Updated README

This commit is contained in:
Dominik Moritz Roth 2024-12-21 18:31:26 +01:00
parent 9fb0014a99
commit de2b9a10d6

View File

@ -65,9 +65,7 @@ pytest tests/test_projections.py
*Note*: The test suite verifies:
1. All projections run without errors and maintain basic properties (shapes, positive definiteness)
2. KL bounds are actually (approximately) met for:
- KL projection (both diagonal and full covariance)
- Wasserstein projection (diagonal covariance only)
2. KL bounds are actually (approximately) met for true KL projection (both diagonal and full covariance)
3. Gradients can be computed through all projections:
- Both through projection operation and trust region loss
- Gradients have correct shapes and are finite