Open geoffwoollard opened 5 years ago
Eg do pixel drop out and see if the ligand binding site / missing chunk of particle is the hot pixels.
Can vectorize by pre-generating mask with NaNs/0s and adding/multiplying. The mask would have the pixel drop out windows and be shape N,N,pixel_drop_outs. Where N is the particle image shape.
https://pypi.org/project/keras-targeted-dropout/
https://stackoverflow.com/questions/45361559/feature-importance-chart-in-neural-network-using-keras-in-python
Eg do pixel drop out and see if the ligand binding site / missing chunk of particle is the hot pixels.