IVAR-experiments

A repository accompanying the paper Inductive Venn–Abers and related regressors containing code for running experiments on synthetic and real datasets and generating formatted result tables for analysis.

---

Installation

Clone the repository and install the required Python dependencies:

pip install -r requirements.txt

---

Running Experiments

All experiments are executed from the src directory using the main module.

python -m main.py --dataset DATASET --noise_level NOISE_LEVEL

Command-Line Arguments

• --dataset
  Specifies the dataset type. Valid options are:

  • synthetic_datasets
  • real_datasets

• --noise_level
  Specifies the noise level applied to the data. Valid options are:

  • 1
  • 3

  Note: This argument is only applicable when using synthetic_datasets. It is ignored when running experiments on real datasets.

---

Example Usage

Run experiments on synthetic datasets with noise level 1:

python -m main.py --dataset synthetic_datasets --noise_level 1 --n_samples 10000

Run experiments on real datasets:

python -m main.py --dataset real_datasets

---

Output

All experiment outputs are saved to the output/ subdirectory.
This directory contains the raw results generated during execution.

---

Results Processing

To process the experimental results and generate tex formatted tables, run the following Jupyter notebook:

process_results.ipynb

The notebook reads data from the output/ directory and produces text-formatted tables suitable for reporting and analysis.

---

Notes

• Ensure commands are executed from the correct directory as specified above.
• Noise levels are only relevant for synthetic datasets.
• The repository is structured to support reproducible experimentation.

---