUROC is calculated via the frequency that a type has a higher value/height compared to a type. Each population is sampled from gaussians with a type-specific mean and a shared standard deviation.

Mean 0 | (25) | Mean 1 | (30) | |||

Standard Deviation | (5) | N | (250) |

Generate new population:

- With above settings
- With ≈ AUROC:

You can sort the blocks by three styles:

- — the prevalence of in the top half of the population compared to is a cue for the AUROC.
- — this is just a look at the joint population.
- — this shows the two populations sorted first by type and then by height.

Population AUROC: Undefined%