Merge pull request #12 from aleph-alpha-intelligence-layer/iterate-on…

…-mindset-classifier

Remove limit from FilterSearch

src/intelligence_layer/connectors/retrievers/in_memory_retriever.py

@@ -126,14 +126,14 @@ def _add_texts_to_memory(self, documents: Sequence[Document]) -> None:
         )
 
     def get_filtered_documents_with_scores(
-        self, query: str, limit: int, filter: models.Filter
+        self, query: str, filter: models.Filter
     ) -> Sequence[SearchResult]:
         """Specific method for `InMemoryRetriever` to support filtering search results."""
         query_embedding = self._embed(query, self._query_representation)
         search_result = self._search_client.search(
             collection_name=self._collection_name,
             query_vector=query_embedding,
-            limit=limit,
+            limit=self._k,
             query_filter=filter,
         )
         return [self._point_to_search_result(point) for point in search_result]

src/intelligence_layer/use_cases/classify/embedding_based_classify.py

@@ -12,7 +12,7 @@
     RetrieverType,
 )
 from intelligence_layer.core.logger import DebugLogger
-from intelligence_layer.core.task import Chunk, Probability, Task
+from intelligence_layer.core.task import Chunk, Probability
 from intelligence_layer.use_cases.classify.classify import (
     Classify,
     ClassifyInput,
@@ -111,7 +111,7 @@ def __init__(
         retriever = InMemoryRetriever(
             client,
             documents=documents,
-            k=len(documents),
+            k=scoring.value,
             retriever_type=RetrieverType.SYMMETRIC,
         )
         self._filter_search = FilterSearch(retriever)
@@ -149,7 +149,6 @@ def _label_search(
     ) -> SearchOutput:
         search_input = FilterSearchInput(
             query=chunk,
-            limit=self._scoring.value,
             filter=models.Filter(
                 must=[
                     models.FieldCondition(

src/intelligence_layer/use_cases/search/filter_search.py

@@ -14,12 +14,10 @@ class FilterSearchInput(BaseModel):
 
     Attributes:
         query: The text to be searched with.
-        limit: The maximum number of items to be retrieved.
         filter: Conditions to filter by as offered by Qdrant.
     """
 
     query: str
-    limit: int
     filter: Filter
 
 
@@ -44,7 +42,6 @@ class FilterSearch(Task[FilterSearchInput, SearchOutput]):
         >>> task = FilterSearch(retriever)
         >>> input = FilterSearchInput(
         >>>     query="When did East and West Germany reunite?"
-        >>>     limit=1,
         >>>     filter=models.Filter(
         >>>         must=[
         >>>             models.FieldCondition(
@@ -64,6 +61,6 @@ def __init__(self, in_memory_retriever: InMemoryRetriever):
 
     def run(self, input: FilterSearchInput, logger: DebugLogger) -> SearchOutput:
         results = self._in_memory_retriever.get_filtered_documents_with_scores(
-            input.query, input.limit, input.filter
+            input.query, input.filter
         )
         return SearchOutput(results=results)

tests/use_cases/classify/test_embedding_based_classify.py

@@ -63,6 +63,39 @@ def test_embedding_based_classify_raises_for_unknown_label(
         embedding_based_classify.run(classify_input, NoOpDebugLogger())
 
 
+def test_embedding_based_classify_works_for_empty_labels_in_request(
+    embedding_based_classify: EmbeddingBasedClassify,
+) -> None:
+    classify_input = ClassifyInput(
+        chunk=Chunk("This is good"),
+        labels=frozenset(),
+    )
+    result = embedding_based_classify.run(classify_input, NoOpDebugLogger())
+    assert result.scores == {}
+
+
+def test_embedding_based_classify_works_without_examples(
+    client: Client,
+) -> None:
+    labels_with_examples = [
+        LabelWithExamples(
+            name="positive",
+            examples=[],
+        ),
+        LabelWithExamples(
+            name="negative",
+            examples=[],
+        ),
+    ]
+    embedding_based_classify = EmbeddingBasedClassify(labels_with_examples, client)
+    classify_input = ClassifyInput(
+        chunk=Chunk("This is good"),
+        labels=frozenset(),
+    )
+    result = embedding_based_classify.run(classify_input, NoOpDebugLogger())
+    assert result.scores == {}
+
+
 def test_can_evaluate_embedding_based_classify(
     embedding_based_classify: EmbeddingBasedClassify,
 ) -> None:

tests/use_cases/search/test_filter_search.py

@@ -22,8 +22,8 @@ def in_memory_retriever_documents() -> Sequence[Document]:
             metadata={"type": "doc"},
         ),
         Document(
-            text="Cats are small animals. Well, I do not fit at all but I am of the correct type.",
-            metadata={"type": "doc"},
+            text="Cats are small animals. Well, I do not fit at all and I am of the correct type.",
+            metadata={"type": "no doc"},
         ),
         Document(
             text="Germany reunited in 1990. This document fits perfectly but it is of the wrong type.",
@@ -44,7 +44,6 @@ def test_filter_search(
 ) -> None:
     search_input = FilterSearchInput(
         query="When did Germany reunite?",
-        limit=1,
         filter=models.Filter(
             must=[
                 models.FieldCondition(