Recurrent Neural Network with PyTorch

In this assignment you will implement recurrent neural networks (RNN / LSTM) using PyTorch for sequence modeling.

You will cover:

• Vanilla RNN: Implement a simple RNN cell manually and understand the hidden-state recurrence \(h_t = \tanh(W_h h_{t-1} + W_x x_t + b)\).
• LSTM: Use nn.LSTM to handle long-range dependencies and compare it with the vanilla RNN.
• Sequence processing: Prepare variable-length sequences, apply padding, and use pack_padded_sequence for efficient batching.
• Training: Train on a sequential dataset (e.g., time-series prediction or character-level language modeling) and analyze gradient flow.
• Evaluation: Measure performance on held-out sequences and visualize predicted vs. true values.

Problems will be released closer to the due date.