ModelLogger

[dmccloskey]

Description

Methods for logging model diagnostics during training to replace verbose outputs to the console

Features

• [x] Model loss/error vs epoch curves
• [x] Train/Val accuracy vs epoch curves
• [x] Node Output/Expected Output vs epoch
• [x] Ratio of weight updates (want weights:updates ~1e-3) vs epoch
• [x] Node output distribution (mean/var) per layer per mini-batch per time-step vs epoch
• [x] Node error distribution (mean/var) per layer per mini-batch per time-step vs epoch
• [x] Epoch iteration vs time
• [x] Weight/Node values per time_step vs. Epoch

Objectives

• [x] ModelLogger class
• [x] getters/setters for ModelLogger for Model
• [x] activation of ModelLogger via ModelTrainer during training or validation
• [x] method to log each of the features above
• [x] tests for each of the features above