How should update_grid be used? For example, should it be used every batch during training? Since the default valid grid range is [-1, 1], if update_grid is not used, the input values may go outside this range, and it seems that fitting might not work properly. However, in the case of the MNIST example, update_grid is not used. Does this mean update_grid is unnecessary?
How should update_grid be used? For example, should it be used every batch during training? Since the default valid grid range is [-1, 1], if update_grid is not used, the input values may go outside this range, and it seems that fitting might not work properly. However, in the case of the MNIST example, update_grid is not used. Does this mean update_grid is unnecessary?