maxwnewcomer / pestpp-tensorboard

Tensorboard extension for Pest++

TensorBoard Tools for PEST++ and NWT files

The goal of this project was to provide users with a more visual understanding of the internal processes within NWT files and PEST model_runs_completed

run_tensorboard.py
run_tensorboard.py provides users with the ability to see live loss, metric, and weight tracking
--inputdir is required when running
--logname is not required, but can be used for logname customization

How can you use this resource?

• By tracking weights and loss live, you can easily stop a failed run before wasting possible hours of computation time
• Easy comparison between runs, allowing for you to tune your models with a visual understanding of your model

understand_NWT.py
understand_NWT.py allows users to understand how seemingly 'black box' variables effect their model runtime
--filepath of NWT_Explore_out.csv needed for visualization

How can you use this resource?

• Use the variable selection tools under the hparam tab to understand how different combinations of variables change the resulting runtime
• Visualize your variables in both line connection plots and scatterplots comparing variable and runtime, allowing you to extract patterns within your variables

Recommended
Use the useful explore_nwt.py program to compare at least 30 different variable combinations for best results

Goals

• Use machine learning to optimize variable selection, minimizing randomness
• Increased depth support for PEST++, specifically pestpp-ies