python-crfsuite

0.9.12 · active · verified Thu Apr 09

python-crfsuite is a Python binding for CRFsuite, a fast implementation of Conditional Random Fields (CRFs) for labeling sequential data. It's widely used in Natural Language Processing (NLP) for tasks like Named Entity Recognition (NER), Part-of-Speech (POS) tagging, and other sequence labeling problems. The current version is 0.9.12, and releases primarily focus on Python version compatibility and stability.

Warnings

breaking Version 0.9.12 dropped support for Python 3.6, 3.7, 3.8, and 3.9. Users on these older Python versions must either upgrade their Python environment or pin to an older `python-crfsuite` version.
Fix: Upgrade your Python environment to 3.10 or newer (3.10, 3.11, 3.12, 3.13, 3.14 are supported). If unable to upgrade, pin to `python-crfsuite<0.9.12`.
gotcha The PyPI package name is `python-crfsuite`, but the module to import in your Python code is `pycrfsuite`.
Fix: Always use `import pycrfsuite` in your Python code, even though you install it with `pip install python-crfsuite`.
gotcha The input data format for `Trainer.append()` and `Tagger.tag()` requires a list of feature lists for each item in the sequence. Each feature list is typically a list of strings (e.g., `[['feature1', 'feature2'], ['feature3']]`). Incorrectly formatted input will lead to errors.
Fix: Ensure your feature sequences are formatted as `list[list[str]]` where the outer list represents the sequence, and each inner list represents the features for a single token/item in that sequence. Refer to the quickstart for an example.

Install

pip install python-crfsuite Install stable version

Imports

Trainer
```
import pycrfsuite
trainer = pycrfsuite.Trainer(...)
```
The PyPI package is 'python-crfsuite', but the Python import module is 'pycrfsuite'.
Tagger
```
import pycrfsuite
tagger = pycrfsuite.Tagger(...)
```
The PyPI package is 'python-crfsuite', but the Python import module is 'pycrfsuite'.

Quickstart

This quickstart demonstrates how to train a Conditional Random Field (CRF) model using `pycrfsuite.Trainer` and then use the trained model with `pycrfsuite.Tagger` to predict labels for new sequences. The example uses a simple list-of-lists format for features and labels, which is common for sequence labeling tasks.

import pycrfsuite
import os

# Sample data (features, labels)
X_train = [
    [['walk', 'big'], ['dog']],
    [['eat', 'apple'], ['red', 'apple']],
    [['run', 'fast'], ['cat']]
]
y_train = [
    ['VERB', 'NOUN'],
    ['VERB', 'NOUN'],
    ['VERB', 'NOUN']
]

# 1. Train a CRF model
trainer = pycrfsuite.Trainer(verbose=False)
for xseq, yseq in zip(X_train, y_train):
    trainer.append(xseq, yseq)

trainer.set_params({
    'c1': 1.0,   # coefficient for L1 penalty
    'c2': 1e-3,  # coefficient for L2 penalty
    'max_iterations': 50, # stop earlier
    'feature.possible_transitions': True
})

model_filename = 'model.crfsuite'
trainer.train(model_filename)

print(f"Model trained and saved to '{model_filename}'")

# 2. Use the trained model for tagging
tagger = pycrfsuite.Tagger()
tagger.open(model_filename)

X_test = [
    [['see', 'small'], ['dog']]
]

predicted_tags = [tagger.tag(xseq) for xseq in X_test]
print(f"Test sequence: {X_test}")
print(f"Predicted tags: {predicted_tags}")

# Clean up the model file
os.remove(model_filename)

view raw JSON →