Does nilearn orthogonalize the regressor in design matrix? How do I cancel it ?
For example I create a design matrix from events, the decision_error should be zero but it is not.
Hi, the value is quasi zero, up to numerical error.
Best,
Bertrand
Hi, did you find a way to cancel the orthogonalization on nilearn?
Thanks,
Tamir
Hi, Tamir.
I didn’t find the function to cancel the orthogonalization on nilearn. But you can change the code of nilearn to turn off the orthogonalization easily.
You can find the code in nilearn.glm.first_level.hemodynamic_models.py.compute_regressor.
And add code “if orth:” which use variable orth to control the orthogonalization.
By the way, I didn’t see the significant difference with previous results on my study. Because the regressors didn’t have high collinearity.
Yukun