Building prediction intervals and predictive distributions using conformal prediction
ICP is a computationally efficient variant of the original transductive conformal prediction framework. Like all other models from the conformal prediction family, ICP is model-agnostic in terms of the underlying point prediction model and data distribution and comes with automatic validity guarantees for final samples of any size.
The key advantage of ICP compared to the original variant of conformal prediction (transductive conformal prediction) is that ICP requires training the underlying regression model only once, leading to efficient computations during the calibration and prediction phases. ICP is highly computationally efficient as the conformal layer requires very little additional computation overhead compared to training the underlying model.
The ICP process involves splitting the dataset into a proper training set and a calibration set. The training set is used to create the initial...
