Robin Kahlow

Made `select_blades` and `keep_blades` take blade indices instead of names in https://github.com/RobinKa/tfga/commit/26ecef36ae0941b12319e312baffb75c2035dadf and added functions taking names `select_blades_with_name` and `keep_blades_with_name` with the old behavior.

For 2D and 3D (including PGA) this seems to work (just took inspiration from the Quaternion logarithm), although I don't know under what conditions this works. ```python def rotor_log(x): norm_x...

General bivector exp and log: https://arxiv.org/abs/2107.03771 (Graded Symmetry Groups: Plane and Simple) - Exp and decomposition implemented in #21

Another idea: - Keep track of blade values and indices like initially done - In the geometric product, instead of using gather on the blade values, create a cayley tensor...

Started some initial work here https://github.com/RobinKa/tfga/tree/feature/faster-cayley-prod Initial results for PGA (16 basis blades), runtime of multiplying two even-graded multivectors (8 basis blades): - Full (old): 1x - Partial cayley (not...

I agree, also here's the recent paper @hugohadfield talked about before https://arxiv.org/abs/2006.07360 which uses many different algebras that could probably be implemented easily using GA. List of some relevant articles...

Hey, I benchmarked both on my GPU and this layout was much faster. I think around 10x for the geometric product, although I don't remember the exact number. (so overall,...

> Harping on about the same topic; as currently implemented, at least one batch axis appears mandatory; I cant just create a single vector and act on it since it...

In my TensorFlow library https://github.com/RobinKa/tfga I chose the dense approach with the 3-tensor with lots of zeros (also see slides here https://tfgap.warlock.ai/#/6/1 about this approach). It was faster than using...