PyMetis

2025.2.2 verified Mon Apr 27 auth: no python

PyMetis is a Python wrapper for METIS, a set of serial graph partitioning algorithms. It provides both low-level and high-level interfaces for partitioning graphs. Version 2025.2.2 supports Python >=3.10. The package is actively maintained with periodic releases.

pip install pymetis

Common errors

error ImportError: No module named pymetis ↓

cause pymetis is not installed or Python version is <3.10.

fix

Install pymetis with pip install pymetis or upgrade Python.

error ValueError: adjacency must be a list of lists of integers ↓

cause Adjacency list contains non-integer elements (e.g., floats) or is not a list of lists.

fix

Ensure adjacency is a list of lists of ints, e.g., [[1,2],[0,2]].

error pymetis.part_graph returns only one argument (unexpected) ↓

cause Using older version where part_graph returned a single value; modern pymetis returns a tuple (cut, parts).

fix

Update code to unpack two values: cut, parts = pymetis.part_graph(...)

Warnings

gotcha METIS requires that graph adjacency lists contain no duplicate edges and no self-loops. pymetis does not validate inputs; passing invalid graphs can cause silent errors or crashes. ↓

fix Preprocess your graph to remove duplicates and self-loops before calling pymetis.

breaking In v2025.2, the zero-copy mode was introduced. If you rely on adjacency being modified by the C code (e.g., reordering), note that zero-copy may change behavior. The option `zero_copy=True` can be passed to `part_graph` to disable copying. ↓

fix Use `zero_copy=False` if you need the old behavior of copying the adjacency list.

gotcha METIS functions can segfault if the graph is too large or if memory is insufficient. pymetis does not provide graceful error handling for such cases. ↓

fix Ensure sufficient memory and consider using iterative partitioning for large graphs.

Imports

pymetis
```
import pymetis
```
Correct import.
pymetis.part_graph
wrong
```
from pymetis import part_graph
```
correct
```
import pymetis
part = pymetis.part_graph(nparts, adjacency=adjacency)
```
part_graph is a function in the pymetis module; direct import may work but less conventional.

Quickstart

Basic graph partitioning using pymetis.

import numpy as np
import pymetis

# Define adjacency list for a small graph
# Each entry is a list of neighbor indices
adjacency = [[1, 2], [0, 2], [0, 1, 3], [2]]
nparts = 2

cut, parts = pymetis.part_graph(nparts, adjacency=adjacency)
print(f"Edge cut: {cut}")
print(f"Partition assignments: {parts}")