Null fix (#41)

* fix empty input

Author: Gautam8387

cyteonto/cyteonto.py

@@ -40,6 +40,10 @@ def _api_key_for_provider(provider: str, fallback: str = "") -> str | None:
     return fallback or None
 
 
+def _is_empty(label: str) -> bool:
+    return not label or not label.strip()
+
+
 class CyteOnto:
     """Compare two sets of cell type annotations against the Cell Ontology.
 
@@ -345,7 +349,9 @@ async def _embed_user_labels(
             lbl: d for lbl, d in (raw_existing or {}).items() if not d.is_blank()
         }
 
-        all_real = use_cache and all(lbl in existing for lbl in labels)
+        all_real = use_cache and all(
+            _is_empty(lbl) or lbl in existing for lbl in labels
+        )
         if all_real:
             cached = storage.load_user_embeddings(emb_path)
             if cached is not None and cached[1] == labels:
@@ -355,7 +361,9 @@ async def _embed_user_labels(
                 return cached[0]
 
         unique_labels = list(dict.fromkeys(labels))
-        missing = [lbl for lbl in unique_labels if lbl not in existing]
+        missing = [
+            lbl for lbl in unique_labels if not _is_empty(lbl) and lbl not in existing
+        ]
         if missing:
             logger.info(
                 f"Generating {len(missing)} new descriptions for '{identifier}' "
@@ -369,13 +377,17 @@ async def _embed_user_labels(
                     existing[lbl] = desc
             storage.save_descriptions(desc_path, existing, self.llm_key)
 
-        # Build the text to embed for every label position. Blanks are not
-        # cached, so the raw label text is used as a fallback to keep the
-        # array aligned. Next compare(...) run will retry description
-        # generation for those labels and overwrite.
-        texts: list[str] = []
+        # Build the text to embed for every label position. Empty labels are
+        # skipped entirely (no description, no embedding) and get a zero
+        # vector below. Blanks are not cached, so the raw label text is used
+        # as a fallback to keep the array aligned. Next compare(...) run will
+        # retry description generation for those labels and overwrite.
+        texts: list[str | None] = []
         fallback_count = 0
         for lbl in labels:
+            if _is_empty(lbl):
+                texts.append(None)
+                continue
             desc = existing.get(lbl, CellDescription.blank(label=lbl))
             if desc is not None and not desc.is_blank():
                 texts.append(desc.to_sentence())
@@ -389,30 +401,37 @@ async def _embed_user_labels(
                 "label text as a fallback; they will be retried on the next run."
             )
 
-        # Only embed each unique text once, then fan results back out so the
-        # final array stays aligned with the original `labels` order/length.
+        # Only embed each unique non-empty text once, then fan results back
+        # out so the final array stays aligned with `labels` order/length.
         text_to_idx: dict[str, int] = {}
         unique_texts: list[str] = []
         for t in texts:
+            if t is None:
+                continue
             if t not in text_to_idx:
                 text_to_idx[t] = len(unique_texts)
                 unique_texts.append(t)
-        if len(unique_texts) < len(texts):
+        if len(unique_texts) < len([t for t in texts if t is not None]):
             logger.info(
                 f"Embedding {len(unique_texts)} unique texts for '{identifier}' "
                 f"({len(texts)} total label positions)"
             )
 
-        unique_embeddings = await self._embed_with_failover(unique_texts)
-        if unique_embeddings is None:
-            raise RuntimeError(f"Failed to embed labels for '{identifier}'")
+        if unique_texts:
+            unique_embeddings = await self._embed_with_failover(unique_texts)
+            if unique_embeddings is None:
+                raise RuntimeError(f"Failed to embed labels for '{identifier}'")
+            dim = unique_embeddings.shape[1]
+        else:
+            dim = self._load_ontology_embeddings()[0].shape[1]
+
         emb_path = self.paths.user_embeddings(
             run_id, kind, identifier, self.llm_key, self.embd_key
         )
-        fan_out_idx = np.fromiter(
-            (text_to_idx[t] for t in texts), dtype=np.int64, count=len(texts)
-        )
-        embeddings = unique_embeddings[fan_out_idx]
+        embeddings = np.zeros((len(labels), dim), dtype=np.float32)
+        for j, t in enumerate(texts):
+            if t is not None:
+                embeddings[j] = unique_embeddings[text_to_idx[t]]
 
         storage.save_user_embeddings(
             emb_path,
@@ -530,6 +549,7 @@ async def compare(
             use_cache=use_cache,
         )
         author_matches = self._match(author_emb, min_similarity=min_match_similarity)
+        author_empty = [_is_empty(lbl) for lbl in author_labels]
         similarity = self._ensure_similarity()
 
         rows: list[dict[str, Any]] = []
@@ -548,18 +568,23 @@ async def compare(
                 use_cache=use_cache,
             )
             algo_matches = self._match(algo_emb, min_similarity=min_match_similarity)
+            algo_empty = [_is_empty(lbl) for lbl in algo_labels]
 
             for i, (a_lbl, g_lbl) in enumerate(zip(author_labels, algo_labels)):
-                a_id, a_sim = author_matches[i]
-                g_id, g_sim = algo_matches[i]
+                a_id, a_sim = ("", 0.0) if author_empty[i] else author_matches[i]
+                g_id, g_sim = ("", 0.0) if algo_empty[i] else algo_matches[i]
                 hier = (
                     similarity.similarity(
                         a_id, g_id, metric=metric, metric_params=metric_params
                     )
                     if a_id and g_id
                     else 0.0
                 )
-                method = self._method_for(a_id, g_id, hier)
+                method = (
+                    "empty"
+                    if (author_empty[i] or algo_empty[i])
+                    else self._method_for(a_id, g_id, hier)
+                )
                 rows.append(
                     {
                         "run_id": run_id,

modal_app/api.py

@@ -133,7 +133,7 @@ def get_result(run_id: str, format: str = "json"):
         if status["state"] != "completed":
             raise HTTPException(
                 409,
-                f"Run is '{status['state']}', not completed",
+                f"Job is '{status['state']}', not completed",
             )
 
         rel = status["resultJsonPath"] if format == "json" else status["resultCsvPath"]

tests/test_cyteonto.py

@@ -1,10 +1,14 @@
 """Tests for cyteonto.cyteonto pure units (no live agents or network)."""
 
 from pathlib import Path
+from unittest.mock import AsyncMock, Mock
 
 import numpy as np
+import pytest
 
-from cyteonto.cyteonto import CyteOnto, _api_key_for_provider
+from cyteonto import storage
+from cyteonto.cyteonto import CyteOnto, _api_key_for_provider, _is_empty
+from cyteonto.models import AgentUsage, CellDescription
 
 
 class TestApiKeyForProvider:
@@ -43,9 +47,7 @@ class TestMatch:
     def _instance(self):
         # Bypass __init__ to test the pure matching logic in isolation.
         inst = object.__new__(CyteOnto)
-        inst._ontology_embeddings = np.array(
-            [[1.0, 0.0], [0.0, 1.0]], dtype=np.float32
-        )
+        inst._ontology_embeddings = np.array([[1.0, 0.0], [0.0, 1.0]], dtype=np.float32)
         inst._ontology_ids = ["CL:0000001", "CL:0000002"]
         return inst
 
@@ -57,9 +59,7 @@ def test_exact_match(self):
 
     def test_below_threshold_returns_none(self):
         inst = self._instance()
-        out = inst._match(
-            np.array([[0.7, 0.7]], dtype=np.float32), min_similarity=0.99
-        )
+        out = inst._match(np.array([[0.7, 0.7]], dtype=np.float32), min_similarity=0.99)
         assert out[0][0] is None
 
     def test_one_dimensional_query_reshaped(self):
@@ -76,3 +76,122 @@ def test_counts_only_files(self, temp_dir: Path):
         nested.mkdir()
         (nested / "b.txt").write_text("y")
         assert CyteOnto._count_files(temp_dir) == 2
+
+
+class TestIsEmpty:
+    def test_empty_string(self):
+        assert _is_empty("") is True
+
+    def test_whitespace_only(self):
+        assert _is_empty("   ") is True
+        assert _is_empty("\t\n") is True
+
+    def test_non_empty(self):
+        assert _is_empty("T cell") is False
+        assert _is_empty(" NK cell ") is False
+
+
+class TestEmbedUserLabelsSkipsEmpty:
+    @pytest.mark.asyncio
+    async def test_empty_labels_not_described_and_zero_vectors(self, monkeypatch):
+        inst = object.__new__(CyteOnto)
+        inst.paths = Mock()
+        inst.paths.user_embeddings.return_value = Path("/tmp/emb.npz")
+        inst.paths.user_descriptions.return_value = Path("/tmp/desc.json")
+        inst.llm_key = Mock()
+        inst.embd_key = Mock()
+        inst.reasoning = False
+        inst.usage = AgentUsage(agentName="CyteOnto")
+
+        described = CellDescription(
+            initialLabel="T cell",
+            descriptiveName="CD4+ helper T lymphocyte",
+            function="Coordinates immune responses",
+            diseaseRelevance="Autoimmune disease",
+            developmentalStage="Mature",
+        )
+
+        describe_mock = AsyncMock(
+            return_value=([described], AgentUsage(agentName="CellDescriptionAgent"))
+        )
+        inst._describe_labels = describe_mock
+        inst._embed_with_failover = AsyncMock(
+            return_value=np.array([[1.0, 2.0]], dtype=np.float32)
+        )
+
+        monkeypatch.setattr(storage, "save_descriptions", lambda *a, **k: None)
+        monkeypatch.setattr(storage, "save_user_embeddings", lambda *a, **k: None)
+
+        result = await inst._embed_user_labels(
+            labels=["T cell", "", "   "],
+            run_id="run-test",
+            kind="author",
+            identifier="author",
+            use_cache=False,
+        )
+
+        # Only the single non-empty label is sent for description generation.
+        describe_mock.assert_awaited_once_with(["T cell"])
+        # Only the non-empty description sentence is embedded.
+        embed_arg = inst._embed_with_failover.await_args.args[0]
+        assert embed_arg == [described.to_sentence()]
+
+        assert result.shape == (3, 2)
+        np.testing.assert_array_equal(result[0], np.array([1.0, 2.0], dtype=np.float32))
+        np.testing.assert_array_equal(result[1], np.zeros(2, dtype=np.float32))
+        np.testing.assert_array_equal(result[2], np.zeros(2, dtype=np.float32))
+
+
+class TestCompareEmptyHandling:
+    @pytest.mark.asyncio
+    async def test_empty_positions_get_blank_id_zero_score_empty_method(self):
+        inst = object.__new__(CyteOnto)
+        inst._embed_user_labels = AsyncMock(
+            return_value=np.zeros((2, 2), dtype=np.float32)
+        )
+        inst._match = Mock(return_value=[("CL:0000001", 0.95), ("CL:0000002", 0.80)])
+        sim = Mock()
+        sim.similarity.return_value = 0.9
+        inst._ensure_similarity = Mock(return_value=sim)
+
+        df = await inst.compare(
+            author_labels=["T cell", ""],
+            algorithms={"algo0": ["B cell", ""], "algo1": ["", "NK cell"]},
+            run_id="run-test",
+        )
+
+        def row(algo: str, idx: int) -> dict:
+            return df[(df.algorithm == algo) & (df.pair_index == idx)].iloc[0].to_dict()
+
+        # Both labels present -> normal cytescore path.
+        r = row("algo0", 0)
+        assert r["author_ontology_id"] == "CL:0000001"
+        assert r["algorithm_ontology_id"] == "CL:0000001"
+        assert r["cytescore_similarity"] == 0.9
+        assert r["similarity_method"] == "cytescore"
+
+        # Both labels empty -> blank ids, zero scores, empty method.
+        r = row("algo0", 1)
+        assert r["author_ontology_id"] == ""
+        assert r["algorithm_ontology_id"] == ""
+        assert r["author_embedding_similarity"] == 0.0
+        assert r["algorithm_embedding_similarity"] == 0.0
+        assert r["cytescore_similarity"] == 0.0
+        assert r["similarity_method"] == "empty"
+
+        # Only algorithm label empty -> author side kept, algorithm blanked.
+        r = row("algo1", 0)
+        assert r["author_ontology_id"] == "CL:0000001"
+        assert r["author_embedding_similarity"] == 0.95
+        assert r["algorithm_ontology_id"] == ""
+        assert r["algorithm_embedding_similarity"] == 0.0
+        assert r["cytescore_similarity"] == 0.0
+        assert r["similarity_method"] == "empty"
+
+        # Only author label empty -> algorithm side kept, author blanked.
+        r = row("algo1", 1)
+        assert r["author_ontology_id"] == ""
+        assert r["algorithm_ontology_id"] == "CL:0000002"
+        assert r["algorithm_embedding_similarity"] == 0.80
+        assert r["cytescore_similarity"] == 0.0
+        assert r["similarity_method"] == "empty"