ponder-lab / Hybridize-Functions-Refactoring / #1551
72%
main: 72%

LAST BUILD BRANCH: test/precision-audit-batch-14
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
#1551

push

travis-pro

khatchad 
Extend precision audit to fixtures 67-70 (`RaggedTensor.from_nested_value_rowids`)

Apply the two-layer audit helper to fixtures 67-70, all of which exercise `tf.RaggedTensor.from_nested_value_rowids(x, nested_value_rowids)` with `x` from `tf.keras.Input(shape=[None], dtype=tf.string)` and two row-id tensors yielding `nrows0 = 4`, `K = 2` ragged dimensions.

Layer 1 (Ariadne): `STRING` dtype, shape `(NumericDim(4), null, null, null)`. Position 0 carries the constant nrows derived from `max(rowids[0]) + 1`; positions 1 and 2 are ragged markers (pre-`RaggedDim`-typed-sentinel, still raw `null` in the consumed 0.44.0 release—see wala/ML#544 for the flip target which has landed on master but not yet in a Hybridize-consumed release); position 3 is the dynamic dim from `keras.Input`'s `shape=[None]` (also raw `null` pre-typed-sentinel—see wala/ML#545).

Layer 2 (Hybridize): the inference algorithm collapses every `null` to `SymbolicDim("?")`, so the per-element raggedness vs. dynamic-dim distinction is lost in the inferred signature. The collapse is correct for the dynamic dim (a `TensorSpec` wildcard is the right representation) but lossy for the two ragged dims (a `RaggedTensorSpec` is the precise emission—see #524).

Static-import `STRING` and shift to `Arrays.asList` per the project's no-FQN style; sweep applies to the previously-inlined fixture-59 / 60-66 anchors as well.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

1188 of 1645 relevant lines covered (72.22%)

0.72 hits per line

Jobs
ID Job ID Ran Files Coverage
1 #1551.1 31
72.22
Source Files on build #1551