immlib 1.0.0rc2

Author: Noah C. Benson <nben@uw.edu>
License: MIT
Documentation

immlib is a lightweight Python library that simplifies the design of application programming interfaces (APIs) for scientific libraries. The name immlib comes from the library’s philosophy of using immutable data to simplify scientific workflows.

What's in it

• immlib.workflow — immutable workflows built from calc units that are assembled into plan DAGs. Inputs and outputs are tracked and wired up automatically, results are computed lazily, and calculations can be cached in memory or on disk.
• immlib.pathlib / immlib.iolib — a single path function that resolves local and remote paths (S3, Google Storage, Azure, and OSF), downloads remote data into a local cache, and save/load functions that read and write a range of file formats.
• quantities and units — immlib.Quantity (a pint.Quantity that also supports having no units at all) and immlib.math, a small common namespace that operates on quantities, NumPy arrays, and PyTorch tensors alike, so the two backends are interchangeable and units are tracked through the arithmetic.

Alongside these are a set of utilities for testing and coercing types, handling persistent and lazy collections, and working with numerical arguments.

Requirements

immlib supports Python 3.10 through 3.14, including the free-threaded (no-GIL) builds, and depends on NumPy, SciPy, Pint, pcollections, docshare, joblib, cloudpathlib, and PyYAML.

PyTorch and pandas are optional. immlib never imports PyTorch at import time, so it can be used without it (only operations that actually need a tensor raise); pandas is needed only by the csv/tsv save/load formats. See the "Stability and Compatibility" page of the documentation for the supported version ranges of each dependency and what the library promises across releases.

immlib is heavily based on the library pimms, which effectively served as a prototype for immlib. Both libraries were motivated by a number of observations about the design of scientific software and are an attempt to make some of these problems easier to manage.

For more information, see the documentation.