wavetorch
wavetorch copied to clipboard
🌊 Numerically solving and backpropagating through the wave equation
import wavetorch File "G:\Anaconda\lib\site-packages\wavetorch\__init__.py", line 1, in from . import rnn, cell, geom, source, probe, data, plot, utils, io ImportError: cannot import name 'data' from 'wavetorch' (G:\Anaconda\lib\site-packages\wavetorch\__init__.py
Should consider the performance benefit of implementing an adjoint calculation for the backward pass through the `forward()` method in `WaveCell`. This would potentially save us on memory during gradient computation...
Thanks for this outstanding work! May I ask why you utilized the class class TimeStep(torch.autograd.Function) with @staticmethod forward and backward (in cell.py)? In this class, some backward gradient is manually...