Currently, the keras and torch constraints are expensive due to the back and forth of the weights between the native and R representations. In fact, keras shows a warning saying that the callback takes too much time compared with the rest. We should see if there's a better way to implement this without bringing the matrices to R.
Currently, the keras and torch constraints are expensive due to the back and forth of the weights between the native and R representations. In fact, keras shows a warning saying that the callback takes too much time compared with the rest. We should see if there's a better way to implement this without bringing the matrices to R.