This is the official reporitory for the code associated with the paper: On the Identifiability of Quantized Factors by Vitória Barin-Pacela, Kartik Ahuja, Simon Lacoste-Julien, Pascal Vincent, Conference on Causal Learning Reasoning (CLeaR), 2024.
It contains notebooks and code for reproducing the figures in the paper.
- Figure 2 is reproduced in the notebook
exoplanet_data.ipynb, it contains evidence of axis-aligned discontinuities in the Nasa Expolanets dataset. - Figure 3 is reproduced in the notebook
results_unfactorized.ipynb, which contains a synthetic dataset with non-factorized support. We provide results for our model (axis alignment), Hausdorff Factorized Support, and Linear ICA. - Figure 4 is reproduced in the notebook
results_factorized.ipynb, which contains a synthetic dataset with factorized support. We provide results for our model (axis alignment) and Linear ICA. - Figure 5 is reproduced in the notebook
mocap_data.ipynb, it contains evidence of axis-aligned discontinuities in the CMU motion capture dataset.
Use requirements.txt.
The majority of quantized_identifiability is licensed under CC-BY-NC, however portions of the project are available under separate license terms: iVAE is licensed under the MIT license.