Roadmap

[LeonieBorne]

Roadmap

This issue contains the roadmap of this project. It's a place to start to investigate the issues that you can contribute to.

Please note that the list of tutorials proposed are by no means exhaustive. If you wish to add/modify some of them, do not hesitate to suggest it by creating a new issue!

General

Here is a (non-exhaustive) list of points to be dealt with before/during/after the tutorials have been written.

• Databases to showcase #4 Find databases to use in the tutorials on OpenNeuro
• Export the tutorials in NeuroLibre

Tutorial 0. Introduction #5

The objective of this introductory tutorial is to explain the general principles of cross-decomposition algorithms, their possible applications and practical considerations. It should introduce and refer to the other tutorials.

This tutorial should also give an overview of the different cross-decomposition algorithms that exist, including CCA, PLS regression, PLS canonical, PLS-PM (for more than 2-blocks of variables), etc.

Useful references

• Cross-decomposition in scikit-learn: scikit-learn documentation for the cross-decomposition module (CCA, PLS regression, PLS canonical). Note that the documentation should be updated soon (see current pull request, corresponding branch).
• PLS-PM: "PLS Path Modeling with R" Gaston Sanchez
• PLS-PM in Python
• PLS methods for neuroimaging: Krishnan, Anjali, et al. "Partial Least Squares (PLS) methods for neuroimaging: a tutorial and review." Neuroimage 56.2 (2011): 455-475.
• CCA for neuroscientists: Wang, Hao-Ting, et al. "Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists." NeuroImage (2020)

Tutorial 1. Data preprocessing #6

This tutorial focus on minimal data preprocessing, usually required as for most machine-learning methods, with among other things:

• z-scoring of each variable,
• outlier detection,
• missing values processing,
• deconfounding procedures.

Useful references

• Section 5.1 CCA for neuroscientists: Wang, Hao-Ting, et al. "Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists." NeuroImage (2020)

Tutorial 2. Data reduction #7

This tutorial focus on dimensionality-reduction techniques (PCA, ICA, etc.) that can provide useful data preprocessing when the number of variables exceeds the number of samples.

Useful references

• Section 5.2 CCA for neuroscientists: Wang, Hao-Ting, et al. "Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists." NeuroImage (2020)

Tutorial 3. Model selection #8

This tutorial introduce to the different techniques used to evaluate/validate/select the model.

• How to choose the optimal number of latent sources of variation to be extracted?
• How to evaluate the contribution of each individual input variable to the overall modeling solution?
• How to compare the models?

Useful references

• Section 5.3 CCA for neuroscientists: Wang, Hao-Ting, et al. "Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists." NeuroImage (2020)
• Section 4.6 PLS-PM: "PLS Path Modeling with R" Gaston Sanchez
• Comparison CCA/PLS: Rahim, Mehdi, Bertrand Thirion, and Gaël Varoquaux. "Multi-output predictions from neuroimaging: assessing reduced-rank linear models." 2017 International Workshop on Pattern Recognition in Neuroimaging (PRNI). IEEE, 2017.
• Permutation inference for CCA: Winkler, Anderson M., et al. "Permutation inference for Canonical Correlation Analysis." arXiv preprint arXiv:2002.10046 (2020).

[htwangtw]

I will add one more paper to tutorial 3:
Permutation Inference for Canonical Correlation Analysis
https://arxiv.org/abs/2002.10046
I would like to implement this method - they have matlab code and pseudocode
However, the implementation could be a project of its own.

[fBeyer89]

I would suggest this tutorial as reference for PLS (Partial Least Squares (PLS) methods for neuroimaging: A tutorial and review) It's quite old but I found it really helpful when I was implementing PLS.

[htwangtw]

Let's welcome @fBeyer89 @bbuckova @diiobo @nadinespy from EMEA to the project!

[LeonieBorne]

@htwangtw @fBeyer89 Thank you so much for the references! I have added them to Tutorial 3 and 0.