A more universal, convenient, and concise version

Combining HOPE with the computational graph provides users with a convenient and efficient interface for computing higher-order derivatives. We provide two implementations for the calculation of high-order derivatives, HOPE and Autograd. Users can conveniently compute high-order derivatives of neural networks and obtain Taylor series approximations for the network by calling hope.py or auto.py, while HOPE has significant advantages in terms of accuracy, speed, and memory cost compared with Autograd. If there are some deep learning modules missing in HOPE, you can add them as needed in ‘’utils/HopeGrad.py‘’.