todo.md

This document contains a list of work that development items that need to be addressed.

• Plot variance around the loss and measure curves

• Removed logging dependency on tensorboard (use a simple in-house logging serialization format)

• Fine-tune default configuration for efficiency and intuitive appeal

• Extend closer functionality

  • Make sure we can close ourself.
  • Allow harn.config to specify other classes to export (besides the model itself)
  • Need to make the closer generally robust.

• Run one validation batch after every training batch in the first epoch to see how a single set of validation examples evolve.