ponder-lab / Hybridize-Functions-Refactoring / #1520
72%

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% (+0.2%) from 72.036%
#1520

push

travis-pro

web-flow 
Precision audit: extend two-layer assertions to fixtures 13-58 + 59 (#522)

* Extend `testHasLikelyTensorParameter11` with per-parameter + input-signature assertions

First fixture of the precision-audit pattern (Project 5 / `Ariadne
precision audit via existing test fixtures`). Reads the `.py`'s
semantics — `add(tf.ones([1, 2]), tf.ones([2, 2]))` — and pins what a
static analysis *should* produce for the two parameters:

- `a`: `TensorType(FLOAT32, [1, 2])`
- `b`: `TensorType(FLOAT32, [2, 2])`

Layer 1 asserts these via `Parameter.getTensorTypes()`; Layer 2 asserts
the combined `Function.inferInputSignature()` output. Both pass against
the current Ariadne 0.44.0 + algorithm — no gap to file for this case.

Also static-imports `DType.FLOAT32` as `FLOAT32` and replaces the 14
existing call sites in the same file. Test bodies get less verbose with
the bare constant.

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

* Extend precision audit to fixtures 13, 14, 15

Continuing the precision-audit pattern started in #520:

- `testHasLikelyTensorParameter13`: two functions (`dense_layer`, `add`)
  threaded through `tf.matmul`. All four parameters (dense_layer's x/w/b,
  add's a/b) carry concrete FLOAT32 with the expected shapes derived from
  `tf.ones([3, 2])`, `tf.ones([2, 2])`, `tf.ones([2])` and the matmul
  output (3, 2).
- `testHasLikelyTensorParameter14`: `conv_fn(image)` with
  `image = tf.zeros([1, 200, 200, 100])`. Single-call, single-parameter
  case with concrete rank-4 shape.
- `testHasLikelyTensorParameter15`: `double(a)` called four times with
  diverging dtypes (INT32, FLOAT32, STRING, STRING). Multi-context dtype
  disagreement; algorithm correctly drops signature.

All three pass against Ariadne 0.44.0 + the current algorithm — no gap
to file.

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

* Extend precision audit to fixtures 16, 19, 20, 21, 22, 23, 24, 25

All eight pass against Ariadne 0.44.0 + the current algorithm. No
... (continued)

1188 of 1645 relevant lines covered (72.22%)

0.72 hits per line

Jobs
ID Job ID Ran Files Coverage
1 #1520.1 31
72.22
Source Files on build #1520