Potentially replace PCA with SVD and focus on U which is what sets the new axis. Then evaluate the similarity between U matrices via norms. Again, if they're super different then they could be on different axes at which point aggregation doesn't make much sense.
Potentially replace PCA with SVD and focus on U which is what sets the new axis. Then evaluate the similarity between U matrices via norms. Again, if they're super different then they could be on different axes at which point aggregation doesn't make much sense.