You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Many operators change the shape/size of the output. This becomes very hard to keep track of unless the user is very comfortable using that specific operator. Here we propose introducing a dimsd attribute to all operators which take dims. Whereas dims refers to the shape of the "model" array (input to forward) prior to flattening, dimsd would refer to the shape of "data" array (input to adjoint). This way, the user could always rely on the following code pattern for an operator Op acting on some dimensions of an N-dimensional array:
Motivation
Many operators change the shape/size of the output. This becomes very hard to keep track of unless the user is very comfortable using that specific operator. Here we propose introducing a
dimsd
attribute to all operators which takedims
. Whereasdims
refers to the shape of the "model" array (input to forward) prior to flattening,dimsd
would refer to the shape of "data" array (input to adjoint). This way, the user could always rely on the following code pattern for an operatorOp
acting on some dimensions of an N-dimensional array:Definition of done
dims
should also havedimsd
.dims_fft
in operators which used other attributes to rely on this functionality.The text was updated successfully, but these errors were encountered: