datachain 0.3.0__py3-none-any.whl → 0.3.2__py3-none-any.whl

datachain/sql/types.py

@@ -17,6 +17,7 @@ from datetime import datetime
 from types import MappingProxyType
 from typing import Any, Union
 
+import sqlalchemy as sa
 from sqlalchemy import TypeDecorator, types
 
 _registry: dict[str, "TypeConverter"] = {}
@@ -28,6 +29,9 @@ read_converter_registry = MappingProxyType(_read_converter_registry)
 _type_defaults_registry: dict[str, "TypeDefaults"] = {}
 type_defaults_registry = MappingProxyType(_type_defaults_registry)
 
+_db_defaults_registry: dict[str, "DBDefaults"] = {}
+db_defaults_registry = MappingProxyType(_db_defaults_registry)
+
 NullType = types.NullType
 
 
@@ -43,6 +47,10 @@ def register_type_defaults(dialect_name: str, td: "TypeDefaults"):
     _type_defaults_registry[dialect_name] = td
 
 
+def register_db_defaults(dialect_name: str, dbd: "DBDefaults"):
+    _db_defaults_registry[dialect_name] = dbd
+
+
 def converter(dialect) -> "TypeConverter":
     name = dialect.name
     try:
@@ -71,6 +79,14 @@ def type_defaults(dialect) -> "TypeDefaults":
         raise ValueError(f"No type defaults registered for dialect: {name!r}") from None
 
 
+def db_defaults(dialect) -> "DBDefaults":
+    name = dialect.name
+    try:
+        return db_defaults_registry[name]
+    except KeyError:
+        raise ValueError(f"No DB defaults registered for dialect: {name!r}") from None
+
+
 class SQLType(TypeDecorator):
     impl: type[types.TypeEngine[Any]] = types.TypeEngine
     cache_ok = True
@@ -97,6 +113,10 @@ class String(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).string()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).string()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).string(value)
 
@@ -115,6 +135,10 @@ class Boolean(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).boolean()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).boolean()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).boolean(value)
 
@@ -133,6 +157,10 @@ class Int(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).int()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).int()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).int(value)
 
@@ -145,6 +173,10 @@ class Int32(Int):
     def default_value(dialect):
         return type_defaults(dialect).int32()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).int32()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).int32(value)
 
@@ -157,6 +189,10 @@ class Int64(Int):
     def default_value(dialect):
         return type_defaults(dialect).int64()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).int64()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).int64(value)
 
@@ -169,12 +205,16 @@ class UInt64(Int):
     def default_value(dialect):
         return type_defaults(dialect).uint64()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).uint64()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).uint64(value)
 
 
 class Float(SQLType):
-    impl = types.INTEGER
+    impl = types.FLOAT
 
     @property
     def python_type(self):
@@ -187,6 +227,10 @@ class Float(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).float()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).float()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).float(value)
 
@@ -199,6 +243,10 @@ class Float32(Float):
     def default_value(dialect):
         return type_defaults(dialect).float32()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).float32()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).float32(value)
 
@@ -211,6 +259,10 @@ class Float64(Float):
     def default_value(dialect):
         return type_defaults(dialect).float64()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).float64()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).float64(value)
 
@@ -247,6 +299,10 @@ class Array(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).array()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).array()
+
     def on_read_convert(self, value, dialect):
         r = read_converter(dialect).array(value, self.item_type, dialect)
         if isinstance(self.item_type, JSON):
@@ -268,6 +324,10 @@ class JSON(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).json()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).json()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).json(value)
 
@@ -286,6 +346,10 @@ class DateTime(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).datetime()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).datetime()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).datetime(value)
 
@@ -304,6 +368,10 @@ class Binary(SQLType):
     def default_value(dialect):
         return type_defaults(dialect).binary()
 
+    @staticmethod
+    def db_default_value(dialect):
+        return db_defaults(dialect).binary()
+
     def on_read_convert(self, value, dialect):
         return read_converter(dialect).binary(value)
 
@@ -328,13 +396,17 @@ class TypeReadConverter:
         return value
 
     def float(self, value):
+        if value is None:
+            return float("nan")
+        if isinstance(value, str) and value.lower() == "nan":
+            return float("nan")
         return value
 
     def float32(self, value):
-        return value
+        return self.float(value)
 
     def float64(self, value):
-        return value
+        return self.float(value)
 
     def array(self, value, item_type, dialect):
         if value is None or item_type is None:
@@ -347,10 +419,9 @@ class TypeReadConverter:
     def datetime(self, value):
         return value
 
-    def uuid(self, value):
-        return value
-
     def binary(self, value):
+        if isinstance(value, str):
+            return value.encode()
         return value
 
 
@@ -415,13 +486,13 @@ class TypeDefaults:
         return None
 
     def float(self):
-        return None
+        return float("nan")
 
     def float32(self):
-        return None
+        return self.float()
 
     def float64(self):
-        return None
+        return self.float()
 
     def array(self):
         return None
@@ -432,11 +503,49 @@ class TypeDefaults:
     def datetime(self):
         return None
 
-    def uuid(self):
+    def binary(self):
         return None
 
+
+class DBDefaults:
+    def string(self):
+        return sa.text("''")
+
+    def boolean(self):
+        return sa.text("False")
+
+    def int(self):
+        return sa.text("0")
+
+    def int32(self):
+        return self.int()
+
+    def int64(self):
+        return self.int()
+
+    def uint64(self):
+        return self.int()
+
+    def float(self):
+        return sa.text("NaN")
+
+    def float32(self):
+        return self.float()
+
+    def float64(self):
+        return self.float()
+
+    def array(self):
+        return sa.text("'[]'")
+
+    def json(self):
+        return sa.text("'{}'")
+
+    def datetime(self):
+        return sa.text("'1970-01-01 00:00:00'")
+
     def binary(self):
-        return None
+        return sa.text("''")
 
 
 TYPES = [

datachain/utils.py

@@ -10,7 +10,7 @@ import sys
 import time
 from collections.abc import Iterable, Iterator, Sequence
 from datetime import date, datetime, timezone
-from itertools import islice
+from itertools import chain, islice
 from typing import TYPE_CHECKING, Any, Optional, TypeVar, Union
 from uuid import UUID
 
@@ -241,7 +241,7 @@ _T_co = TypeVar("_T_co", covariant=True)
 
 
 def batched(iterable: Iterable[_T_co], n: int) -> Iterator[tuple[_T_co, ...]]:
-    "Batch data into tuples of length n. The last batch may be shorter."
+    """Batch data into tuples of length n. The last batch may be shorter."""
     # Based on: https://docs.python.org/3/library/itertools.html#itertools-recipes
     # batched('ABCDEFG', 3) --> ABC DEF G
     if n < 1:
@@ -251,6 +251,21 @@ def batched(iterable: Iterable[_T_co], n: int) -> Iterator[tuple[_T_co, ...]]:
         yield batch
 
 
+def batched_it(iterable: Iterable[_T_co], n: int) -> Iterator[Iterator[_T_co]]:
+    """Batch data into iterators of length n. The last batch may be shorter."""
+    # batched('ABCDEFG', 3) --> ABC DEF G
+    if n < 1:
+        raise ValueError("Batch size must be at least one")
+    it = iter(iterable)
+    while True:
+        chunk_it = islice(it, n)
+        try:
+            first_el = next(chunk_it)
+        except StopIteration:
+            return
+        yield chain((first_el,), chunk_it)
+
+
 def flatten(items):
     for item in items:
         if isinstance(item, list):

{datachain-0.3.0.dist-info → datachain-0.3.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: datachain
-Version: 0.3.0
+Version: 0.3.2
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License: Apache-2.0
@@ -55,6 +55,15 @@ Requires-Dist: mkdocs-material >=9.3.1 ; extra == 'docs'
 Requires-Dist: mkdocs-section-index >=0.3.6 ; extra == 'docs'
 Requires-Dist: mkdocstrings-python >=1.6.3 ; extra == 'docs'
 Requires-Dist: mkdocs-literate-nav >=0.6.1 ; extra == 'docs'
+Provides-Extra: examples
+Requires-Dist: datachain[tests] ; extra == 'examples'
+Requires-Dist: numpy <2,>=1 ; extra == 'examples'
+Requires-Dist: defusedxml ; extra == 'examples'
+Requires-Dist: accelerate ; extra == 'examples'
+Requires-Dist: unstructured[pdf] ; extra == 'examples'
+Requires-Dist: pdfplumber ==0.11.3 ; extra == 'examples'
+Requires-Dist: huggingface-hub[hf_transfer] ; extra == 'examples'
+Requires-Dist: nltk ==3.8.1 ; extra == 'examples'
 Provides-Extra: remote
 Requires-Dist: lz4 ; extra == 'remote'
 Requires-Dist: msgpack <2,>=1.0.4 ; extra == 'remote'
@@ -100,102 +109,78 @@ Requires-Dist: usearch ; extra == 'vector'
 AI 🔗 DataChain
 ----------------
 
-DataChain is a data-frame library designed for AI-specific scenarios. It helps ML and
-AI engineers build a metadata layer on top of unstructured files and analyze data using
-this layer.
+DataChain is a modern Pythonic data-frame library designed for artificial intelligence.
+It is made to organize your unstructured data into datasets and wrangle it at scale on
+your local machine.
 
-📂 **Raw Files Processing**
-   Process raw files (images, video, text, PDFs) directly from storage (S3, GCP, Azure,
-   Local), version and update datasets.
+Key Features
+============
 
-🌟 **Metadata layer.**
-   Build a metadata layer on top of files using structured sources like CSV, Parquet,
-   and JSON files.
+📂 **Storage as a Source of Truth.**
+   - Process unstructured data without redundant copies: S3, GCP, Azure, and local
+     file systems.
+   - Multimodal data: images, video, text, PDFs, JSONs, CSVs, parquet.
+   - Join files and metadata together into persistent, versioned, columnar datasets.
 
-⭐ **Metadata enrichment.**
-   Enhance the metadata layer with outputs from local ML model inferences and LLM calls.
+🐍 **Python-friendly data pipelines.**
+   - Operate on Python objects and object fields.
+   - Built-in parallelization and out-of-memory compute without a need in SQL or
+     Spark jobs.
 
-🛠️ **Data Transformation.**
-   Transform metadata using traditional methods like filtering, grouping, joining, and
-   others.
+🧠 **Data Enrichment and Processing.**
+   - Generate metadata columns using local AI models and LLM APIs.
+   - Filter, join, and group by AI metadata. Vector similarity search.
+   - Pass datasets to Pytorch and Tensorflow, or export back into storage.
 
-🐍 **User-friendly interface.**
-   Operate efficiently with familiar Python objects and object fields, eliminating the
-   need for SQL.
+🚀 **Efficiency.**
+   - Parallelization, out-of-memory workloads and data caching.
+   - Vectorized operations on Python object fields: sum, count, avg, etc.
+   - Vector search on embeddings.
 
 
+Quick Start
+-----------
+
 .. code:: console
 
    $ pip install datachain
 
 
-Data Structures
-===============
-
-DataChain introduces expressive data structures tailored for AI-specific workload:
-
-- **Dataset:** Preserves the file-references and meta-information. Takes care of Python
-  object serialization, dataset versioning and difference. Operations on dataset:
-
-  - **Transformations:** traditional data-frame or SQL operations such as filtering,
-    grouping, joining.
-  - **Enrichments:** mapping, aggregating and generating using customer’s Python
-    code. This is needed to work with ML inference and LLM calls.
-
-- **Chain** is a sequence of operations on datasets. Chain executes operations in lazy
-  mode - only when needed.
-
-DataChain name comes from these major data structures: dataset and chaining.
-
+Selecting files using JSON metadata
+======================================
 
-What’s new in DataChain?
-========================
+A storage consists of images of cats and dogs (`dog.1048.jpg`, `cat.1009.jpg`),
+annotated with ground truth and model inferences in the 'json-pairs' format,
+where each image has a matching JSON file like `cat.1009.json`:
 
-The project combines multiple ideas from different areas in order to simplify AI
-use-cases and at the same time to fit it into traditional data infrastructure.
+.. code:: json
 
-- **Python-Native for AI.** Utilizes Python instead of SQL for data manipulation as the
-  native language for AI. It’s powered by `Pydantic`_ data models.
-- **Separation of CPU-GPU workloads.** Distinguishes CPU-heavy transformations (filter,
-  group_by, join) from GPU heavy enrichments (ML-inference or LLM calls). That’s mostly
-  needed for distributed computations.
-- **Resuming data processing** (in development). Introduces idempotent operations,
-  allowing data processing to resume from the last successful process file/record/batch
-  if it fails due to issues like failed LLM calls, ML inference or file download.
+    {
+        "class": "cat", "id": "1009", "num_annotators": 8,
+        "inference": {"class": "dog", "confidence": 0.68}
+    }
 
-Additional relatively new ideas:
+Example of downloading only high-confidence cat images using JSON metadata:
 
-- **Functional style data processing.** Using a functional/chaining approach to data
-  processing rather than declarative SQL, inspired by R-dplyr and some Python libraries.
-- **Data Versioning.** Treats raw files in cloud storage as the source of truth for data
-  and implements data versioning, extending ideas from DVC (developed by the same team).
 
+.. code:: py
 
-What DataChain is NOT?
-======================
-
-- **Not a database** (Postgres, MySQL). Instead, it uses databases under the hood:
-  `SQLite`_ in open-source and ClickHouse and other data warehouses for the commercial
-  version.
-- **Not a data processing tool / data warehouse** (Spark, Snowflake, Big Query) since
-  it delegates heavy data transformations to underlying data warehouses and focuses on
-  AI specific data enrichments and orchestrating all the pieces together.
-
+    from datachain import Column, DataChain
 
-Quick Start
------------
+    meta = DataChain.from_json("gs://datachain-demo/dogs-and-cats/*json", object_name="meta")
+    images = DataChain.from_storage("gs://datachain-demo/dogs-and-cats/*jpg")
 
-Data curation with a local model
-=================================
+    images_id = images.map(id=lambda file: file.path.split('.')[-2])
+    annotated = images_id.merge(meta, on="id", right_on="meta.id")
 
-We will evaluate chatbot dialogs stored as text files in Google Cloud Storage
-- 50 files total in this example.
-These dialogs involve users chatting with a bot while looking for better wireless plans.
-Our goal is to identify the successful dialogs.
+    likely_cats = annotated.filter((Column("meta.inference.confidence") > 0.93) \
+                                   & (Column("meta.inference.class_") == "cat"))
+    likely_cats.export_files("high-confidence-cats/", signal="file")
 
-The data used in the examples is `publicly available`_. The sample code is designed to run on a local machine.
 
-First, we'll show batch inference with a simple sentiment model using the `transformers` library:
+Data curation with a local AI model
+===================================
+Batch inference with a simple sentiment model using the `transformers` library:
 
 .. code:: shell
 
@@ -246,30 +231,30 @@ LLM judging chatbots
 =============================
 
 LLMs can work as efficient universal classifiers. In the example below,
-we employ a free API from Mistral to judge the chatbot performance. Please get a free
+we employ a free API from Mistral to judge the `publicly available`_ chatbot dialogs. Please get a free
 Mistral API key at https://console.mistral.ai
 
+
 .. code:: shell
 
-    $ pip install mistralai
+    $ pip install mistralai (Requires version >=1.0.0)
     $ export MISTRAL_API_KEY=_your_key_
 
 DataChain can parallelize API calls; the free Mistral tier supports up to 4 requests at the same time.
 
 .. code:: py
 
-    from mistralai.client import MistralClient
-    from mistralai.models.chat_completion import ChatMessage
+    from mistralai import Mistral
     from datachain import File, DataChain, Column
 
     PROMPT = "Was this dialog successful? Answer in a single word: Success or Failure."
 
     def eval_dialogue(file: File) -> bool:
-         client = MistralClient()
-         response = client.chat(
+         client = Mistral()
+         response = client.chat.complete(
              model="open-mixtral-8x22b",
-             messages=[ChatMessage(role="system", content=PROMPT),
-                       ChatMessage(role="user", content=file.read())])
+             messages=[{"role": "system", "content": PROMPT},
+                       {"role": "user", "content": file.read()}])
          result = response.choices[0].message.content
          return result.lower().startswith("success")
 
@@ -309,8 +294,8 @@ Instead of extracting this information from the Mistral response data structure
 
 .. code:: py
 
-    from mistralai.client import MistralClient
-    from mistralai.models.chat_completion import ChatMessage, ChatCompletionResponse
+    from mistralai import Mistral
+    from mistralai.models import ChatCompletionResponse
     from datachain import File, DataChain, Column
 
     PROMPT = "Was this dialog successful? Answer in a single word: Success or Failure."
@@ -319,8 +304,8 @@ Instead of extracting this information from the Mistral response data structure
          client = MistralClient()
          return client.chat(
              model="open-mixtral-8x22b",
-             messages=[ChatMessage(role="system", content=PROMPT),
-                       ChatMessage(role="user", content=file.read())])
+             messages=[{"role": "system", "content": PROMPT},
+                       {"role": "user", "content": file.read()}])
 
     chain = (
        DataChain.from_storage("gs://datachain-demo/chatbot-KiT/", object_name="file")
@@ -438,7 +423,10 @@ Tutorials
 ---------
 
 * `Getting Started`_
-* `Multimodal <examples/multimodal/clip_fine_tuning.ipynb>`_ (try in `Colab <https://colab.research.google.com/github/iterative/datachain/blob/main/examples/multimodal/clip_fine_tuning.ipynb>`__)
+* `Multimodal <https://github.com/iterative/datachain-examples/blob/main/multimodal/clip_fine_tuning.ipynb>`_ (try in `Colab <https://colab.research.google.com/github/iterative/datachain-examples/blob/main/multimodal/clip_fine_tuning.ipynb>`__)
+* `LLM evaluations <https://github.com/iterative/datachain-examples/blob/main/llm/llm_chatbot_evaluation.ipynb>`_ (try in `Colab <https://colab.research.google.com/github/iterative/datachain-examples/blob/main/llm/llm_chatbot_evaluation.ipynb>`__)
+* `Reading JSON metadata <https://github.com/iterative/datachain-examples/blob/main/formats/json-metadata-tutorial.ipynb>`_ (try in `Colab <https://colab.research.google.com/github/iterative/datachain-examples/blob/main/formats/json-metadata-tutorial.ipynb>`__)
+
 
 Contributions
 -------------