Accessing (for manipulation in python/Saving to csv etc) Optimization History (Training History) from Tensorboard files

Physics Simulation PhysicsNeMo-ML Model Framework Technical Support (PhysicsNeMo Only)
1

Hi, title of the question sums up my question: I want to access (for manipulation in python/Saving to csv etc) Optimization History (Training History) from Tensorboard files.

Is there a way to do this? Thanks. I am not very familiar with tensorboard, have simulation results, and would like to use the loss function values over the course of (say, adam and lbfgs) optimization

Thanks!

2

Hi @ctalbot

I know the Tensorboard package has some methods for accessing the raw data as Pandas dataframes. Perhaps you could use the information in this tutorial: Accessing TensorBoard Data as DataFrames  |  TensorFlow

Otherwise you could edit the trainer class to save the loss value to a CSV manually which would be pretty straight forward.

1 Like