Search⌘ K
AI Features
Home
Courses
Deep Learning with PyTorch Step-by-Step: Part I - Fundamentals

SummaryWriter

Explore how to use PyTorch's SummaryWriter to create meaningful logs for TensorBoard. Learn to set up organized log directories, add model graphs, and track training and validation losses over epochs. Understand best practices for managing multiple runs to avoid corrupted visualizations and enhance model training insights.

We'll cover the following...

Overview of SummaryWriter

It all starts with the creation of a SummaryWriter:

Python 3.5
# setting SummaryWriter up
writer = SummaryWriter('runs/test')

If we try printing this, we get the following output:

Python 3.5
# setting SummaryWriter up
writer = SummaryWriter('runs/test')
print(writer)

Since we told TensorBoard to look for logs inside the runs folder, it only makes sense to actually log to that folder. Moreover, to be able to distinguish between different experiments or models, we should also specify a sub-folder: test.





If we do not specify any folder, TensorBoard will default to runs/CURRENT_DATETIME_HOSTNAME, which is not such a great name if you are looking for your experiment results in the future.

So, it is recommended to try to name it in a more meaningful way like runs/test or runs/simple_linear_regression. It will then create a subfolder inside runs (the folder we specified when we started TensorBoard).

Even better, you should name it in a meaningful way, and add DateTime or a sequential number as a suffix like runs/test_001 or runs/test_20200502172130 to avoid writing data of multiple runs into the same folder (we will see why this is bad in the add_scalars section below).

SummaryWriter methods

The SummaryWriter class implements several methods to allow us to send information to TensorBoard:
...