ponder-lab / Hybridize-Functions-Refactoring / #1569
72%
main: 72%

LAST BUILD BRANCH: gh-readonly-queue/main/pr-556-e9efb0117db7248a8cfe020d8d0b96e8b63be1e0
DEFAULT BRANCH: main
Ran
Jobs 1
Files 31
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 72.219%. Remained the same
#1569

Pull #534

travis-pro

khatchad 
Extend precision audit to fixtures 94-99 (`sparse.eye` and `linalg.eye`)

Six fixtures across two TF APIs. Numerically all six are IDEAL (Layer 1 and Layer 2 both match the Python runtime ground truth), but the sparse half surfaces a runtime-semantics finding worth anchoring.

**`linalg.eye(2)`** (97-99) — dense identity matrix, `FLOAT32, (2, 2)`. Clean IDEAL case.

**`sparse.eye(2, 3)`** (94-96) — Ariadne emits `FLOAT32, (2, 3)` matching the static shape exactly, and the inference algorithm passes it through unchanged. The numbers are right. The runtime semantics aren't: a `tf.function(input_signature=[TensorSpec(shape=(2, 3), dtype=float32)])` rejects a `SparseTensor` argument, since TF distinguishes `SparseTensor` from `Tensor` at the API level. The precise emission is `tf.SparseTensorSpec`, sibling to the `RaggedTensorSpec` case (#524).

Filed as #533 with `SparseTensorSpec` as the proposed flip target. Inline TODO comments at each of 94-96 reference the issue; the assertion itself pins current behavior unchanged.

Distinct from the ragged case: ragged-marker `null` surfaces at the per-dim level in `TensorType.dims`, so the imprecision is visible in the audit assertions. Sparse-vs-dense is currently invisible at the `TensorType` level—Ariadne emits the same shape and dtype for `sparse.eye(2, 3)` as it would for a dense (2, 3) tensor. The signal needs to come from elsewhere (the producing tensor generator, or a sparse-aware extension of Ariadne's representation).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Pull Request #534: Extend precision audit to fixtures 94-99 (`sparse.eye` and `linalg.eye`)

1188 of 1645 relevant lines covered (72.22%)

0.72 hits per line

Jobs
ID Job ID Ran Files Coverage
1 #1569.1 31
72.22
Source Files on build #1569