lgeiger / mypy-einsum

MyPy Type Checking for NumPy/Jax/PyTorch Einsum Operations

MyPy Type Checking for NumPy/Jax/PyTorch Einsum Operations

mypy_einsum is a Mypy plugin for type checking np.einsum, jax.numpy.einsum, and torch.einsum operations.

The Einstein summation convention can be used to compute many multi-dimensional, linear algebraic array operations. einsum provides a succinct way of representing these.

However, since einsum equations are passed as a string, it is very easy to overlook typos or other bugs as linters are unable to help. mypy_einsum is a Mypy plugin that that is able to statically verify the correctness of einsum equations with needing to execute the code.

Installation

mypy_einsum can be installed with pip:

pip install mypy-einsum

Setup

To enable the plugin, add it to you projects Mypy configuration file. Usually mypy.ini:

[mypy]
plugins = mypy_einsum

or pyproject.toml:

[tool.mypy]
plugins = ["mypy_einsum"]

Example

Can you spot the 🐛 without running the code?

import numpy as np

a = np.arange(9).reshape(3, 3)

np.einsum("ik,kj->ij", a)

mypy_einsum will catch it for you:

❯ mypy example.py --pretty
example.py:5: error: Number of einsum subscripts must be equal to the
number of operands.  [einsum]
    np.einsum("ik,kj->ij", a)
              ^~~~~~~~~~~
Found 1 error in 1 file (checked 1 source file)

After fixing it mypy will succeed 🎉:

np.einsum("ik,kj->ij", a, a)

❯ mypy example.py
Success: no issues found in 1 source file

Supported Operations

• np.einsum
• np.einsum_path
• jax.numpy.einsum
• jax.numpy.einsum_path
• torch.einsum

Reporting Issues and Contributing

mypy_einsum aims to never raise warnings for valid einsum operations. If you encounter a warning that you believe is incorrect, or think mypy_einsum is not reporting an error please let us know. Contributions are very welcome!

mypy-einsum's Issues

Add tests

Add support for einops

https://github.com/arogozhnikov/einops

Add Readme