A guideline project demonstrating how to organize and build recurrent neural networks (RNNs)—specifically LSTM-based classifiers—on imbalanced data. This repo provides a clear directory structure, configuration patterns, and training/evaluation scripts so you can adapt the pattern to your own sequence-modeling tasks.
- Modular code layout: clear separation of data loading, model definition, training loop, and evaluation
- Imbalanced-data handling: built-in support for class weights and oversampling
- Config‐driven: all hyperparameters, paths, and training options live in a single YAML file
- Logging & checkpoints: automated saving of best models
- Easy experiment tracking: built-in scripts to reproduce experiments from a single command