dnn cli

[brettviren]

We want to capture workflow for training and validating DNNROI and future deep learning components in the WC ecosystem so that they may be (re)exercised in the future and extended to new detectors.

To do that we want a CLI which can perform tasks in the workflow for existing DL (DNNROI) and future models in a general, extensible way.

This issue holds an outline which may evolve over the course of development. We want these features:

• a train subcommand to exercise training on the named network
• a plot subcommand to plot training metadata and app-specific plots
• implementation of U-Net and the variant that DNNROI uses
• a general abstract API (Python module pattern) for each application with general base imp and derived imp for DNNROI
  • network constructor
  • dataset and any transformations
  • training loop
  • training metric plots
  • model specific validation plots
• tracking and plotting of training metadata
• persist and load the model