This is my (unofficial) implementation and evaluation of PVR-MNIST example on

A. Geiger, Z. Wu, H. Lu, J. Rozner, E. Kreiss, T. Icard, N. D. Goodman, and C. Potts, “Inducing causal structure for interpretable neural networks,” 2021, Paper in arXiv.

• Getting Started
  • Prerequisites
  • Installation
• Results
• Usage
• Contributing
  • Reporting
  • Pull requests

Follow these instructions to get a copy of the project up and running on your local machine for development and testing purposes.

• Python 3.7
• Pip 22.2.2

Provide step-by-step instructions on how to install and set up the project. Include any necessary configuration steps.

1. Clone the repository:

git clone [email protected]:aespogom/causal_PVR_MNIST.git

2. Navigate to the project

cd CAUSAL_PVR_MNIST

3. Create a virtual environment
4. Install dependencies

pip install -r requirements.txt

5. Run the project You can run the following line with the default parameters to get the results of model_04, which resembles A. Geiger et al results.

python -m causal_PVR_MNIST.py

The main script is placed at CAUSAL_PVR_MNIST/causal_PVR_MNIST.py. This script is meant to load the data, create the dataloaders, load the models, train and evaluate the results.

Firstly, the data is downloaded in CAUSAL_PVR_MNIST/data/.

Then, the dataset model is placed at CAUSAL_PVR_MNIST/dataset/BlockStylePVR.py, following the example from OfirKedem.

Both models, student and teacher are placed at CAUSAL_PVR_MNIST/models/.

• In the case of the student, we are using the Resnet18 architecture as the paper states. The forward and backward implementation is customized to satisfy the IIT.
• In the case of the teacher, an oracle model is implemented as a look-up table.

The required configurations are placed at CAUSAL_PVR_MNIST/training_configs/.

• /MNIST.nm provides the required information for the interchange interventions.

The final weights for the student model are stored in CAUSAL_PVR_MNIST/results/

We welcome and encourage community contributions to improve this project. To contribute, please follow these guidelines:

If you encounter any issues or have suggestions for improvements, please check the Issues section to see if the topic has already been discussed. If not, open a new issue, including:

• A clear and descriptive title.
• A detailed description of the issue or suggestion.
• Steps to reproduce the issue if it's a bug.
• Your environment (e.g., operating system, browser, or version of the project).
• Any relevant screenshots or error messages.

We accept contributions in the form of pull requests. To submit a pull request, please follow these steps:

1. Fork the repository.
2. Create a new branch with a descriptive name for your feature or bug fix.
3. Make your changes and ensure they are well-documented.
4. Run any necessary tests and ensure they pass.
5. Commit your changes with a clear and concise commit message.
6. Push your branch to your fork.
7. Create a pull request, including a description of your changes.