Probabilistic PCA for missing data: learning curves shows a phase transition and missing rate acts as an effective reduction in the signal-to-noise ratio, not the sample size.
Code accompanying the ICML 2019 paper
Ipsen, Niels Bruun, and Hansen, Lars Kai.
Phase transition in PCA with missing data: Reduced signal-to-noise ratio, not sample size!.
arXiv preprint arXiv:1905.00709 (2019).
In task01.py learning curves on artificial data are generated, while task02.py works on the Olivetti Faces dataset