A toolbox of common types, protocols, and tooling to support AI test and evaluation workflows.
Check out the documentation and examples for more information.
MAITE is a library of common types, protocols (a.k.a. structural subtypes), and utilities for the test and evaluation (T&E) of supervised machine learning models. It is being developed under the Joint AI T&E Infrastructure Capability (JATIC) program. Its goal is to streamline the development of JATIC Python projects by ensuring seamless, synergistic workflows when working with MAITE-conforming Python packages for different T&E tasks. To this end, MAITE seeks to eliminate redundancies that would otherwise be shared across – and burden – separate efforts in machine learning test and evaluation. MAITE is designed to be a low-dependency, frequently-improved Python package that is installed by JATIC projects. The following is a brief overview of the current state of its submodules.
To install from the Python Package Index (PyPI), run:
pip install maiteℹ️ You can install MAITE for a given release tag, e.g.
v0.6.1, by running:$ pip install git+ssh://git@github.com/mit-ll-ai-technology/maite.git@v0.6.1
To clone this repository and install from source, run:
$ git clone https://github.com/mit-ll-ai-technology/maite
$ cd maite
$ pip install .Common types for machine learning test and evaluation
The protocols subpackage defines common types – such as an inference-mode object detector – to be leveraged across JATIC projects. These are specifically designed to be Python protocol classes, which support structural subtyping. As a result, developers and users can satisfy MAITE-typed interfaces without having to explicitly subclass. This ability helps to promote common interfaces across JATIC projects without introducing explicit inter-dependencies between them.
Support for rigorous software testing
The testing subpackage is designed to help developers create a rigorous automated test suite for their project. These include:
- Pytest fixtures for initializing test functions with common models, datasets, and other inputs that are useful for testing machine learning code.
- Functions running static type checking tests using pyright in a pytest test suite, including scans of both source code and example documentation code blocks.
- Hypothesis strategies for driving property-based tests of interfaces that leverage MAITE protocols.
>>> def f(x: str):
... return 1 + x
>>> pyright_analyze(f)
{'version': '1.1.281',
'time': '1669686515154',
'generalDiagnostics': [{'file': 'source.py',
'severity': 'error',
'message': 'Operator "+" not supported for types "Literal[1]" and "str"\n\xa0\xa0Operator "+" not supported for types "Literal[1]" and "str"',
'range': {'start': {'line': 1, 'character': 11},
'end': {'line': 1, 'character': 16}},
'rule': 'reportGeneralTypeIssues'}],
'summary': {'filesAnalyzed': 20,
'errorCount': 1,
'warningCount': 0,
'informationCount': 0,
'timeInSec': 0.319}}General utilities
- Functions for validating the types and values of user arguments, with explicit and consistent user-error messages, that raise MAITE-customized exceptions.
- Specialized PyTorch utilities to help facilitate safe and ergonomic code patterns for manipulating stateful torch objects
- Other quality assurance and convenience functions that may be widely useful across projects
DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.
© 2024 MASSACHUSETTS INSTITUTE OF TECHNOLOGY
- Subject to FAR 52.227-11 – Patent Rights – Ownership by the Contractor (May 2014)
- SPDX-License-Identifier: MIT
This material is based upon work supported by the Under Secretary of Defense for Research and Engineering under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Under Secretary of Defense for Research and Engineering.
The software/firmware is provided to you on an As-Is basis