An introduction to conformal prediction

Slides of presentation: Conformal prediction

Video of the presentation: Leveraging conformal prediction for calibrated probabilistic time series forecasts - Inge van den Ende, PyData Eindhoven 2023

Workshop summary in a blog post

Workshop description

During the workshop we will apply conformal prediction:

1. Add a prediction interval to a point forecast
2. Calibrate a probabilistic forecast (conformalized quantile regression)

Open the notebooks in the workshop folder to make the assignment. You can find the full code in the notebooks in the solutions folder. Both notebooks show how implement that step in the crepes package, the MAPIE package and manually by yourself.

Requirements

Python >= 3.10 and experience with making a regression forecast. In the workshop we take a LightGBM model as an example to make a forecast, but this model is not explained in the workshop.

Usage

• Clone the repository
• Install requirements with pip install -r requirements.txt
• Start jupyter lab and navigate to the workshop folder
• Open the first workshop notebook

Video record

Re-watch this YouTube stream

Credits

This workshop was set up by @pyladiesams and @ingevandenende.