linkml-store 0.2.0__tar.gz → 0.2.2__tar.gz

{linkml_store-0.2.0 → linkml_store-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: linkml-store
-Version: 0.2.0
+Version: 0.2.2
 Summary: linkml-store
 License: MIT
 Author: Author 1
@@ -18,6 +18,7 @@ Provides-Extra: chromadb
 Provides-Extra: fastapi
 Provides-Extra: frictionless
 Provides-Extra: h5py
+Provides-Extra: ibis
 Provides-Extra: llm
 Provides-Extra: map
 Provides-Extra: mongodb
@@ -34,7 +35,9 @@ Requires-Dist: duckdb (>=0.10.1)
 Requires-Dist: duckdb-engine (>=0.11.2)
 Requires-Dist: fastapi ; extra == "fastapi"
 Requires-Dist: frictionless ; extra == "frictionless"
+Requires-Dist: gcsfs ; extra == "ibis"
 Requires-Dist: h5py ; extra == "h5py"
+Requires-Dist: ibis-framework[duckdb,examples] (>=9.3.0) ; extra == "ibis"
 Requires-Dist: jinja2 (>=3.1.4,<4.0.0)
 Requires-Dist: jsonlines (>=4.0.0,<5.0.0)
 Requires-Dist: linkml (>=1.8.0) ; extra == "validation"
@@ -43,6 +46,7 @@ Requires-Dist: linkml_map ; extra == "map"
 Requires-Dist: linkml_renderer ; extra == "renderer"
 Requires-Dist: llm ; extra == "llm"
 Requires-Dist: matplotlib ; extra == "analytics"
+Requires-Dist: multipledispatch ; extra == "ibis"
 Requires-Dist: neo4j ; extra == "neo4j"
 Requires-Dist: networkx ; extra == "neo4j"
 Requires-Dist: pandas (>=2.2.1) ; extra == "analytics"
@@ -52,6 +56,7 @@ Requires-Dist: pyarrow ; extra == "pyarrow"
 Requires-Dist: pydantic (>=2.0.0,<3.0.0)
 Requires-Dist: pymongo ; extra == "mongodb"
 Requires-Dist: pystow (>=0.5.4,<0.6.0)
+Requires-Dist: ruff (>=0.6.2) ; extra == "tests"
 Requires-Dist: scikit-learn ; extra == "scipy"
 Requires-Dist: scipy ; extra == "scipy"
 Requires-Dist: seaborn ; extra == "analytics"

{linkml_store-0.2.0 → linkml_store-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "linkml-store"
-version = "0.2.0"
+version = "0.2.2"
 description = "linkml-store"
 authors = ["Author 1 <author@org.org>"]
 license = "MIT"
@@ -20,6 +20,7 @@ seaborn = { version = "*", optional = true }
 plotly = { version = "*", optional = true }
 pystow = "^0.5.4"
 black = { version=">=24.0.0", optional = true }
+ruff = { version=">=0.6.2", optional = true }
 llm = { version="*", optional = true }
 tiktoken = { version="*", optional = true }
 pymongo = { version="*", optional = true }
@@ -35,6 +36,9 @@ linkml = { version=">=1.8.0", optional = true }
 linkml_map = { version="*", optional = true }
 linkml_renderer = { version="*", optional = true }
 frictionless = { version="*", optional = true }
+ibis-framework = { version=">=9.3.0", extras = ["duckdb", "examples"], optional = true }
+gcsfs = { version="*", optional = true }
+multipledispatch = { version="*" }
 pandas = ">=2.2.1"
 jinja2 = "^3.1.4"
 jsonlines = "^4.0.0"
@@ -69,7 +73,7 @@ numpy = [
 [tool.poetry.extras]
 analytics = ["pandas", "matplotlib", "seaborn", "plotly"]
 app = ["streamlit"]
-tests = ["black"]
+tests = ["black", "ruff"]
 llm = ["llm", "tiktoken"]
 mongodb = ["pymongo"]
 neo4j = ["neo4j", "py2neo", "networkx"]
@@ -82,6 +86,7 @@ renderer = ["linkml_renderer"]
 fastapi = ["fastapi", "uvicorn"]
 frictionless = ["frictionless"]
 scipy = ["scipy", "scikit-learn"]
+ibis = ["ibis-framework", "multipledispatch", "gcsfs"]
 
 [tool.poetry.scripts]
 linkml-store = "linkml_store.cli:cli"
@@ -119,27 +124,13 @@ extend-exclude = [
 ]
 force-exclude = true
 line-length = 120
-extend-ignore = ["E203"]
-select = [
+lint.extend-ignore = ["E203"]
+lint.select = [
   "E",  # pycodestyle errors
   "F",  # Pyflakes
   "I",  # isort
 ]
-# Assume Python 3.8
-target-version = "py38"
 
-[tool.ruff.per-file-ignores]
-# These templates can have long lines
-"linkml/generators/sqlalchemy/sqlalchemy_declarative_template.py" = ["E501"]
-"linkml/generators/sqlalchemy/sqlalchemy_imperative_template.py" = ["E501"]
-
-# Notebooks can have unsorted imports
-"tests/test_notebooks/input/*" = ["E402"]
-
-
-[tool.ruff.mccabe]
-# Unlike Flake8, default to a complexity level of 10.
-max-complexity = 10
 
 
 [tool.codespell]

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/api/collection.py

@@ -226,6 +226,18 @@ class Collection(Generic[DatabaseType]):
         self._initialized = True
         patches = [{"op": "add", "path": "/0", "value": obj} for obj in objs]
         self._broadcast(patches, **kwargs)
+        self._post_modification_hook(**kwargs)
+
+    def _post_delete_hook(self, **kwargs):
+        self._post_modification_hook(**kwargs)
+
+    def _post_modification_hook(self, **kwargs):
+        for indexer in self.indexers.values():
+            ix_collection_name = self.get_index_collection_name(indexer)
+            ix_collection = self.parent.get_collection(ix_collection_name)
+            # Currently updating the source triggers complete reindexing
+            # TODO: make this more efficient by only deleting modified
+            ix_collection.delete_where({})
 
     def delete(self, objs: Union[OBJECT, List[OBJECT]], **kwargs) -> Optional[int]:
         """
@@ -458,6 +470,7 @@ class Collection(Generic[DatabaseType]):
         where: Optional[Any] = None,
         index_name: Optional[str] = None,
         limit: Optional[int] = None,
+        mmr_relevance_factor: Optional[float] = None,
         **kwargs,
     ) -> QueryResult:
         """
@@ -476,7 +489,7 @@ class Collection(Generic[DatabaseType]):
         Now let's index, using the simple trigram-based index
 
         >>> index = get_indexer("simple")
-        >>> collection.attach_indexer(index)
+        >>> _ = collection.attach_indexer(index)
 
         Now let's find all objects:
 
@@ -514,12 +527,15 @@ class Collection(Generic[DatabaseType]):
         if ix_coll.size() == 0:
             logger.info(f"Index {index_name} is empty; indexing all objects")
             all_objs = self.find(limit=-1).rows
-            self.index_objects(all_objs, index_name, replace=True, **kwargs)
+            if all_objs:
+                # print(f"Index {index_name} is empty; indexing all objects {len(all_objs)}")
+                self.index_objects(all_objs, index_name, replace=True, **kwargs)
+                assert ix_coll.size() > 0
         qr = ix_coll.find(where=where, limit=-1, **kwargs)
         index_col = ix.index_field
         # TODO: optimize this for large indexes
         vector_pairs = [(row, np.array(row[index_col], dtype=float)) for row in qr.rows]
-        results = ix.search(query, vector_pairs, limit=limit)
+        results = ix.search(query, vector_pairs, limit=limit, mmr_relevance_factor=mmr_relevance_factor, **kwargs)
         for r in results:
             del r[1][index_col]
         new_qr = QueryResult(num_rows=len(results))
@@ -648,7 +664,31 @@ class Collection(Generic[DatabaseType]):
         """
         return self.find({}, limit=1).num_rows
 
-    def attach_indexer(self, index: Union[Indexer, str], name: Optional[str] = None, auto_index=True, **kwargs):
+    def rows_iter(self) -> Iterable[OBJECT]:
+        """
+        Return an iterator over the objects in the collection.
+
+        :return:
+        """
+        yield from self.find({}, limit=-1).rows
+
+    def rows(self) -> List[OBJECT]:
+        """
+        Return a list of objects in the collection.
+
+        :return:
+        """
+        return list(self.rows_iter())
+
+    def ranked_rows(self) -> List[Tuple[float, OBJECT]]:
+        """
+        Return a list of objects in the collection, with scores.
+        """
+        return [(n, obj) for n, obj in enumerate(self.rows_iter())]
+
+    def attach_indexer(
+        self, index: Union[Indexer, str], name: Optional[str] = None, auto_index=True, **kwargs
+    ) -> Indexer:
         """
         Attach an index to the collection.
 
@@ -669,8 +709,8 @@ class Collection(Generic[DatabaseType]):
         >>> full_index.name = "full"
         >>> name_index = get_indexer("simple", text_template="{name}")
         >>> name_index.name = "name"
-        >>> collection.attach_indexer(full_index)
-        >>> collection.attach_indexer(name_index)
+        >>> _ = collection.attach_indexer(full_index)
+        >>> _ = collection.attach_indexer(name_index)
 
         Now let's find objects using the full index, using the string "France".
         We expect the country France to be the top hit, but the score will
@@ -713,6 +753,10 @@ class Collection(Generic[DatabaseType]):
             all_objs = self.find(limit=-1).rows
             logger.info(f"Auto-indexing {len(all_objs)} objects")
             self.index_objects(all_objs, index_name, replace=True, **kwargs)
+        return index
+
+    def get_index_collection_name(self, indexer: Indexer) -> str:
+        return self._index_collection_name(indexer.name)
 
     def _index_collection_name(self, index_name: str) -> str:
         """

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/api/database.py

@@ -268,7 +268,7 @@ class Database(ABC, Generic[CollectionType]):
         metadata: Optional[CollectionConfig] = None,
         recreate_if_exists=False,
         **kwargs,
-    ) -> CollectionType:
+    ) -> Collection:
         """
         Create a new collection in the current database.
 
@@ -760,6 +760,12 @@ class Database(ABC, Generic[CollectionType]):
         """
         Export a database to a file or location.
 
+        >>> from linkml_store.api.client import Client
+        >>> client = Client()
+        >>> db = client.attach_database("duckdb", alias="test")
+        >>> db.import_database("tests/input/iris.csv", Format.CSV, collection_name="iris")
+        >>> db.export_database("/tmp/iris.yaml", Format.YAML)
+
         :param location: location of the file
         :param target_format: target format
         :param kwargs: additional arguments

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/api/queries.py

@@ -40,7 +40,9 @@ class FacetCountResult(BaseModel):
 
 class QueryResult(BaseModel):
     """
-    A query result
+    A query result.
+
+    TODO: make this a subclass of Collection
     """
 
     query: Optional[Query] = None

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/api/stores/duckdb/duckdb_collection.py

@@ -50,8 +50,9 @@ class DuckDBCollection(Collection):
         if not isinstance(objs, list):
             objs = [objs]
         cd = self.class_definition()
-        if not cd:
+        if not cd or not cd.attributes:
             cd = self.induce_class_definition_from_objects(objs)
+        assert cd.attributes
         table = self._sqla_table(cd)
         engine = self.parent.engine
         with engine.connect() as conn:
@@ -61,7 +62,8 @@ class DuckDBCollection(Collection):
                 stmt = stmt.compile(engine)
                 conn.execute(stmt)
                 conn.commit()
-        return
+        self._post_delete_hook()
+        return None
 
     def delete_where(self, where: Optional[Dict[str, Any]] = None, missing_ok=True, **kwargs) -> Optional[int]:
         logger.info(f"Deleting from {self.target_class_name} where: {where}")
@@ -87,6 +89,7 @@ class DuckDBCollection(Collection):
             if deleted_rows_count == 0 and not missing_ok:
                 raise ValueError(f"No rows found for {where}")
             conn.commit()
+            self._post_delete_hook()
             return deleted_rows_count if deleted_rows_count > -1 else None
 
     def query_facets(

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/cli.py

@@ -1,8 +1,9 @@
 import logging
 import sys
 import warnings
+from collections import defaultdict
 from pathlib import Path
-from typing import Optional
+from typing import Optional, Tuple, Any
 
 import click
 import yaml
@@ -76,6 +77,8 @@ class ContextSettings(BaseModel):
         if name is None:
             # if len(self.database.list_collections()) > 1:
             #    raise ValueError("Collection must be specified if there are multiple collections.")
+            if not self.database:
+                return None
             if not self.database.list_collections():
                 return None
             name = list(self.database.list_collections())[0]
@@ -218,7 +221,10 @@ def insert(ctx, files, object, format):
 @click.option("--object", "-i", multiple=True, help="Input object as YAML")
 @click.pass_context
 def store(ctx, files, object, format):
-    """Store objects from files (JSON, YAML, TSV) into the specified collection."""
+    """Store objects from files (JSON, YAML, TSV) into the database.
+
+    Note: this is similar to insert, but a collection does not need to be specified
+    """
     settings = ctx.obj["settings"]
     db = settings.database
     if not files and not object:
@@ -410,14 +416,6 @@ def list_collections(ctx, **kwargs):
 def fq(ctx, where, limit, columns, output_type, wide, output):
     """
     Query facets from the specified collection.
-
-    :param ctx:
-    :param where:
-    :param limit:
-    :param columns:
-    :param output_type:
-    :param output:
-    :return:
     """
     collection = ctx.obj["settings"].collection
     where_clause = yaml.safe_load(where) if where else None
@@ -483,6 +481,41 @@ def describe(ctx, where, output_type, output, limit):
     write_output(df.describe(include="all").transpose(), output_type, target=output)
 
 
+@cli.command()
+@click.option("--where", "-w", type=click.STRING, help="WHERE clause for the query")
+@click.option("--limit", "-l", type=click.INT, help="Maximum number of results to return")
+@click.option("--output-type", "-O", type=format_choice, default="json", help="Output format")
+@click.option("--output", "-o", type=click.Path(), help="Output file path")
+@click.option("--index", "-I", help="Attributes to index on in pivot")
+@click.option("--columns", "-A", help="Attributes to use as columns in pivot")
+@click.option("--values", "-V", help="Attributes to use as values in pivot")
+@click.pass_context
+def pivot(ctx, where, limit, index, columns, values, output_type, output):
+    collection = ctx.obj["settings"].collection
+    where_clause = yaml.safe_load(where) if where else None
+    column_atts = columns.split(",") if columns else None
+    value_atts = values.split(",") if values else None
+    index_atts = index.split(",") if index else None
+    results = collection.find(where_clause, limit=limit)
+    pivoted = defaultdict(dict)
+    for row in results.rows:
+        index_key = tuple([row.get(att) for att in index_atts])
+        column_key = tuple([row.get(att) for att in column_atts])
+        value_key = tuple([row.get(att) for att in value_atts])
+        pivoted[index_key][column_key] = value_key
+    pivoted_objs = []
+    def detuple(t: Tuple) -> Any:
+        if len(t) == 1:
+            return t[0]
+        return str(t)
+    for index_key, data in pivoted.items():
+        obj = {att: key for att, key in zip(index_atts, index_key)}
+        for column_key, value_key in data.items():
+            obj[detuple(column_key)] = detuple(value_key)
+        pivoted_objs.append(obj)
+    write_output(pivoted_objs, output_type, target=output)
+
+
 @cli.command()
 @click.option("--output-type", "-O", type=format_choice, default=Format.YAML.value, help="Output format")
 @click.option("--output", "-o", type=click.Path(), help="Output file path")
@@ -499,6 +532,7 @@ def describe(ctx, where, output_type, output, limit):
     "--predictor-type", "-t", default="sklearn", show_default=True, type=click.STRING, help="Type of predictor"
 )
 @click.option("--evaluation-count", "-n", type=click.INT, help="Number of examples to evaluate over")
+@click.option("--evaluation-match-function", help="Name of function to use for matching objects in eval")
 @click.option("--query", "-q", type=click.STRING, help="query term")
 @click.pass_context
 def infer(
@@ -506,6 +540,7 @@ def infer(
     inference_config_file,
     query,
     evaluation_count,
+    evaluation_match_function,
     training_test_data_split,
     predictor_type,
     target_attribute,
@@ -549,7 +584,10 @@ def infer(
     else:
         query_obj = None
     collection = ctx.obj["settings"].collection
-    atts = collection.class_definition().attributes.keys()
+    if collection:
+        atts = collection.class_definition().attributes.keys()
+    else:
+        atts = []
     if feature_attributes:
         features = feature_attributes.split(",")
         features = [f.strip() for f in features]
@@ -575,7 +613,8 @@ def infer(
         if training_test_data_split:
             config.train_test_split = training_test_data_split
         predictor = get_inference_engine(predictor_type, config=config)
-        predictor.load_and_split_data(collection)
+        if collection:
+            predictor.load_and_split_data(collection)
         predictor.initialize_model()
     if export_model:
         logger.info(f"Exporting model to {export_model} in {model_format}")
@@ -584,8 +623,14 @@ def infer(
         if not export_model and not evaluation_count:
             raise ValueError("Query or evaluate must be specified if not exporting model")
     if evaluation_count:
+        if evaluation_match_function == "score_text_overlap":
+            match_function_fn = score_text_overlap
+        elif evaluation_match_function is not None:
+            raise ValueError(f"Unknown match function: {evaluation_match_function}")
+        else:
+            match_function_fn = None
         outcome = evaluate_predictor(
-            predictor, target_attributes, evaluation_count=evaluation_count, match_function=score_text_overlap
+            predictor, target_attributes, evaluation_count=evaluation_count, match_function=match_function_fn
         )
         print(f"Outcome: {outcome} // accuracy: {outcome.accuracy}")
     if query_obj:

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/index/implementations/llm_indexer.py

@@ -1,11 +1,13 @@
 import logging
 from pathlib import Path
-from typing import TYPE_CHECKING, List
+from typing import TYPE_CHECKING, List, Optional
 
 import numpy as np
+from tiktoken import encoding_for_model
 
 from linkml_store.api.config import CollectionConfig
 from linkml_store.index.indexer import INDEX_ITEM, Indexer
+from linkml_store.utils.llm_utils import get_token_limit, render_formatted_text
 
 if TYPE_CHECKING:
     import llm
@@ -29,6 +31,7 @@ class LLMIndexer(Indexer):
     cached_embeddings_database: str = None
     cached_embeddings_collection: str = None
     cache_queries: bool = False
+    truncation_method: Optional[str] = None
 
     @property
     def embedding_model(self):
@@ -62,6 +65,21 @@ class LLMIndexer(Indexer):
         """
         logging.info(f"Converting {len(texts)} texts to vectors")
         model = self.embedding_model
+        token_limit = get_token_limit(model.model_id)
+        encoding = encoding_for_model("gpt-4o")
+
+        def truncate_text(text: str) -> str:
+            # split into tokens every 1000 chars:
+            parts = [text[i : i + 1000] for i in range(0, len(text), 1000)]
+            return render_formatted_text(
+                lambda x: "".join(x),
+                parts,
+                encoding,
+                token_limit,
+            )
+
+        texts = [truncate_text(text) for text in texts]
+
         if self.cached_embeddings_database and (cache is None or cache or self.cache_queries):
             model_id = model.model_id
             if not model_id:
@@ -88,7 +106,7 @@ class LLMIndexer(Indexer):
                 embeddings_collection = embeddings_db.create_collection(coll_name, metadata=config)
             else:
                 embeddings_collection = embeddings_db.create_collection(coll_name, metadata=config)
-            texts = list(texts)
+
             embeddings = list([None] * len(texts))
             uncached_texts = []
             n = 0

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/index/indexer.py

@@ -3,6 +3,7 @@ from enum import Enum
 from typing import Any, Callable, Dict, List, Optional, Tuple
 
 import numpy as np
+from linkml_store.utils.vector_utils import pairwise_cosine_similarity, mmr_diversified_search
 from pydantic import BaseModel
 
 INDEX_ITEM = np.ndarray
@@ -19,23 +20,57 @@ class TemplateSyntaxEnum(str, Enum):
     fstring = "fstring"
 
 
-def cosine_similarity(vector1, vector2) -> float:
+class Indexer(BaseModel):
     """
-    Calculate the cosine similarity between two vectors
+    An indexer operates on a collection in order to search for objects.
 
-    :param vector1:
-    :param vector2:
-    :return:
-    """
-    dot_product = np.dot(vector1, vector2)
-    norm1 = np.linalg.norm(vector1)
-    norm2 = np.linalg.norm(vector2)
-    return dot_product / (norm1 * norm2)
+    You should use a subcllass of this; this can be looked up dynqamically:
 
+    >>> from linkml_store.index import get_indexer
+    >>> indexer = get_indexer("simple")
+
+    You can customize how objects are indexed by passing in a text template.
+    For example, if your collection has objects with "name" and "profession" attributes,
+    you can index them as "{name} {profession}".
+
+    >>> indexer = get_indexer("simple", text_template="{name} :: {profession}")
+
+    By default, python fstrings are assumed.
+
+    We can test this works using the :ref:`object_to_text` method (normally
+    you would never need to call this directly, but it's useful for testing):
+
+    >>> obj = {"name": "John", "profession": "doctor"}
+    >>> indexer.object_to_text(obj)
+    'John :: doctor'
+
+    You can also use Jinja2 templates; this gives more flexibility and logic,
+    e.g. conditional formatting:
+
+    >>> tmpl = "{{name}}{% if profession %} :: {{profession}}{% endif %}"
+    >>> indexer = get_indexer("simple", text_template=tmpl, text_template_syntax=TemplateSyntaxEnum.jinja2)
+    >>> indexer.object_to_text(obj)
+    'John :: doctor'
+    >>> indexer.object_to_text({"name": "John"})
+    'John'
+
+    You can also specify which attributes to index:
+
+    >>> indexer = get_indexer("simple", index_attributes=["name"])
+    >>> indexer.object_to_text(obj)
+    'John'
+
+    The purpose of an indexer is to translate a collection of objects into a collection of objects
+    such as vectors for purposes such as search. Unless you are implementing your own indexer, you
+    generally don't need to use the methods that return vectors, but we can examine their behavior
+    to get a sense of how they work.
+
+    >>> vectors = indexer.objects_to_vectors([{"name": "Aardvark"}, {"name": "Aardwolf"}, {"name": "Zesty"}])
+    >>> assert pairwise_cosine_similarity(vectors[0], vectors[1]) > pairwise_cosine_similarity(vectors[0], vectors[2])
+
+    Note you should consult the documentation for the specific indexer you are using for more details on
+    how text is converted to vectors.
 
-class Indexer(BaseModel):
-    """
-    An indexer operates on a collection in order to search for objects.
     """
 
     name: Optional[str] = None
@@ -119,10 +154,13 @@ class Indexer(BaseModel):
         return str(obj)
 
     def search(
-        self, query: str, vectors: List[Tuple[str, INDEX_ITEM]], limit: Optional[int] = None
+        self, query: str, vectors: List[Tuple[str, INDEX_ITEM]], limit: Optional[int] = None,
+            mmr_relevance_factor: Optional[float] = None
     ) -> List[Tuple[float, Any]]:
         """
-        Search the index for a query string
+        Use the indexer to search against a database of vectors.
+
+        Note: this is a low-level method, typically you would use the :ref:`search` method on a :ref:`Collection`.
 
         :param query: The query string to search for
         :param vectors: A list of indexed items, where each item is a tuple of (id, vector)
@@ -133,13 +171,29 @@ class Indexer(BaseModel):
         # Convert the query string to a vector
         query_vector = self.text_to_vector(query, cache=False)
 
+        if mmr_relevance_factor is not None:
+            vlist = [v for _, v in vectors]
+            idlist = [id for id, _ in vectors]
+            sorted_indices = mmr_diversified_search(
+                query_vector, vlist,
+                relevance_factor=mmr_relevance_factor, top_n=limit)
+            results = []
+            # TODO: this is inefficient when limit is high
+            for i in range(limit):
+                if i >= len(sorted_indices):
+                    break
+                pos = sorted_indices[i]
+                score = pairwise_cosine_similarity(query_vector, vlist[pos])
+                results.append((score, idlist[pos]))
+            return results
+
         distances = []
 
         # Iterate over each indexed item
         for item_id, item_vector in vectors:
             # Calculate the Euclidean distance between the query vector and the item vector
             # distance = 1-np.linalg.norm(query_vector - item_vector)
-            distance = cosine_similarity(query_vector, item_vector)
+            distance = pairwise_cosine_similarity(query_vector, item_vector)
             distances.append((distance, item_id))
 
         # Sort the distances in ascending order

{linkml_store-0.2.0 → linkml_store-0.2.2}/src/linkml_store/inference/evaluation.py

@@ -20,6 +20,8 @@ def score_match(target: Optional[Any], candidate: Optional[Any], match_function:
     1.0
     >>> score_match("a", "b")
     0.0
+    >>> score_match("abcd", "abcde")
+    0.0
     >>> score_match("a", None)
     0.0
     >>> score_match(None, "a")
@@ -52,7 +54,7 @@ def score_match(target: Optional[Any], candidate: Optional[Any], match_function:
 
     :param target:
     :param candidate:
-    :param match_function:
+    :param match_function: defaults to struct
     :return:
     """
     if target == candidate:
@@ -99,7 +101,8 @@ def evaluate_predictor(
     :param predictor:
     :param target_attributes:
     :param feature_attributes:
-    :param evaluation_count:
+    :param evaluation_count: max iterations
+    :param match_function: function to use for matching
     :return:
     """
     n = 0
@@ -113,8 +116,8 @@ def evaluate_predictor(
         else:
             test_obj = row
         result = predictor.derive(test_obj)
-        logger.info(f"Predicted: {result.predicted_object} Expected: {expected_obj}")
         tp += score_match(result.predicted_object, expected_obj, match_function)
+        logger.info(f"TP={tp} MF={match_function} Predicted: {result.predicted_object} Expected: {expected_obj}")
         n += 1
         if evaluation_count is not None and n >= evaluation_count:
             break
@@ -125,6 +128,9 @@ def score_text_overlap(str1: Any, str2: Any) -> float:
     """
     Compute the overlap score between two strings.
 
+    >>> score_text_overlap("abc", "bcde")
+    0.5
+
     :param str1:
     :param str2:
     :return: