ponder-lab / Hybridize-Functions-Refactoring / #2061

80%

main: 79%

push

github

Emit `SparseTensorSpec` for sparse tensor parameters (#644)

* Emit SparseTensorSpec for sparse tensor parameters (#533)

With Ariadne 0.51.0 typing SparseTensor parameters as sparse, route them to
`tf.SparseTensorSpec` in the inferred input signature, mirroring the ragged
path (#524):

- `InputSignature.specTypeName` returns `SparseTensorSpec` when the
  parameter's `TensorType.isSparse()`, ahead of the ragged and dense
  branches; `toTensorSpecList` and `requiredSpecTypeNames` follow from it.
- `Function.inferSpec` carries a sparseness consensus (every context sparse)
  onto the reduced `TensorType` via `asSparse()`, so the inferred signature
  preserves sparseness rather than collapsing it to a dense `TensorSpec`. A
  mixed sparse/dense parameter falls through to dense (pre-#533 behavior).

The nine `testHasLikelyTensorParameter*` sparse audits flip from the
`(sparse, dense)` divergence pinned in #636 to a single sparse expectation
now that both layers agree. Adds `InputSignatureTest` coverage for the
`SparseTensorSpec` emission and the sparse `requiredSpecTypeNames`.

Closes #533.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Update the sparse-test Javadocs for SparseTensorSpec emission

The six `sparse.eye`/`sparse.SparseTensor` test Javadocs described the
pre-#533 behavior (the inferred dense `TensorSpec` rejecting `SparseTensor`
arguments, emission "tracked separately"). With this PR emitting
`SparseTensorSpec`, reword them to state the parameter is inferred as sparse
and the signature emits a `SparseTensorSpec` that admits the sparse inputs.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

7 of 7 new or added lines in 2 files covered. (100.0%)

1600 of 2000 relevant lines covered (80.0%)

0.8 hits per line