This repository contains the code for reproducing the experiments and figures presented in the paper E-Values Expand the Scope of Conformal Prediction.

The repository is structured into three main folders, each corresponding to one of the three methods presented in the paper: batch anytime-valid conformal prediction (batch-anytime-valid-cp), fixed-size conformal sets (fixed-size-conformal-sets), and conformal prediction under ambiguous ground truth (monte-carlo-cp). Each folder is self-contained and independent of the others.

1. Run the load_dataset.ipynb notebook. This will generate a femnist.csv file.

2. Run the split_dataset.ipynb notebook. This will create 2 files: train.csv (training set) and test.csv (test set).

3. (Optional) Run the model_train.ipynb notebook to re-train the model f. The training weights will be saved in the weights/ folder, and the training history will be stored in the results/ folder.

4. Execute the batch-anytime-valid-cp.ipynb notebook to reproduce the experiments from the paper.

1. Run the load_dataset.ipynb notebook. This will generate a femnist.csv file.

2. Run the split_dataset.ipynb notebook. This will create 2 files: train.csv (training set) and test.csv (test set).

3. (Optional) Run the model_train.ipynb notebook to re-train the model f. The training weights will be saved in the weights/ folder, and the training history will be stored in the results/ folder.

4. Execute the fixed-size-cp.ipynb notebook to reproduce the experiments from the paper.

1. Manually download the two files cifar10h-counts.npy and cifar10h-probs.npy from the CIFAR-10H dataset and place them in the data/ folder.

2. Manually download the file cifar-10-python.tar.gz from the CIFAR-10 dataset and add it to the data/ folder. This step should be automatically performed when running either model_train.ipynb or monte-carlo-cp.ipynb.

3. (Optional) Run the model_train.ipynb notebook to re-train the model f. The training weights will be saved in the weights/ folder, and the training history will be stored in the results/ folder.

4. Execute the monte-carlo-cp.ipynb notebook to reproduce the experiments from the paper.

5. Visualize the results using the plot.ipynb notebook.