Numpy with manual gradient calculations for TPTT with RNNs for a class project.
If you want the PyTorch version, view the old/tptt_mnist.py file.
Much of the code was taken from https://github.com/nmanchev/MM-TPTT-RNN/blob/main/task_a/rnn_stptt.py, but we removed the stochastic nature of that code as well and modified it to address our project's requirement of no machine learning frameworks and manual graident calculations.
Paper link - https://www.jmlr.org/papers/volume21/18-141/18-141.pdf
Original Github repo by authors - https://github.com/nmanchev/tptt?tab=readme-ov-file#recurrent-neural-network-with-target-propagation-through-time
Note : The original code is in Theano