Given an IMU message containing orientation, angular_velocity and linear_acceleration, if I change the orientation (i.e. apply a rotation), then to get the linear_acceleration in the new frame of reference I simply rotate the linear_acceleration vector by the same rotation matrix.

Does the same hold for the angular_velocity?

My understanding is that the angular_velocity found in ROS messages (IMU, twist, etc.) is expressed as time derivative of rpy angles, so it is not an element of a vector space. Does it make sense to multiply it by a rotation matrix to get its representation with respect to a new frame? If angular_velocity was a real vector (commonly denoted as omega), then that process would be correct. But I'm not sure how I can apply a rotation in the case of time derivative of rpy.