makassar-ml - Release for MS Thesis Defense

Release for MS Thesis Defense

Release of codebase at time of MS thesis defense. Also includes collection of all tuned models, and selection of the best tuned models as conveyed in our paper.

Tuning Results

For convenience, we provide pre-trained models in 2 separate files:

File | Size | Description ------ | ---- | ---- all_models.tar.gz | 1.2 GB | Collection of all tuned models (larger file size). best_models.tar.gz | 335 MB | Selection of the best models, as discussed in our paper (smaller file size).

The tuning results consist of several files for each model:

• <model>.h5 - Trained TensorFlow model file
• <model>_hparams.json - Hyperparameter selection for model
• <model>_history.csv - Training history CSV file (useful for plotting)
• <model>_metrics.json - Final metric values after training (useful for tabular display)

The best models contained within this release are:

• LSTM baseline
• InceptionV3 baseline
• FoT model 9
• ViT model 37
• FuT classifier model 43
• FuT regression model 20
• FuT multi-task model 42

- Jupyter Notebook
Published by zanderman almost 4 years ago