Dear All,
If I have several datasets, such as Dataset 1, 2, and 3. I have pooled all the three datasets together and want to perform an age prediction task and I want to use the tangent space vectors as features. So how should I calculate the tangent space projection. I think about two options:

I could calculate the projection in each dataset by using the reference mean matrix within each dataset;

I just pooled three dataset together and calculate the tangent space on the train data split and fit_transform on the test data split.
Thansk.