LSE Omnibus embedding #835
Conversation
graspologic/embed/omni.py
Outdated
| @@ -145,6 +168,10 @@ class OmnibusEmbed(BaseEmbedMulti): | |||
| Only applicable for ``algorithm="randomized"``; allows you to seed the | |||
| randomized svd solver for deterministic, albeit pseudo-randomized behavior. | |||
|
|
|||
| lse : bool, optional (default False) | |||
There was a problem hiding this comment.
I pulled this from the earlier LSE Omni PR -- Is this the api we can all agree on or does someone have other suggestions?
In topologic, I believe we used an enum here (ase or lse, default ase.) Other devs may prefer a magic string but I'd prefer not.
We can also leave this as is and change it in the future if it turns out we want to support something other than ase/lse.
| Lbar = (L1 + L2) / 2 | ||
|
|
||
| omni = OmnibusEmbed(n_components=3, diag_aug=diag_aug, lse=True) | ||
| OmniBar = compute_bar(omni.fit_transform([csr_matrix(L1), csr_matrix(L2)])) |
There was a problem hiding this comment.
Works with sparse.
|
implementation looks fine, my continued queasyness is just that I don't have a simulation / data example that I can show someone that says when LSE omni would be preferred |
what level of queasiness are we talking about? like, we're currently using it to measure fluidity in our viva insights product to calculate https://docs.microsoft.com/en-us/workplace-analytics/azure-templates/ona-metric-calculations#fluidity or are you more wondering if there's an academic paper on it? because I don't think we have that edit: jlar has an entire slide deck he can go through with you if that would help. |
not extremely, just think it would be good to have a public-facing reason for using this stuff as there is no associated publication, as you mention |
|
should @nyecarr reference that page in the docs as well? |
Reference Issues/PRs
What does this implement/fix? Briefly explain your changes.
Add ability to create LSE omnibus embeddings.
Any other comments?
Based off of @zeou1 PR