soupdb-vector 0.1.0__tar.gz

soupdb_vector-0.1.0/.gitignore

@@ -0,0 +1,270 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+docs/PROCESAMIENTO-CONSULTAS.md
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+shell.nix
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+/docs/**/*.md
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+# .idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#   that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#   you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# Artefactos generados
+paper/soupdb.pdf
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml
+
+# Agents
+.agents/
+.claude/
+.cursor/
+.codeium/
+.windsurf/
+
+# Agentes autónomos y CLI de IA
+.aider*
+.gpt-pilot/
+.openinterpreter/
+
+# Caché de LLMs y Bases de Datos Vectoriales Locales
+.langchain/
+.llamaindex/
+*.db
+*.log
+.chroma/
+
+# Historiales de chat generados por extensiones (Copilot, Claude, Gemini)
+*.chat.md
+*.chat.txt
+.chat-history/
+.copilot_history/
+copilot-chat/
+
+# Configuraciones de agentes locales (para evitar filtrar API keys de OpenAI/Anthropic)
+.llm_config.json
+.oai_keys
+aider.conf.yml
+
+# OS
+.DS_Store
+Thumbs.db
+
+# IDEs genéricos
+.vscode/*
+!.vscode/settings.json
+!.vscode/tasks.json
+!.vscode/launch.json
+.idea/
+
+# Datos multimedia generados (manifests, caché de descriptores, codebooks, vectores.bin).
+# Se regeneran con benchmarks/scripts/{prepare_multimedia_datasets,build_multimedia_vectors}.py
+benchmarks/data/

soupdb_vector-0.1.0/PKG-INFO

@@ -0,0 +1,13 @@
+Metadata-Version: 2.5
+Name: soupdb-vector
+Version: 0.1.0
+Summary: SoupDB vector plugin: persistent IVF/HNSW indices and SIFT/MFCC features.
+License-Expression: MIT
+Requires-Python: >=3.11
+Requires-Dist: librosa>=0.11.0
+Requires-Dist: numpy>=2.4.6
+Requires-Dist: opencv-python-headless>=5.0.0.93
+Requires-Dist: scikit-learn>=1.9.1
+Requires-Dist: soundfile>=0.14.0
+Requires-Dist: soupdb
+Requires-Dist: threadpoolctl>=3.7.0

soupdb_vector-0.1.0/pyproject.toml

@@ -0,0 +1,26 @@
+[project]
+name = "soupdb-vector"
+version = "0.1.0"
+description = "SoupDB vector plugin: persistent IVF/HNSW indices and SIFT/MFCC features."
+requires-python = ">=3.11"
+license = "MIT"
+dependencies = [
+  "soupdb",
+  "librosa>=0.11.0",
+  "numpy>=2.4.6",
+  "opencv-python-headless>=5.0.0.93",
+  "scikit-learn>=1.9.1",
+  "soundfile>=0.14.0",
+  "threadpoolctl>=3.7.0",
+]
+
+[project.entry-points."soupdb.plugins"]
+ivf = "soupdb_vector.plugin:IVFPlugin"
+hnsw = "soupdb_vector.plugin:HNSWPlugin"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["soupdb_vector"]

soupdb_vector-0.1.0/soupdb_vector/__init__.py

@@ -0,0 +1,3 @@
+"""SoupDB vector plugin package: IVF/HNSW indices and multimedia features."""
+
+__all__ = ["indexes", "multimedia", "plugin"]

soupdb_vector-0.1.0/soupdb_vector/indexes/__init__.py

@@ -0,0 +1,18 @@
+"""Vector indexes for Parte 4: paged vector store, brute force, IVF and HNSW."""
+
+from soupdb_vector.indexes.brute_force import BruteForceIndex
+from soupdb_vector.indexes.hnsw import HNSWIndex
+from soupdb_vector.indexes.ivf import IVFIndex
+from soupdb_vector.indexes.metrics import COSINE, EUCLIDEAN, INNER_PRODUCT, METRICS
+from soupdb_vector.indexes.store import VectorStore
+
+__all__ = [
+    "COSINE",
+    "EUCLIDEAN",
+    "INNER_PRODUCT",
+    "METRICS",
+    "BruteForceIndex",
+    "HNSWIndex",
+    "IVFIndex",
+    "VectorStore",
+]

soupdb_vector-0.1.0/soupdb_vector/indexes/_record_file.py

@@ -0,0 +1,154 @@
+"""Array of fixed-size records over pages of a DiskManager.
+
+Page 0 is a header (magic, record size, record count and a free area for the
+owner's metadata); record ``i`` lives at page ``1 + i // per_page``. Fixed
+size lets a record be addressed by arithmetic instead of a slot directory,
+which is what vectors and HNSW adjacency lists need.
+"""
+
+import struct
+from collections.abc import Iterator
+
+from engine.storage.buffer_manager import BufferManager
+from engine.storage.disk_manager import DiskManager
+
+_HEADER = struct.Struct("<4sHIQ")
+_VERSION = 1
+
+
+def peek_meta(buffer_manager: BufferManager, magic: bytes, layout: struct.Struct) -> tuple:
+    """Read the owner's metadata of an existing file before knowing its record size."""
+    frame = buffer_manager.pin(0)
+    try:
+        stored_magic = _HEADER.unpack_from(frame, 0)[0]
+        values = layout.unpack_from(frame, _HEADER.size)
+    finally:
+        buffer_manager.unpin(0)
+    if stored_magic != magic:
+        raise ValueError(f"not a {magic!r} file")
+    return values
+
+
+class FixedRecordFile:
+    """Append-only array of fixed-size records, read and written in place."""
+
+    def __init__(
+        self,
+        disk_manager: DiskManager,
+        buffer_manager: BufferManager,
+        *,
+        record_size: int,
+        magic: bytes,
+    ) -> None:
+        page_size = disk_manager.page_size
+        if record_size <= 0 or record_size > page_size:
+            raise ValueError(
+                f"record of {record_size} bytes does not fit in a {page_size}-byte page"
+            )
+        if len(magic) != 4:
+            raise ValueError("magic must be exactly 4 bytes")
+
+        self._disk = disk_manager
+        self._buffer = buffer_manager
+        self._record_size = record_size
+        self._per_page = page_size // record_size
+        self._magic = magic
+
+        if disk_manager.page_count == 0:
+            disk_manager.allocate_page()
+            self._count = 0
+            self._write_header()
+        else:
+            frame = buffer_manager.pin(0)
+            try:
+                stored_magic, version, stored_size, count = _HEADER.unpack_from(frame, 0)
+            finally:
+                buffer_manager.unpin(0)
+            if stored_magic != magic or version != _VERSION:
+                raise ValueError(f"{disk_manager.path} is not a {magic!r} file")
+            if stored_size != record_size:
+                raise ValueError(f"record size on disk is {stored_size}, expected {record_size}")
+            self._count = count
+
+    def __len__(self) -> int:
+        return self._count
+
+    @property
+    def per_page(self) -> int:
+        return self._per_page
+
+    @property
+    def meta_capacity(self) -> int:
+        return self._disk.page_size - _HEADER.size
+
+    def read_meta(self, layout: struct.Struct) -> tuple:
+        frame = self._buffer.pin(0)
+        try:
+            return layout.unpack_from(frame, _HEADER.size)
+        finally:
+            self._buffer.unpin(0)
+
+    def write_meta(self, layout: struct.Struct, *values: object) -> None:
+        if layout.size > self.meta_capacity:
+            raise ValueError("metadata does not fit in the header page")
+        frame = self._buffer.pin(0)
+        layout.pack_into(frame, _HEADER.size, *values)
+        self._buffer.unpin(0, dirty=True)
+
+    def append(self, layout: struct.Struct, *values: object) -> int:
+        index = self._count
+        page_id, offset = self._locate(index)
+        if page_id == self._disk.page_count:
+            self._disk.allocate_page()
+        frame = self._buffer.pin(page_id)
+        layout.pack_into(frame, offset, *values)
+        self._buffer.unpin(page_id, dirty=True)
+        self._count += 1
+        self._write_header()
+        return index
+
+    def unpack(self, index: int, layout: struct.Struct, extra: int = 0) -> tuple:
+        """Unpack ``layout`` at byte ``extra`` of record ``index``."""
+        page_id, offset = self._locate(self._check(index))
+        frame = self._buffer.pin(page_id)
+        try:
+            return layout.unpack_from(frame, offset + extra)
+        finally:
+            self._buffer.unpin(page_id)
+
+    def pack(self, index: int, layout: struct.Struct, *values: object, extra: int = 0) -> None:
+        page_id, offset = self._locate(self._check(index))
+        frame = self._buffer.pin(page_id)
+        layout.pack_into(frame, offset + extra, *values)
+        self._buffer.unpin(page_id, dirty=True)
+
+    def scan(self, layout: struct.Struct) -> Iterator[tuple[int, tuple]]:
+        """Yield (index, values) for every record, pinning each page once."""
+        index = 0
+        page_id = 1
+        while index < self._count:
+            frame = self._buffer.pin(page_id)
+            try:
+                in_page = min(self._per_page, self._count - index)
+                for slot in range(in_page):
+                    yield index + slot, layout.unpack_from(frame, slot * self._record_size)
+            finally:
+                self._buffer.unpin(page_id)
+            index += in_page
+            page_id += 1
+
+    def flush(self) -> None:
+        self._buffer.flush_all()
+
+    def _locate(self, index: int) -> tuple[int, int]:
+        return 1 + index // self._per_page, (index % self._per_page) * self._record_size
+
+    def _check(self, index: int) -> int:
+        if not 0 <= index < self._count:
+            raise IndexError(f"record {index} out of range (count {self._count})")
+        return index
+
+    def _write_header(self) -> None:
+        frame = self._buffer.pin(0)
+        _HEADER.pack_into(frame, 0, self._magic, _VERSION, self._record_size, self._count)
+        self._buffer.unpin(0, dirty=True)

soupdb_vector-0.1.0/soupdb_vector/indexes/brute_force.py

@@ -0,0 +1,68 @@
+"""Exact k-NN by scanning every stored vector.
+
+It is the sequential baseline of the comparison 2.4.4 and the oracle the
+recall of IVF and HNSW is measured against.
+"""
+
+import heapq
+from collections.abc import Sequence
+from pathlib import Path
+from typing import Any
+
+from engine.common.rid import RID
+from engine.indexes.base import RankedIndex
+from soupdb_vector.indexes.constants import DEFAULT_METRIC, VECTOR_BUFFER_CAPACITY
+from soupdb_vector.indexes.metrics import norm, score_fn
+from soupdb_vector.indexes.store import VectorStore
+
+
+class BruteForceIndex(RankedIndex):
+    """RankedIndex that scores every vector; memory is O(k), not O(N)."""
+
+    def __init__(self, store: VectorStore) -> None:
+        self._store = store
+
+    @classmethod
+    def open(
+        cls,
+        path: str | Path,
+        dim: int | None = None,
+        *,
+        page_size: int = 4096,
+        buffer_capacity: int = VECTOR_BUFFER_CAPACITY,
+    ) -> "BruteForceIndex":
+        return cls(
+            VectorStore.open(path, dim, page_size=page_size, buffer_capacity=buffer_capacity)
+        )
+
+    @property
+    def store(self) -> VectorStore:
+        return self._store
+
+    def __len__(self) -> int:
+        return len(self._store)
+
+    def insert(self, vector: Sequence[float], rid: RID) -> None:
+        self._store.append(vector, rid)
+
+    def top_k(
+        self, query: Sequence[float], k: int, *, metric: str = DEFAULT_METRIC
+    ) -> list[tuple[RID, float]]:
+        if k < 0:
+            raise ValueError(f"k must be non-negative, got {k}")
+        self._store.check_dim(query)
+        scorer = score_fn(metric)
+        query_norm = norm(query)
+        scored = (
+            (scorer(query, query_norm, vector, vector_norm), rid)
+            for _vid, rid, vector, vector_norm in self._store.scan()
+        )
+        best = heapq.nlargest(k, scored, key=_by_score)
+        return [(rid, value) for value, rid in best]
+
+    def close(self) -> None:
+        self._store.close()
+
+
+def _by_score(pair: tuple[float, Any]) -> float:
+    return pair[0]

soupdb_vector-0.1.0/soupdb_vector/indexes/constants.py

@@ -0,0 +1,34 @@
+"""Tunable parameters of the vector indexes.
+
+Kept apart so the benchmarks can vary them without touching the structures.
+"""
+
+#: Métrica por defecto cuando la consulta no indica ``metric``.
+DEFAULT_METRIC = "euclidean"
+
+#: Páginas en el buffer pool de cada archivo cuando el índice abre sus propios managers.
+VECTOR_BUFFER_CAPACITY = 64
+
+#: Número de listas (centroides) del IVF; la regla habitual es ~sqrt(N).
+IVF_NLIST = 32
+
+#: Listas que se visitan por consulta.
+IVF_NPROBE = 4
+
+#: Iteraciones de Lloyd al entrenar los centroides.
+IVF_KMEANS_ITERATIONS = 20
+
+#: Tope de la muestra con la que se entrena K-Means; acota la memoria del entrenamiento.
+IVF_TRAIN_SAMPLE = 10_000
+
+#: Vecinos por nodo en las capas superiores de HNSW; la capa 0 guarda el doble.
+HNSW_M = 16
+
+#: Tamaño de la lista de candidatos al insertar.
+HNSW_EF_CONSTRUCTION = 100
+
+#: Tamaño de la lista de candidatos al consultar.
+HNSW_EF_SEARCH = 64
+
+#: Tope de niveles del grafo; con M=16 un nivel 16 es astronómicamente improbable.
+HNSW_MAX_LEVEL = 16