Gradient-audit README: final mesh-gradients assessment, regularization matrix, Delaunay shipped - #116
Merged
Merged
Conversation
Records the 2026-07-26 re-certification and linear-algebra precision probe in the gradient-audit README: all four jax_grad pixelization variants pass strict FD post-consolidation (production shape <= 6.2e-8 rel err); slogdet vs cholesky log-det gradients agree to 2e-15; the relaxed-KKT NNLS backward pass contributes <= 3e-10 relative gradient error at the default target_kappa; the os_pix=1 einstein_radius FD poisoning persists under the unconstrained solver, re-confirming PyAutoArray#377's solver exoneration. Verdict: no linear-algebra reformulation warranted for gradient precision. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
Records the 2026-07-26 jax_grad/knn.py certification in the gradient-audit README: KNearestNeighbor (Wendland) and KNNBarycentric FD-certified strict on all parameters under split-family regularization; neighbour-based regularization pinned as the scipy TracerArrayConversionError boundary; the KNNBarycentric science-gate caveat (PyAutoArray#317) carried alongside its gradient status. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
Adds the 2026-07-26 sweep results to the gradient-audit README: the full
al.reg x {rectangular, KNN} compatibility matrix (FD-certified / AD-live-
but-FD-limited / hard-error / incompatible, with mechanisms), the Matern
verdict (tfp bessel_kve gradients work; nu=2.5 strict-certified on the
rectangular mesh), the kernel-scheme conditioning note (explicit C^-1 at
cond ~1e9 on clustered traced vertices is the one linear-algebra
reformulation candidate the sweep surfaced), and a findings-log entry.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
Updates the regularization x mesh matrix: ExponentialKernel moves from hard-error to works/FD-limited on both mesh families after the PyAutoArray xp port (same kernel-family noise floor as Matern nu=0.5, which it equals mathematically). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
…is unblockable The 2026-07-26 probe shows stop_gradient on the tables pure_callback input unlocks jax.grad through the full Delaunay likelihood: with the visibility walk in-graph, only the int32 tables (piecewise-constant in the vertices) are frozen, so this is the exact almost-everywhere derivative. 14/14 params live on the production shape, FD median 9.6e-6 / max 2.1e-3, eager==jit. Updates the status row and findings log; ship task filed to PyAutoMind. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
… update Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Updates
jax_profiling/gradient/README.mdwith the 2026-07-26 mesh-gradients assessment (library: PyAutoLabs/PyAutoArray#408; certification: PyAutoLabs/autolens_workspace_test#225):🤖 Generated with Claude Code
https://claude.ai/code/session_013FSqnkgZv97PU9JdkCcthy
Generated by Claude Code