Closed kdmalc closed 3 weeks ago
Make sure that you split this for IMU and EMG. Eg in the combined case (we may end up just doing EMG), run mean subtraction on just the EMG portion, and then mean subtraction on just the IMU portion. And standardize separately
Momona said to use her standardization scheme from her UIST paper. Idk what that is. I'll ask her for it / for the code ig. She said the other standardization methods didn't work (assuming she meant they gave bad results)
I added this in utils.preprocessing.py, but all the NBs (the early ones like 200/300, not sure if it comes up elsewhere) need to be switched to using this. Use Standard Scaler (is currently an option you can toggle with a string).
Run the dim reduc eval metric on the reduced forms of both 1) the non-normalized and 2) the normalized (to see which is better). Can also compare against just the mean-subtracted reduced as well.
Eventually repeat this but for the clustering results (eg does normalization affect the clustering results).