[FEA] PCA initialization for TSNE

[cjnolet]

UMAP uses a spectral embedding to initialize points and it was recently demonstrated in this paper that the choice of initialization is important to preserving global structure. The paper also demonstrates that TSNE can preserve global structure fairly well by using PCA to initialize the embeddings.

The PCA initialization is done directly on the training data (using only the first 2 principal components).

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[cjnolet]

Linking #1029 and preferring use of this issue since we have made progress on sparse PCA since the original issue was opened.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.