Push

.env

@@ -0,0 +1,19 @@
+# To overwrite these variables, create a .env.local file
+
+# The path to the directory where the data will be downloaded on machine
+LILAC_DATA_PATH=./data
+
+# Set to 1 for duckdb to use views instead of materialized tables (lower memory usage, but slower).
+DUCKDB_USE_VIEWS=0
+
+# Variables that can be set in .env.local
+#
+# Get key from https://www.cohere.ai/api-keys
+# COHERE_API_KEY=
+
+# GCS_REGION=
+# GCS_ACCESS_KEY=
+# GCS_SECRET_KEY=
+
+# Get key from https://platform.openai.com/account/api-keys
+# OPENAI_API_KEY=

Dockerfile

@@ -0,0 +1,24 @@
+# NOTE: When we upgrade to 3.11 we can use a slimmer docker image which comes with gcc.
+FROM python:3.9-bullseye
+
+# Allow statements and log messages to immediately appear in the Knative logs
+ENV PYTHONUNBUFFERED True
+
+# Set the working directory in the container.
+WORKDIR /code
+
+COPY .env .
+COPY LICENSE .
+
+# Copy static files.
+COPY /web/blueprint/build ./web/blueprint/build
+
+# Copy python files.
+COPY /src ./src/
+
+# Install the dependencies. This requires exporting requirements.txt from poetry first, which
+# happens from ./build_docker.sh.
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+CMD ["uvicorn", "src.server:app", "--host", "0.0.0.0", "--port", "5432"]

LICENSE

@@ -0,0 +1,201 @@
+ Apache License
+ Version 2.0, January 2004
+ http://www.apache.org/licenses/
+
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+ 1. Definitions.
+
+ "License" shall mean the terms and conditions for use, reproduction,
+ and distribution as defined by Sections 1 through 9 of this document.
+
+ "Licensor" shall mean the copyright owner or entity authorized by
+ the copyright owner that is granting the License.
+
+ "Legal Entity" shall mean the union of the acting entity and all
+ other entities that control, are controlled by, or are under common
+ control with that entity. For the purposes of this definition,
+ "control" means (i) the power, direct or indirect, to cause the
+ direction or management of such entity, whether by contract or
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
+ outstanding shares, or (iii) beneficial ownership of such entity.
+
+ "You" (or "Your") shall mean an individual or Legal Entity
+ exercising permissions granted by this License.
+
+ "Source" form shall mean the preferred form for making modifications,
+ including but not limited to software source code, documentation
+ source, and configuration files.
+
+ "Object" form shall mean any form resulting from mechanical
+ transformation or translation of a Source form, including but
+ not limited to compiled object code, generated documentation,
+ and conversions to other media types.
+
+ "Work" shall mean the work of authorship, whether in Source or
+ Object form, made available under the License, as indicated by a
+ copyright notice that is included in or attached to the work
+ (an example is provided in the Appendix below).
+
+ "Derivative Works" shall mean any work, whether in Source or Object
+ form, that is based on (or derived from) the Work and for which the
+ editorial revisions, annotations, elaborations, or other modifications
+ represent, as a whole, an original work of authorship. For the purposes
+ of this License, Derivative Works shall not include works that remain
+ separable from, or merely link (or bind by name) to the interfaces of,
+ the Work and Derivative Works thereof.
+
+ "Contribution" shall mean any work of authorship, including
+ the original version of the Work and any modifications or additions
+ to that Work or Derivative Works thereof, that is intentionally
+ submitted to Licensor for inclusion in the Work by the copyright owner
+ or by an individual or Legal Entity authorized to submit on behalf of
+ the copyright owner. For the purposes of this definition, "submitted"
+ means any form of electronic, verbal, or written communication sent
+ to the Licensor or its representatives, including but not limited to
+ communication on electronic mailing lists, source code control systems,
+ and issue tracking systems that are managed by, or on behalf of, the
+ Licensor for the purpose of discussing and improving the Work, but
+ excluding communication that is conspicuously marked or otherwise
+ designated in writing by the copyright owner as "Not a Contribution."
+
+ "Contributor" shall mean Licensor and any individual or Legal Entity
+ on behalf of whom a Contribution has been received by Licensor and
+ subsequently incorporated within the Work.
+
+ 2. Grant of Copyright License. Subject to the terms and conditions of
+ this License, each Contributor hereby grants to You a perpetual,
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+ copyright license to reproduce, prepare Derivative Works of,
+ publicly display, publicly perform, sublicense, and distribute the
+ Work and such Derivative Works in Source or Object form.
+
+ 3. Grant of Patent License. Subject to the terms and conditions of
+ this License, each Contributor hereby grants to You a perpetual,
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+ (except as stated in this section) patent license to make, have made,
+ use, offer to sell, sell, import, and otherwise transfer the Work,
+ where such license applies only to those patent claims licensable
+ by such Contributor that are necessarily infringed by their
+ Contribution(s) alone or by combination of their Contribution(s)
+ with the Work to which such Contribution(s) was submitted. If You
+ institute patent litigation against any entity (including a
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
+ or a Contribution incorporated within the Work constitutes direct
+ or contributory patent infringement, then any patent licenses
+ granted to You under this License for that Work shall terminate
+ as of the date such litigation is filed.
+
+ 4. Redistribution. You may reproduce and distribute copies of the
+ Work or Derivative Works thereof in any medium, with or without
+ modifications, and in Source or Object form, provided that You
+ meet the following conditions:
+
+ (a) You must give any other recipients of the Work or
+ Derivative Works a copy of this License; and
+
+ (b) You must cause any modified files to carry prominent notices
+ stating that You changed the files; and
+
+ (c) You must retain, in the Source form of any Derivative Works
+ that You distribute, all copyright, patent, trademark, and
+ attribution notices from the Source form of the Work,
+ excluding those notices that do not pertain to any part of
+ the Derivative Works; and
+
+ (d) If the Work includes a "NOTICE" text file as part of its
+ distribution, then any Derivative Works that You distribute must
+ include a readable copy of the attribution notices contained
+ within such NOTICE file, excluding those notices that do not
+ pertain to any part of the Derivative Works, in at least one
+ of the following places: within a NOTICE text file distributed
+ as part of the Derivative Works; within the Source form or
+ documentation, if provided along with the Derivative Works; or,
+ within a display generated by the Derivative Works, if and
+ wherever such third-party notices normally appear. The contents
+ of the NOTICE file are for informational purposes only and
+ do not modify the License. You may add Your own attribution
+ notices within Derivative Works that You distribute, alongside
+ or as an addendum to the NOTICE text from the Work, provided
+ that such additional attribution notices cannot be construed
+ as modifying the License.
+
+ You may add Your own copyright statement to Your modifications and
+ may provide additional or different license terms and conditions
+ for use, reproduction, or distribution of Your modifications, or
+ for any such Derivative Works as a whole, provided Your use,
+ reproduction, and distribution of the Work otherwise complies with
+ the conditions stated in this License.
+
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
+ any Contribution intentionally submitted for inclusion in the Work
+ by You to the Licensor shall be under the terms and conditions of
+ this License, without any additional terms or conditions.
+ Notwithstanding the above, nothing herein shall supersede or modify
+ the terms of any separate license agreement you may have executed
+ with Licensor regarding such Contributions.
+
+ 6. Trademarks. This License does not grant permission to use the trade
+ names, trademarks, service marks, or product names of the Licensor,
+ except as required for reasonable and customary use in describing the
+ origin of the Work and reproducing the content of the NOTICE file.
+
+ 7. Disclaimer of Warranty. Unless required by applicable law or
+ agreed to in writing, Licensor provides the Work (and each
+ Contributor provides its Contributions) on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ implied, including, without limitation, any warranties or conditions
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+ PARTICULAR PURPOSE. You are solely responsible for determining the
+ appropriateness of using or redistributing the Work and assume any
+ risks associated with Your exercise of permissions under this License.
+
+ 8. Limitation of Liability. In no event and under no legal theory,
+ whether in tort (including negligence), contract, or otherwise,
+ unless required by applicable law (such as deliberate and grossly
+ negligent acts) or agreed to in writing, shall any Contributor be
+ liable to You for damages, including any direct, indirect, special,
+ incidental, or consequential damages of any character arising as a
+ result of this License or out of the use or inability to use the
+ Work (including but not limited to damages for loss of goodwill,
+ work stoppage, computer failure or malfunction, or any and all
+ other commercial damages or losses), even if such Contributor
+ has been advised of the possibility of such damages.
+
+ 9. Accepting Warranty or Additional Liability. While redistributing
+ the Work or Derivative Works thereof, You may choose to offer,
+ and charge a fee for, acceptance of support, warranty, indemnity,
+ or other liability obligations and/or rights consistent with this
+ License. However, in accepting such obligations, You may act only
+ on Your own behalf and on Your sole responsibility, not on behalf
+ of any other Contributor, and only if You agree to indemnify,
+ defend, and hold each Contributor harmless for any liability
+ incurred by, or claims asserted against, such Contributor by reason
+ of your accepting any such warranty or additional liability.
+
+ END OF TERMS AND CONDITIONS
+
+ APPENDIX: How to apply the Apache License to your work.
+
+ To apply the Apache License to your work, attach the following
+ boilerplate notice, with the fields enclosed by brackets "[]"
+ replaced with your own identifying information. (Don't include
+ the brackets!) The text should be enclosed in the appropriate
+ comment syntax for the file format. We also recommend that a
+ file or class name and description of purpose be included on the
+ same "printed page" as the copyright notice for easier
+ identification within third-party archives.
+
+ Copyright 2023 Lilac AI Inc.
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.

requirements.txt

@@ -0,0 +1,130 @@
+aiohttp==3.8.4 ; python_version >= "3.9" and python_version < "3.10"
+aiosignal==1.3.1 ; python_version >= "3.9" and python_version < "3.10"
+anyio==3.7.0 ; python_version >= "3.9" and python_version < "3.10"
+async-timeout==4.0.2 ; python_version >= "3.9" and python_version < "3.10"
+attrs==23.1.0 ; python_version >= "3.9" and python_version < "3.10"
+blis==0.7.9 ; python_version >= "3.9" and python_version < "3.10"
+cachetools==5.3.1 ; python_version >= "3.9" and python_version < "3.10"
+catalogue==2.0.8 ; python_version >= "3.9" and python_version < "3.10"
+certifi==2023.5.7 ; python_version >= "3.9" and python_version < "3.10"
+charset-normalizer==3.1.0 ; python_version >= "3.9" and python_version < "3.10"
+click==8.1.3 ; python_version >= "3.9" and python_version < "3.10"
+cloudpickle==2.2.1 ; python_version >= "3.9" and python_version < "3.10"
+cohere==3.10.0 ; python_version >= "3.9" and python_version < "3.10"
+colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.10" and (platform_system == "Windows" or sys_platform == "win32")
+confection==0.0.4 ; python_version >= "3.9" and python_version < "3.10"
+cymem==2.0.7 ; python_version >= "3.9" and python_version < "3.10"
+cytoolz==0.12.1 ; python_version >= "3.9" and python_version < "3.10"
+dask==2023.5.1 ; python_version >= "3.9" and python_version < "3.10"
+datasets==2.12.0 ; python_version >= "3.9" and python_version < "3.10"
+decorator==5.1.1 ; python_version >= "3.9" and python_version < "3.10"
+dill==0.3.6 ; python_version >= "3.9" and python_version < "3.10"
+distributed==2023.5.1 ; python_version >= "3.9" and python_version < "3.10"
+duckdb==0.8.0 ; python_version >= "3.9" and python_version < "3.10"
+email-reply-parser==0.5.12 ; python_version >= "3.9" and python_version < "3.10"
+exceptiongroup==1.1.1 ; python_version >= "3.9" and python_version < "3.10"
+fastapi==0.95.2 ; python_version >= "3.9" and python_version < "3.10"
+filelock==3.12.0 ; python_version >= "3.9" and python_version < "3.10"
+floret==0.10.3 ; python_version >= "3.9" and python_version < "3.10"
+frozenlist==1.3.3 ; python_version >= "3.9" and python_version < "3.10"
+fsspec==2023.5.0 ; python_version >= "3.9" and python_version < "3.10"
+fsspec[http]==2023.5.0 ; python_version >= "3.9" and python_version < "3.10"
+gcsfs==2023.5.0 ; python_version >= "3.9" and python_version < "3.10"
+google-api-core==2.11.0 ; python_version >= "3.9" and python_version < "3.10"
+google-api-python-client==2.88.0 ; python_version >= "3.9" and python_version < "3.10"
+google-auth-httplib2==0.1.0 ; python_version >= "3.9" and python_version < "3.10"
+google-auth-oauthlib==1.0.0 ; python_version >= "3.9" and python_version < "3.10"
+google-auth==2.19.1 ; python_version >= "3.9" and python_version < "3.10"
+google-cloud-core==2.3.2 ; python_version >= "3.9" and python_version < "3.10"
+google-cloud-storage==2.9.0 ; python_version >= "3.9" and python_version < "3.10"
+google-crc32c==1.5.0 ; python_version >= "3.9" and python_version < "3.10"
+google-resumable-media==2.5.0 ; python_version >= "3.9" and python_version < "3.10"
+googleapis-common-protos==1.59.0 ; python_version >= "3.9" and python_version < "3.10"
+h11==0.14.0 ; python_version >= "3.9" and python_version < "3.10"
+httplib2==0.22.0 ; python_version >= "3.9" and python_version < "3.10"
+httptools==0.5.0 ; python_version >= "3.9" and python_version < "3.10"
+huggingface-hub==0.15.1 ; python_version >= "3.9" and python_version < "3.10"
+idna==3.4 ; python_version >= "3.9" and python_version < "3.10"
+importlib-metadata==6.6.0 ; python_version >= "3.9" and python_version < "3.10"
+jellyfish==0.11.2 ; python_version >= "3.9" and python_version < "3.10"
+jinja2==3.1.2 ; python_version >= "3.9" and python_version < "3.10"
+joblib==1.2.0 ; python_version >= "3.9" and python_version < "3.10"
+langcodes==3.3.0 ; python_version >= "3.9" and python_version < "3.10"
+locket==1.0.0 ; python_version >= "3.9" and python_version < "3.10"
+markupsafe==2.1.3 ; python_version >= "3.9" and python_version < "3.10"
+mpmath==1.3.0 ; python_version >= "3.9" and python_version < "3.10"
+msgpack==1.0.5 ; python_version >= "3.9" and python_version < "3.10"
+multidict==6.0.4 ; python_version >= "3.9" and python_version < "3.10"
+multiprocess==0.70.14 ; python_version >= "3.9" and python_version < "3.10"
+murmurhash==1.0.9 ; python_version >= "3.9" and python_version < "3.10"
+networkx==3.1 ; python_version >= "3.9" and python_version < "3.10"
+nltk==3.8.1 ; python_version >= "3.9" and python_version < "3.10"
+numpy==1.24.3 ; python_version >= "3.9" and python_version < "3.10"
+oauthlib==3.2.2 ; python_version >= "3.9" and python_version < "3.10"
+openai-function-call==0.0.5 ; python_version >= "3.9" and python_version < "3.10"
+openai==0.27.8 ; python_version >= "3.9" and python_version < "3.10"
+orjson==3.9.0 ; python_version >= "3.9" and python_version < "3.10"
+packaging==23.1 ; python_version >= "3.9" and python_version < "3.10"
+pandas==2.0.2 ; python_version >= "3.9" and python_version < "3.10"
+partd==1.4.0 ; python_version >= "3.9" and python_version < "3.10"
+pathy==0.10.1 ; python_version >= "3.9" and python_version < "3.10"
+pillow==9.5.0 ; python_version >= "3.9" and python_version < "3.10"
+preshed==3.0.8 ; python_version >= "3.9" and python_version < "3.10"
+protobuf==4.23.2 ; python_version >= "3.9" and python_version < "3.10"
+psutil==5.9.5 ; python_version >= "3.9" and python_version < "3.10"
+pyarrow==9.0.0 ; python_version >= "3.9" and python_version < "3.10"
+pyasn1-modules==0.3.0 ; python_version >= "3.9" and python_version < "3.10"
+pyasn1==0.5.0 ; python_version >= "3.9" and python_version < "3.10"
+pydantic==1.10.9 ; python_version >= "3.9" and python_version < "3.10"
+pyparsing==3.0.9 ; python_version >= "3.9" and python_version < "3.10"
+pyphen==0.14.0 ; python_version >= "3.9" and python_version < "3.10"
+python-dateutil==2.8.2 ; python_version >= "3.9" and python_version < "3.10"
+python-dotenv==1.0.0 ; python_version >= "3.9" and python_version < "3.10"
+pytz==2023.3 ; python_version >= "3.9" and python_version < "3.10"
+pyyaml==6.0 ; python_version >= "3.9" and python_version < "3.10"
+regex==2023.6.3 ; python_version >= "3.9" and python_version < "3.10"
+requests-oauthlib==1.3.1 ; python_version >= "3.9" and python_version < "3.10"
+requests==2.31.0 ; python_version >= "3.9" and python_version < "3.10"
+responses==0.18.0 ; python_version >= "3.9" and python_version < "3.10"
+rsa==4.9 ; python_version >= "3.9" and python_version < "3.10"
+scikit-learn==1.2.2 ; python_version >= "3.9" and python_version < "3.10"
+scipy==1.10.1 ; python_version >= "3.9" and python_version < "3.10"
+sentence-transformers==2.2.2 ; python_version >= "3.9" and python_version < "3.10"
+sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.10"
+setuptools==65.7.0 ; python_version >= "3.9" and python_version < "3.10"
+six==1.16.0 ; python_version >= "3.9" and python_version < "3.10"
+smart-open==6.3.0 ; python_version >= "3.9" and python_version < "3.10"
+sniffio==1.3.0 ; python_version >= "3.9" and python_version < "3.10"
+sortedcontainers==2.4.0 ; python_version >= "3.9" and python_version < "3.10"
+spacy-legacy==3.0.12 ; python_version >= "3.9" and python_version < "3.10"
+spacy-loggers==1.0.4 ; python_version >= "3.9" and python_version < "3.10"
+spacy==3.5.3 ; python_version >= "3.9" and python_version < "3.10"
+srsly==2.4.6 ; python_version >= "3.9" and python_version < "3.10"
+starlette==0.27.0 ; python_version >= "3.9" and python_version < "3.10"
+sympy==1.12 ; python_version >= "3.9" and python_version < "3.10"
+tblib==1.7.0 ; python_version >= "3.9" and python_version < "3.10"
+textacy==0.13.0 ; python_version >= "3.9" and python_version < "3.10"
+thinc==8.1.10 ; python_version >= "3.9" and python_version < "3.10"
+threadpoolctl==3.1.0 ; python_version >= "3.9" and python_version < "3.10"
+tokenizers==0.13.3 ; python_version >= "3.9" and python_version < "3.10"
+toolz==0.12.0 ; python_version >= "3.9" and python_version < "3.10"
+torch==2.0.1 ; python_version >= "3.9" and python_version < "3.10"
+torchvision==0.15.2 ; python_version >= "3.9" and python_version < "3.10"
+tornado==6.3.2 ; python_version >= "3.9" and python_version < "3.10"
+tqdm==4.65.0 ; python_version >= "3.9" and python_version < "3.10"
+transformers==4.29.2 ; python_version >= "3.9" and python_version < "3.10"
+typer==0.7.0 ; python_version >= "3.9" and python_version < "3.10"
+types-psutil==5.9.5.13 ; python_version >= "3.9" and python_version < "3.10"
+typing-extensions==4.6.3 ; python_version >= "3.9" and python_version < "3.10"
+tzdata==2023.3 ; python_version >= "3.9" and python_version < "3.10"
+uritemplate==4.1.1 ; python_version >= "3.9" and python_version < "3.10"
+urllib3==1.26.16 ; python_version >= "3.9" and python_version < "3.10"
+uvicorn[standard]==0.20.0 ; python_version >= "3.9" and python_version < "3.10"
+uvloop==0.17.0 ; (sys_platform != "win32" and sys_platform != "cygwin") and platform_python_implementation != "PyPy" and python_version >= "3.9" and python_version < "3.10"
+wasabi==1.1.1 ; python_version >= "3.9" and python_version < "3.10"
+watchfiles==0.19.0 ; python_version >= "3.9" and python_version < "3.10"
+websockets==11.0.3 ; python_version >= "3.9" and python_version < "3.10"
+xxhash==3.2.0 ; python_version >= "3.9" and python_version < "3.10"
+yarl==1.9.2 ; python_version >= "3.9" and python_version < "3.10"
+zict==3.0.0 ; python_version >= "3.9" and python_version < "3.10"
+zipp==3.15.0 ; python_version >= "3.9" and python_version < "3.10"

src/concepts/concept.py

@@ -0,0 +1,271 @@
+"""Defines the concept and the concept models."""
+import random
+from typing import Iterable, Literal, Optional, Union
+
+import numpy as np
+from pydantic import BaseModel, validator
+from sklearn.exceptions import NotFittedError
+from sklearn.linear_model import LogisticRegression
+
+from ..db_manager import get_dataset
+from ..embeddings.embedding import get_embed_fn
+from ..schema import Path, RichData, SignalInputType, normalize_path
+from ..signals.signal import TextEmbeddingSignal, get_signal_cls
+from ..utils import DebugTimer
+
+LOCAL_CONCEPT_NAMESPACE = 'local'
+
+# Number of randomly sampled negative examples to use for training. This is used to obtain a more
+# balanced model that works with a specific dataset.
+DEFAULT_NUM_NEG_EXAMPLES = 300
+
+
+class ConceptColumnInfo(BaseModel):
+ """Information about a dataset associated with a concept."""
+ # Namespace of the dataset.
+ namespace: str
+ # Name of the dataset.
+ name: str
+ # Path holding the text to use for negative examples.
+ path: Path
+
+ num_negative_examples = DEFAULT_NUM_NEG_EXAMPLES
+
+
+class ExampleOrigin(BaseModel):
+ """The origin of an example."""
+ # The namespace that holds the dataset.
+ dataset_namespace: str
+
+ # The name of the dataset.
+ dataset_name: str
+
+ # The id of row in the dataset that the example was added from.
+ dataset_row_id: str
+
+
+DraftId = Union[Literal['main'], str]
+DRAFT_MAIN = 'main'
+
+
+class ExampleIn(BaseModel):
+ """An example in a concept without the id (used for adding new examples)."""
+ label: bool
+ text: Optional[str]
+ img: Optional[bytes]
+ origin: Optional[ExampleOrigin]
+ # The name of the draft to put the example in. If None, puts it in the main draft.
+ draft: Optional[DraftId] = DRAFT_MAIN
+
+ @validator('text')
+ def parse_text(cls, text: str) -> str:
+ """Fixes surrogate errors in text: https://github.com/ijl/orjson/blob/master/README.md#str ."""
+ return text.encode('utf-8', 'replace').decode('utf-8')
+
+
+class Example(ExampleIn):
+ """A single example in a concept used for training a concept model."""
+ id: str
+
+
+class Concept(BaseModel):
+ """A concept is a collection of examples."""
+ # The namespace of the concept.
+ namespace: str = LOCAL_CONCEPT_NAMESPACE
+ # The name of the concept.
+ concept_name: str
+ # The type of the data format that this concept represents.
+ type: SignalInputType
+ data: dict[str, Example]
+ version: int = 0
+
+ def drafts(self) -> list[DraftId]:
+ """Gets all the drafts for the concept."""
+ drafts: set[DraftId] = set([DRAFT_MAIN]) # Always return the main draft.
+ for example in self.data.values():
+ if example.draft:
+ drafts.add(example.draft)
+ return list(sorted(drafts))
+
+
+class LogisticEmbeddingModel(BaseModel):
+ """A model that uses logistic regression with embeddings."""
+
+ class Config:
+ arbitrary_types_allowed = True
+ underscore_attrs_are_private = True
+
+ version: int = -1
+
+ # The following fields are excluded from JSON serialization, but still pickleable.
+ # See `notebooks/Toxicity.ipynb` for an example of training a concept model.
+ _model: LogisticRegression = LogisticRegression(
+ class_weight=None, C=30, tol=1e-5, warm_start=True, max_iter=1_000, n_jobs=-1)
+
+ def score_embeddings(self, embeddings: np.ndarray) -> np.ndarray:
+ """Get the scores for the provided embeddings."""
+ try:
+ return self._model.predict_proba(embeddings)[:, 1]
+ except NotFittedError:
+ return np.random.rand(len(embeddings))
+
+ def fit(self, embeddings: np.ndarray, labels: list[bool], sample_weights: list[float]) -> None:
+ """Fit the model to the provided embeddings and labels."""
+ if len(set(labels)) < 2:
+ return
+ if len(labels) != len(embeddings):
+ raise ValueError(
+ f'Length of embeddings ({len(embeddings)}) must match length of labels ({len(labels)})')
+ if len(sample_weights) != len(labels):
+ raise ValueError(
+ f'Length of sample_weights ({len(sample_weights)}) must match length of labels '
+ f'({len(labels)})')
+ self._model.fit(embeddings, labels, sample_weights)
+
+
+def draft_examples(concept: Concept, draft: DraftId) -> dict[str, Example]:
+ """Get the examples in the provided draft by overriding the main draft."""
+ draft_examples: dict[str, dict[str, Example]] = {}
+ for id, example in concept.data.items():
+ draft_examples.setdefault(example.draft or DRAFT_MAIN, {})[example.id] = example
+
+ if draft == DRAFT_MAIN:
+ return draft_examples.get(DRAFT_MAIN, {})
+
+ if draft not in draft_examples:
+ raise ValueError(
+ f'Draft {draft} not found in concept. Found drafts: {list(draft_examples.keys())}')
+
+ # Map the text of the draft to its id so we can dedup with main.
+ draft_text_ids = {example.text: id for id, example in draft_examples[draft].items()}
+
+ # Write each of examples from main to the draft examples only if the text does not appear in the
+ # draft.
+ for id, example in draft_examples[DRAFT_MAIN].items():
+ if example.text not in draft_text_ids:
+ draft_examples[draft][id] = example
+
+ return draft_examples[draft]
+
+
+class ConceptModel(BaseModel):
+ """A concept model. Stores all concept model drafts and manages syncing."""
+ # The concept that this model is for.
+ namespace: str
+ concept_name: str
+
+ # The name of the embedding for this model.
+ embedding_name: str
+ version: int = -1
+
+ # The following fields are excluded from JSON serialization, but still pickleable.
+ # Maps a concept id to the embeddings.
+ _embeddings: dict[str, np.ndarray] = {}
+ _logistic_models: dict[DraftId, LogisticEmbeddingModel] = {}
+ _negative_vectors: Optional[np.ndarray] = None
+
+ class Config:
+ arbitrary_types_allowed = True
+ underscore_attrs_are_private = True
+
+ def calibrate_on_dataset(self, column_info: ConceptColumnInfo) -> None:
+ """Calibrate the model on the embeddings in the provided vector store."""
+ db = get_dataset(column_info.namespace, column_info.name)
+ vector_store = db.get_vector_store(normalize_path(column_info.path))
+ keys = vector_store.keys()
+ num_samples = min(column_info.num_negative_examples, len(keys))
+ sample_keys = random.sample(keys, num_samples)
+ self._negative_vectors = vector_store.get(sample_keys)
+
+ def score_embeddings(self, draft: DraftId, embeddings: np.ndarray) -> np.ndarray:
+ """Get the scores for the provided embeddings."""
+ return self._get_logistic_model(draft).score_embeddings(embeddings)
+
+ def score(self, draft: DraftId, examples: Iterable[RichData]) -> list[float]:
+ """Get the scores for the provided examples."""
+ embedding_signal = get_signal_cls(self.embedding_name)()
+ if not isinstance(embedding_signal, TextEmbeddingSignal):
+ raise ValueError(f'Only text embedding signals are currently supported for concepts. '
+ f'"{self.embedding_name}" is a {type(embedding_signal)}.')
+
+ embed_fn = get_embed_fn(self.embedding_name)
+ embeddings = np.array(embed_fn(examples))
+ return self._get_logistic_model(draft).score_embeddings(embeddings).tolist()
+
+ def coef(self, draft: DraftId) -> np.ndarray:
+ """Get the coefficients of the underlying ML model."""
+ return self._get_logistic_model(draft)._model.coef_.reshape(-1)
+
+ def _get_logistic_model(self, draft: DraftId) -> LogisticEmbeddingModel:
+ """Get the logistic model for the provided draft."""
+ if draft not in self._logistic_models:
+ self._logistic_models[draft] = LogisticEmbeddingModel(
+ namespace=self.namespace,
+ concept_name=self.concept_name,
+ embedding_name=self.embedding_name,
+ version=-1)
+ return self._logistic_models[draft]
+
+ def sync(self, concept: Concept) -> bool:
+ """Update the model with the latest labeled concept data."""
+ if concept.version == self.version:
+ # The model is up to date.
+ return False
+
+ concept_path = (f'{self.namespace}/{self.concept_name}/'
+ f'{self.embedding_name}')
+ with DebugTimer(f'Computing embeddings for "{concept_path}"'):
+ self._compute_embeddings(concept)
+
+ # Fit each of the drafts, sort by draft name for deterministic behavior.
+ for draft in concept.drafts():
+ examples = draft_examples(concept, draft)
+ embeddings = np.array([self._embeddings[id] for id in examples.keys()])
+ labels = [example.label for example in examples.values()]
+ num_pos_labels = len([x for x in labels if x])
+ num_neg_labels = len([x for x in labels if not x])
+ sample_weights = [(1.0 / num_pos_labels if x else 1.0 / num_neg_labels) for x in labels]
+ if self._negative_vectors is not None:
+ num_implicit_labels = len(self._negative_vectors)
+ embeddings = np.concatenate([self._negative_vectors, embeddings])
+ labels = [False] * num_implicit_labels + labels
+ sample_weights = [1.0 / num_implicit_labels] * num_implicit_labels + sample_weights
+
+ model = self._get_logistic_model(draft)
+ with DebugTimer(f'Fitting model for "{concept_path}"'):
+ model.fit(embeddings, labels, sample_weights)
+
+ # Synchronize the model version with the concept version.
+ model.version = concept.version
+
+ # Synchronize the model version with the concept version.
+ self.version = concept.version
+
+ return True
+
+ def _compute_embeddings(self, concept: Concept) -> None:
+ embedding_signal = get_signal_cls(self.embedding_name)()
+ if not isinstance(embedding_signal, TextEmbeddingSignal):
+ raise ValueError(f'Only text embedding signals are currently supported for concepts. '
+ f'"{self.embedding_name}" is a {type(embedding_signal)}.')
+
+ embed_fn = get_embed_fn(self.embedding_name)
+ concept_embeddings: dict[str, np.ndarray] = {}
+
+ # Compute the embeddings for the examples with cache miss.
+ texts_of_missing_embeddings: dict[str, str] = {}
+ for id, example in concept.data.items():
+ if id in self._embeddings:
+ # Cache hit.
+ concept_embeddings[id] = self._embeddings[id]
+ else:
+ # Cache miss.
+ # TODO(smilkov): Support images.
+ texts_of_missing_embeddings[id] = example.text or ''
+
+ missing_ids = texts_of_missing_embeddings.keys()
+ missing_embeddings = embed_fn(list(texts_of_missing_embeddings.values()))
+
+ for id, embedding in zip(missing_ids, missing_embeddings):
+ concept_embeddings[id] = embedding
+ self._embeddings = concept_embeddings