KEMBAR78
Debiasing for Tensorboard smoothing by alexirpan · Pull Request #639 · tensorflow/tensorboard · GitHub
Skip to content

Conversation

@alexirpan
Copy link
Contributor

This takes the existing smoothing algorithm, and updates it to avoid getting biased on the initial value. Addresses #610. At low smoothing values, the difference is small, but as smoothing approaches 1, the difference becomes clearer. At smoothing=1, the old code draws a straight line at data[0] - this code draws a straight line at 0 instead, just because of how the initialization changes.

Below are several pictures of the same curves with different smoothing values. The first image (teal curve) is before the change, and the second image (orange curve) is after the change.

Smoothing=0.0
0before
0after

Smoothing=0.1
0_1before
0_1after

Smoothing=0.6
0_6before
0_6after

Smoothing=0.9
0_9before
0_9after

Smoothing=0.999
0_999before
0_999after

Smoothing=1
1before
1after

Copy link

@teamdandelion teamdandelion left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you for this change, and for the clear before/after analysis on a number of different charts.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants