Skip to content

float precision check in rational quadratic spline discriminant #71

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
francesco-vaselli wants to merge 3 commits into
base: master
Choose a base branch
from
Open

float precision check in rational quadratic spline discriminant #71

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d8b0858
float precision check
francesco-vaselli Mar 15, 2023
4934a85
comments and torch.where
francesco-vaselli Sep 30, 2023
4bcc21d
comments 1
francesco-vaselli Sep 30, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions nflows/transforms/splines/rational_quadratic.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,6 +139,10 @@ def rational_quadratic_spline(
c = -input_delta * (inputs - input_cumheights)

discriminant = b.pow(2) - 4 * a * c

float_precision_mask = (torch.abs(discriminant)/(b.pow(2) + 1e-8)) < 1e-6
Copy link
Contributor

@arturbekasov arturbekasov Sep 24, 2023

Choose a reason for hiding this comment

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

Could you add a comment explaining where this particular formula comes from, and define constants/parameters for the magic numbers?

Copy link
Contributor Author

@francesco-vaselli francesco-vaselli Sep 30, 2023

Choose a reason for hiding this comment

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

I have added a comment trying to explain the logic of the operation without being too verbose.
I am afraid the numbers are manly motivated by heuristics to manage numerical stability in our case.

discriminant[float_precision_mask] = 0

assert (discriminant >= 0).all()

root = (2 * c) / (-b - torch.sqrt(discriminant))
Expand Down