tablassert 7.4.7__tar.gz → 7.4.8__tar.gz

{tablassert-7.4.7 → tablassert-7.4.8}/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to this project are documented in this file.
 
+## 7.4.8 - 2026-05-12
+
+### Changes
+- Expanded `fullmap_audit()` failure logging in `qc.py` to include the underlying score values that caused each rejection. Failed CURIE log lines now carry `FUZZ_RATIO`, `FUZZ_PARTIAL`, and (when the BERT stage ran) `BERT_SIMILARITY` alongside the existing `STORE`/`CONFIG`/`COL`/`ORIGINAL`/`PREFERRED`/`CURIE` fields, making it easier to diagnose why a term was dropped.
+- Attached the per-row fuzzy and BERT scores as columns on the pending frame before masking, and switched the intermediate `pl.concat()` calls to `how="diagonal"` so the score columns survive concatenation with the already-passed rows.
+
 ## 7.4.7 - 2026-05-11
 
 ### Changes

{tablassert-7.4.7 → tablassert-7.4.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tablassert
-Version: 7.4.7
+Version: 7.4.8
 Summary: Extract knowledge assertions from tabular data into NCATS Translator-compliant KGX NDJSON — declaratively, with entity resolution and quality control built in.
 Project-URL: Homepage, https://github.com/SkyeAv/Tablassert
 Project-URL: Source, https://github.com/SkyeAv/Tablassert

{tablassert-7.4.7 → tablassert-7.4.8}/docs/changelog.md

@@ -4,10 +4,10 @@ The canonical release history lives in the repository root at [`CHANGELOG.md`](h
 
 ## Current Release Notes
 
-## 7.4.7 - 2026-05-11
+## 7.4.8 - 2026-05-12
 
 ### Changes
 
-- Added an `is_valid_pmc_id` model validator on `Provenance` that enforces `publication` starts with `PMC` followed by digits when `repo` is `PMC`. The constraint was previously documented in 7.3.6 but only now enforced at parse time.
+- Expanded `fullmap_audit()` failure logging in `qc.py` so each rejected CURIE log line now carries the underlying `FUZZ_RATIO`, `FUZZ_PARTIAL`, and (when the BERT stage ran) `BERT_SIMILARITY` score values, making it easier to diagnose why a term was dropped during QC.
 
 For older releases and the full project history, open the root `CHANGELOG.md` in the repository.

{tablassert-7.4.7 → tablassert-7.4.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "tablassert"
-version = "7.4.7"
+version = "7.4.8"
 description = "Extract knowledge assertions from tabular data into NCATS Translator-compliant KGX NDJSON — declaratively, with entity resolution and quality control built in."
 authors = [
     { name = "Skye Lane Goetz", email = "sgoetz@isbscience.org" }
@@ -106,7 +106,6 @@ markers = ["network: requires internet", "gpu: requires CUDAExecutionProvider"]
 line-length = 120
 indent-width = 4
 target-version = "py311"
-exclude = ["*.tcss"]
 
 [tool.ruff.format]
 quote-style = "double"

{tablassert-7.4.7 → tablassert-7.4.8}/src/tablassert/qc.py

@@ -133,7 +133,7 @@ def fullmap_audit(
     passed: pl.DataFrame = pairs.filter(pl.col(out))
     pending: pl.DataFrame = pairs.filter(~pl.col(out))
 
-    exempt_curies: str = r"^CHEBI|^PR|^UniProtKB|^NCBIGene|^UMLS"
+    exempt_curies: str = r"^CHEBI|^PR|^UniProtKB|^NCBIGene|^UMLS|^UNII|^PUBCHEM|^MONDO"
     is_exempt: pl.DataFrame = pending.with_columns(pl.col(cols[0]).str.contains(exempt_curies).alias(out))
     pairs = pl.concat((passed, is_exempt))
 
@@ -160,9 +160,11 @@ def fullmap_audit(
     ratio_scores: object = cpdist(originals, preferreds, scorer=fuzz.ratio)
     partial_scores: object = cpdist(originals, preferreds, scorer=fuzz.partial_token_sort_ratio)
 
-    fuzz_mask: pl.Series = pl.Series(out, (ratio_scores >= 20) | (partial_scores >= 20), dtype=pl.Boolean)
+    pending = pending.with_columns([pl.Series("fuzz_ratio", ratio_scores), pl.Series("fuzz_partial", partial_scores)])
+
+    fuzz_mask: pl.Series = pl.Series(out, (ratio_scores >= 20) | (partial_scores >= 30), dtype=pl.Boolean)
     masked_fuzz: pl.DataFrame = pending.with_columns(fuzz_mask)
-    pairs = pl.concat((passed, masked_fuzz))
+    pairs = pl.concat((passed, masked_fuzz), how="diagonal")
 
     passed = pairs.filter(pl.col(out))
     pending = pairs.filter(~pl.col(out))
@@ -179,22 +181,33 @@ def fullmap_audit(
     n: int = len(originals)
     similarity: object = cosine_similarity(embeddings[:n], embeddings[n:]).diagonal()  # pyright: ignore
 
+    pending = pending.with_columns(pl.Series("bert_similarity", similarity))  # pyright: ignore
+
     bert_mask: pl.Series = pl.Series(out, similarity >= 0.2, dtype=pl.Boolean)  # pyright: ignore
     BERT_fuzz: pl.DataFrame = pending.with_columns(bert_mask)
-    pairs = pl.concat((passed, BERT_fuzz))
+    pairs = pl.concat((passed, BERT_fuzz), how="diagonal")
 
     passed = pairs.filter(pl.col(out))
     pending = pairs.filter(~pl.col(out))
 
     # * Add Logging For Failed CURIES
     if log and pending.height > 0:
-        for c, o, p in zip(
-            pending.get_column(col).to_list(),
-            pending.get_column(original).to_list(),
-            pending.get_column(preferred).to_list(),
-        ):
-            logger.info(
-                f"FAILED | STORE: {section_hash} | CONFIG: {config_file} | COL: {col} | ORIGINAL: {o!r} | PREFERRED: {p!r} | CURIE: {c!r}"
+        has_bert: bool = "bert_similarity" in pending.columns
+        curies: list[str] = pending.get_column(col).to_list()
+        originals_list: list[str] = pending.get_column(original).to_list()
+        preferreds_list: list[str] = pending.get_column(preferred).to_list()
+        fuzz_ratios: list[object] = pending.get_column("fuzz_ratio").to_list()
+        fuzz_partials: list[object] = pending.get_column("fuzz_partial").to_list()
+        bert_sims: list[object] = pending.get_column("bert_similarity").to_list() if has_bert else []
+
+        for i, c in enumerate(curies):
+            msg: str = (
+                f"FAILED | STORE: {section_hash} | CONFIG: {config_file} | COL: {col}"
+                f" | ORIGINAL: {originals_list[i]!r} | PREFERRED: {preferreds_list[i]!r} | CURIE: {c!r}"
+                f" | FUZZ_RATIO: {fuzz_ratios[i]} | FUZZ_PARTIAL: {fuzz_partials[i]}"
             )
+            if has_bert:
+                msg = f"{msg} | BERT_SIMILARITY: {bert_sims[i]}"
+            logger.info(msg)
 
     return df.join(passed.select(col), on=col, how="semi").lazy()

{tablassert-7.4.7 → tablassert-7.4.8}/tests/test_qc.py

@@ -108,9 +108,7 @@ def test_fullmap_audit_suppresses_logs(monkeypatch: Any) -> None:
     monkeypatch.setattr(rf_process, "cpdist", fake_cpdist)
     monkeypatch.setattr(pairwise, "cosine_similarity", fake_cosine_similarity)
 
-    lf: pl.LazyFrame = pl.DataFrame(
-        {"subject": ["MONDO:1"], "original subject": ["foo"], "subject name": ["bar"]}
-    ).lazy()
+    lf: pl.LazyFrame = pl.DataFrame({"subject": ["FOO:1"], "original subject": ["foo"], "subject name": ["bar"]}).lazy()
     result: pl.DataFrame = qc.fullmap_audit(lf, "subject", "store123", "config.yaml", log=False).collect()
 
     assert result.height == 0
@@ -141,9 +139,7 @@ def test_fullmap_audit_logs_failures(monkeypatch: Any) -> None:
     monkeypatch.setattr(rf_process, "cpdist", fake_cpdist)
     monkeypatch.setattr(pairwise, "cosine_similarity", fake_cosine_similarity)
 
-    lf: pl.LazyFrame = pl.DataFrame(
-        {"subject": ["MONDO:1"], "original subject": ["foo"], "subject name": ["bar"]}
-    ).lazy()
+    lf: pl.LazyFrame = pl.DataFrame({"subject": ["FOO:1"], "original subject": ["foo"], "subject name": ["bar"]}).lazy()
     result: pl.DataFrame = qc.fullmap_audit(lf, "subject", "store123", "config.yaml", log=True).collect()
 
     assert result.height == 0
@@ -151,6 +147,42 @@ def test_fullmap_audit_logs_failures(monkeypatch: Any) -> None:
     assert "FAILED" in messages[0]
     assert "STORE: store123" in messages[0]
     assert "CONFIG: config.yaml" in messages[0]
+    assert "FUZZ_RATIO:" in messages[0]
+    assert "FUZZ_PARTIAL:" in messages[0]
+    assert "BERT_SIMILARITY:" in messages[0]
+
+
+# ? fullmap_audit Log Message Contains Expected Score Values
+def test_fullmap_audit_log_score_values(monkeypatch: Any) -> None:
+    messages: list[str] = []
+
+    class DummyLogger:
+        def info(self, message: str) -> None:
+            messages.append(message)
+
+    class DummyBioBERT:
+        def encode(self, values: list[str]) -> object:
+            return np.array([[0.0], [1.0]])
+
+    def fake_cpdist(left: list[str], right: list[str], scorer: Any) -> object:
+        return np.array([5.0]) if scorer == fuzz.ratio else np.array([8.0])
+
+    def fake_cosine_similarity(left: object, right: object) -> object:
+        return np.array([[0.05]])
+
+    monkeypatch.setattr(qc, "logger", DummyLogger())
+    monkeypatch.setattr(qc, "get_biobert", lambda provider=None: DummyBioBERT())
+    monkeypatch.setattr(rf_process, "cpdist", fake_cpdist)
+    monkeypatch.setattr(pairwise, "cosine_similarity", fake_cosine_similarity)
+
+    lf: pl.LazyFrame = pl.DataFrame({"subject": ["X:1"], "original subject": ["foo"], "subject name": ["bar"]}).lazy()
+    result: pl.DataFrame = qc.fullmap_audit(lf, "subject", "store456", "cfg.yaml", log=True).collect()
+
+    assert result.height == 0
+    assert len(messages) == 1
+    assert "FUZZ_RATIO: 5.0" in messages[0]
+    assert "FUZZ_PARTIAL: 8.0" in messages[0]
+    assert "BERT_SIMILARITY: 0.05" in messages[0]
 
 
 # ? GPU Runtime Can Be Forced To CPU

{tablassert-7.4.7 → tablassert-7.4.8}/uv.lock

@@ -2360,7 +2360,7 @@ wheels = [
 
 [[package]]
 name = "tablassert"
-version = "7.4.7"
+version = "7.4.8"
 source = { editable = "." }
 dependencies = [
     { name = "cyclopts" },