Skip to content

Latest commit

 

History

History
28 lines (21 loc) · 687 Bytes

README.md

File metadata and controls

28 lines (21 loc) · 687 Bytes
Raw

Keras Scaffold

The following repo is a starting scaffold for modeling projects.

Implemented Models

  • CycleGAN
  • Pix2Pix

Implemented Blocks

  • Unet Attention Block
  • Residual Block

Regularizations

  • TODO: focal loss
  • TODO: diversity sensitive GAN

Implemented Callbacks

  • log_code.py, logs code to GCP bucket before training
  • copy_keras_model.py, copies keras models to GCP bucket
  • save_multi_model.py, saves multiple keras models each epoch

Custom Callbacks

  • generate_images.py, writes images to tensorboard
  • get_csv_metrics.py, writes metrics of interest to a csv file.

Hyperparam tuning

  • GCP hyperparam tuning
  • TODO: Axios