nagata

Version
Status
Documentation
Tools
Compatibility
Stats

What is nagata?

This package is under heavy construction. Use at your own risk

nagata is a small Python package for loading and saving files with one common, intuitive syntax. Instead of remembering pandas.read_parquet, yaml.safe_load, numpy.load, pickle.load, and so on, you create a FileManager once and call load and save. nagata picks the right reader or writer from the file extension and resolves folders for you.

"Don’t decide for them. Tell them the truth and let them decide for themselves." - Naomi Nagata

Why use nagata?

• One syntax for many formats: load and save work the same way for tabular data, arrays, configuration files, figures, and pickled objects.
• Format detection: the file extension selects the format. You can also name the format explicitly.
• Project folders: set a root, input, interim, and output folder once and refer to them by name ('input', 'output') or just by file name.
• Shared defaults: settings such as text encoding and index handling live in one place (FileFramework.settings) and are applied to every format.
• Extensible: subclass FileFormat and your format is registered automatically.
• Light by default: nagata only imports pandas, numpy, pyyaml, and the other libraries when you use a format that needs them.

You may not want nagata if you need lazy or out-of-core loading, remote storage (S3, GCS, etc.), or fine-grained control over every reader option in the underlying libraries. For those cases, use those libraries (or something like fsspec) directly. nagata does let you pass any extra keyword arguments through to the underlying reader or writer.

Getting started

Requirements

nagata requires Python 3.11 or later and runs on Linux, macOS, and Windows. It is tested on Python 3.11 through 3.14. Its only required dependency is wonka. Support for most file formats relies on optional third-party libraries, described below.

Installation

To install nagata, use pip:

pip install nagata

Formats that rely on other libraries need those libraries installed. You can install them yourself or use one of nagata's extras:

pip install "nagata[pandas]"   # CSV, TSV, JSON, JSON Lines, Stata
pip install "nagata[excel]"    # Excel (pandas and openpyxl)
pip install "nagata[parquet]"  # Parquet and Feather (pandas and pyarrow)
pip install "nagata[all]"      # everything below

Text, pickle, and TOML loading need nothing beyond the standard library.

Usage

Create a FileManager with the folders for your project. Relative folder names are placed inside the root folder, and any folders that do not exist are created.

import nagata

files = nagata.FileManager(
    root_folder = 'project',
    input_folder = 'raw',
    output_folder = 'processed')

Loading and saving

The file extension tells nagata which format to use. By default, files are loaded from the input folder and saved to the output folder.

df = files.load(file_name = 'survey.csv')        # project/raw/survey.csv
files.save(df, file_name = 'survey.parquet')     # project/processed/survey.parquet

Additional keyword arguments are passed to the underlying library:

df = files.load(file_name = 'survey.csv', usecols = ['age', 'income'])
files.save(df, file_name = 'survey.csv', index = True)

Choosing folders

Pass folder to use a different location. It can be the name of one of the manager's folders ('input', 'interim', 'output', 'root'), or a folder name inside the root folder.

df = files.load(file_name = 'survey', file_format = 'parquet', folder = 'output')
files.save(df, file_name = 'survey_backup.csv', folder = 'root')

Or skip the folder logic and pass a complete path (the folder must already exist):

df = files.load(file_path = 'project/raw/survey.csv')
files.save(df, file_path = 'project/processed/survey.tsv')

Different kinds of data

The same two methods work for configuration files, arrays, models, and figures.

files.save({'seed': 42, 'folds': 5}, file_name = 'params.yaml')
params = files.load(file_name = 'params.yaml', folder = 'output')

files.save(features, file_name = 'features.npy')       # NumPy array
files.save(model, file_name = 'model.joblib')          # fitted scikit-learn model
files.save(figure, file_name = 'accuracy.png')         # matplotlib or seaborn figure
files.save(notes, file_name = 'notes.txt')             # plain text

Supported file formats

You can see every registered format with nagata.FileFramework.formats. If a file has no extension, or the extension is not the one you want, pass file_format with the format's name (for example, file_format = 'json_lines').

Adding your own format

Subclassing FileFormat registers the new format automatically. Give it the extension(s) it handles and load and save methods:

import dataclasses
import pathlib
from typing import ClassVar

import nagata


@dataclasses.dataclass
class FileFormatShout(nagata.FileFormat):
    """Stores text in upper case and loads it in lower case."""

    extensions: ClassVar[str] = 'shout'
    load_parameters: ClassVar[dict] = {}
    save_parameters: ClassVar[dict] = {}

    def load(self, path, **kwargs):
        return pathlib.Path(path).read_text().lower()

    def save(self, item, path, **kwargs):
        pathlib.Path(path).write_text(item.upper())


files = nagata.FileManager(root_folder = 'project')
files.save('hello', file_name = 'greeting.shout')
files.load(file_name = 'greeting.shout')  # 'hello'

load_parameters and save_parameters map an argument of your load or save method to a key in nagata.FileFramework.settings, so that shared defaults such as 'file_encoding' are applied automatically. The built-in CSV format is a good example.

Contributing

Contributors are always welcome. Feel free to grab an issue to work on or make a suggested improvement. If you wish to contribute, please read the Contribution Guide and Code of Conduct.

Similar Projects

nagata builds on, and is not a replacement for, the excellent libraries it wraps. If it is not quite what you need, these projects solve related problems:

• pandas I/O: the reader and writer functions that nagata uses for most tabular formats.
• fsspec: a uniform interface to local and remote filesystems.
• Kedro's Data Catalog and Intake: declarative catalogs of datasets for larger data pipelines.
• joblib: efficient persistence for Python objects, including large NumPy arrays and fitted models.

Acknowledgments

Thanks to the maintainers and contributors of pandas, NumPy, seaborn and matplotlib, PyYAML, joblib, and the other open-source libraries that do the heavy lifting behind nagata's formats. The quotation at the top of this README is from James S. A. Corey's The Expanse series.

License

Use of this repository is authorized under the Apache Software License 2.0.