iden 0.4.1

simple library to manage a dataset of shards to train machine learning models

iden

Overview

iden is a simple Python library to manage a dataset of shards when training a machine learning model. iden uses a lazy loading approach to load the shard's data, so it is easy to manage shards without loading their data. iden supports different formats to store shards on disk.

Key Features

• Lazy Loading: Shards are loaded only when needed, enabling efficient memory management
• Multiple Formats: Support for JSON, YAML, Pickle, PyTorch, safetensors, and more
• Flexible Dataset Management: Organize data into splits (train/val/test) with associated assets
• URI-based Identification: Each shard has a unique URI for easy persistence and loading
• Caching Support: Optional in-memory caching for frequently accessed shards
• Extensible: Easy to add custom shard types and loaders

Quick Example

import tempfile
from pathlib import Path
from iden.dataset import create_vanilla_dataset
from iden.shard import create_json_shard, create_shard_dict, create_shard_tuple

# Create a simple dataset
with tempfile.TemporaryDirectory() as tmpdir:
    # Create shards
    train_tuple = create_shard_tuple(
        [
            create_json_shard(
                [1, 2, 3], uri=Path(tmpdir).joinpath("train1.json").as_uri()
            ),
            create_json_shard(
                [4, 5, 6], uri=Path(tmpdir).joinpath("train2.json").as_uri()
            ),
        ],
        uri=Path(tmpdir).joinpath("train_tuple").as_uri(),
    )
    val_tuple = create_shard_tuple(
        [create_json_shard([7, 8, 9], uri=Path(tmpdir).joinpath("val1.json").as_uri())],
        uri=Path(tmpdir).joinpath("val_tuple").as_uri(),
    )

    # Organize shards into splits
    shards = create_shard_dict(
        shards={"train": train_tuple, "val": val_tuple},
        uri=Path(tmpdir).joinpath("shards").as_uri(),
    )
    assets = create_shard_dict(shards={}, uri=Path(tmpdir).joinpath("assets").as_uri())

    # Create dataset
    dataset = create_vanilla_dataset(
        shards=shards,
        assets=assets,
        uri=Path(tmpdir).joinpath("my_dataset").as_uri(),
    )

    # Access data
    train_shards = dataset.get_shards("train")
    print(train_shards[0].get_data())  # Output: [1, 2, 3]

Installation

We highly recommend installing a virtual environment. iden can be installed from pip using the following command:

uv pip install iden

To make the package as slim as possible, only the minimal packages required to use iden are installed. To include all the dependencies, the following command can be used:

uv pip install iden[all]

Please check the get started page to see how to install only some specific dependencies or other alternatives to install the library.

Documentation

• Get Started: Installation instructions
• User Guide: Learn about shards and datasets
• How-to Guides: Step-by-step guides for common tasks
• API Reference: Complete API documentation
• Examples: Practical code examples

Basic Usage

Working with Shards

from iden.shard import create_json_shard

# Create a shard
shard = create_json_shard(data={"key": "value"}, uri="file:///path/to/data.json")

# Get data from shard
data = shard.get_data()

# Cache data for faster access
data = shard.get_data(cache=True)

Managing Datasets

from iden.dataset import create_vanilla_dataset
from iden.shard import create_json_shard, create_shard_dict, create_shard_tuple

# Create a dataset with train/val splits
train_tuple = create_shard_tuple([shard1, shard2, shard3], uri="file:///train_tuple")
val_tuple = create_shard_tuple([shard4, shard5], uri="file:///val_tuple")

shards = create_shard_dict(
    shards={"train": train_tuple, "val": val_tuple},
    uri="file:///shards",
)
assets = create_shard_dict(shards={}, uri="file:///assets")

dataset = create_vanilla_dataset(
    shards=shards,
    assets=assets,
    uri="file:///path/to/dataset",
)

# Access shards
train_shards = dataset.get_shards("train")
first_shard_data = train_shards[0].get_data()

The following is the corresponding iden versions and tested dependencies.

iden	coola	objectory	numpy*	pyyaml*	safetensors*	torch*	python
main	>=1.1,<2.0	>=0.3,<1.0	>=1.24,<2.0	>=6.0,<7.0	>=0.6,<1.0	>=2.0,<3.0	>=3.10
0.4.1	>=1.1,<2.0	>=0.3,<1.0	>=1.24,<2.0	>=6.0,<7.0	>=0.6,<1.0	>=2.0,<3.0	>=3.10
0.4.0	>=1.0,<2.0	>=0.3,<1.0	>=1.24,<2.0	>=6.0,<7.0	>=0.6,<1.0	>=2.0,<3.0	>=3.10
0.3.0	>=0.11.0,<1.0	>=0.3,<1.0	>=1.24,<2.0	>=6.0,<7.0	>=0.6,<1.0	>=2.0,<3.0	>=3.10
0.2.0	>=0.8.4,<1.0	>=0.2,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<3.0	>=3.9,<3.14
0.1.0	>=0.8.4,<1.0	>=0.2,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<3.0	>=3.9,<3.14

iden	cloudpickle*	joblib*
main	>=3.0,<4.0	>=1.3,<2.0
0.4.0	>=3.0,<4.0	>=1.3,<2.0
0.3.0	>=3.0,<4.0	>=1.3,<2.0

^* indicates an optional dependency

older versions

iden	coola	objectory	numpy*	pyyaml*	safetensors*	torch*	python
0.0.4	>=0.3,<1.0	>=0.1,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<3.0	>=3.9,<3.13
0.0.3	>=0.3,<1.0	>=0.1,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<3.0	>=3.9,<3.12
0.0.2	>=0.4,<1.0	>=0.1,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<2.1	>=3.9,<3.12
0.0.1	>=0.4,<1.0	>=0.1,<1.0	>=1.22,<2.0	>=6.0,<7.0	>=0.4,<1.0	>=2.0,<2.1	>=3.9,<3.12

Contributing

Please check the instructions in CONTRIBUTING.md.

Suggestions and Communication

Everyone is welcome to contribute to the community. For any questions or suggestions, Github Issues can be submitted. All issues will be addressed as soon as possible.

API stability

:warning: While iden is in development stage, no API is guaranteed to be stable from one release to the next. In fact, it is very likely that the API will change multiple times before a stable 1.0.0 release. In practice, this means that upgrading iden to a new version will possibly break any code that was using the old version of iden.

License

iden is licensed under BSD 3-Clause "New" or "Revised" license available in LICENSE file.