This is the code release accompanying the paper Class-Conditional Conformal Prediction with Many Classes

Citation:

@article{ding2023classconditional,
  title={Class-Conditional Conformal Prediction with Many Classes},
  author={Ding, Tiffany and Angelopoulos, Anastasios N and Bates, 
          Stephen and Jordan, Michael I and Tibshirani, Ryan J},
  journal={arXiv preprint arXiv:2306.09335},
  year={2023}
}

Setup

First, create a virtual environment and install the necessary packages by running

conda create --name env
conda activate env
pip install -r requirements.txt

To make the environment accessible from Jupyter notebooks, run

ipython3 kernel install --user --name=conformal_env

This adds a kernel called conformal_env to your list of Jupyter kernels.

Download the datasets by running

sh download_data.sh

which will create a folder called data/ and download the data described in the following section.

Data description

1. imagenet (4.62 GB): (115301, 1000) array of softmax scores and (115301,) array of labels
2. cifar-100 (0.01 GB): (30000, 100) array of softmax scores and (30000,) array of labels
3. places365 (0.54 GB): (183996, 365) array of softmax scores and (183996,) array of labels
4. inaturalist (6.72 GB): (1324900, 633) array of softmax scores and (1324900,) array of labels

The code for training models on the raw datasets to produce the softmax scores is located in generate_scores/

Running Clustered Conformal

See example.ipynb for an example of how to run clustered conformal prediction.

Reproducing our experiments

Run sh run_experiments.sh to run our main set of experiments. Run sh run_heatmap_experiments.sh for experiments that test the sensitivity of clustered conformal to the hyperparameter values. To view the main results, run jupyter notebook from Terminal, then run the notebooks in the notebooks/ directory.