Skip to content

HOW IT'S MADE

How it works

Small models on silly topics. The idea is a joke, the work behind it is real.

  1. 01

    Human behavior

    Something small, silly and very relatable.

  2. 02

    Data

    Made-up data with a bit of noise.

  3. 03

    Features

    Raw inputs like time and apps turned into numbers.

  4. 04

    Model

    Logistic regression or random forest, whichever did better.

  5. 05

    Prediction

    A probability, right on the page.

Where the data comes from

There's no real dataset for "will I open Instagram", so I wrote small programs that make one. Each has a hidden trait like discipline or spice tolerance, a few sensible rules, and some random noise. The code and the dataset.csv are inside each notebook folder.

What accuracy means here

The scores show how well a model learns the made-up data, not how well it knows you. Each notebook also shows a best possible score, so you can see how close the model got.

Building the models

Every experiment has its own notebook. It makes the data, cleans it, tries logistic regression and a random forest, and keeps whichever did better.

Using them on the site

The finished models are saved as small JSON files and the page reads them directly. The answers are checked against the Python version so both match.

The "Why?" box

For every prediction, one input is changed at a time to see how much the answer moves. The ones that move it most are shown first.

Privacy

Nothing you enter is saved. No accounts, no tracking. The reply experiment only uses made-up data.