LSTM model

[mkingopng]

Model is completing its training but performance is pretty poor.

Needs experimentation to lift performance

• find the correct sequence length
• hyperparameter tuning: optuna or wandb sweeps
• try different loss functions

Model Features

• improve wandb visualisations
• show feature importance
• improve feature engineering
• add attention mechanism to the model
• gradient clipping
• data augmentation

Completed

• increase the number of LSTM layers
• set seed
• save trained model to pt
• add LR scheduler
• early stopping
• add inference to round out the process

[mkingopng]

completed:

• set random seed function
• save file to pt
• early stopping class
• learning rate annealing

[mkingopng]

• completed sequence length optimisation
• completed testing of LR scheduler and LR annealing scheduler
• completed testing of additional LSTM layers

[mkingopng]

done