Boost NLP Speed 100× with Cython: A Practical Guide

Why Cython for Faster NLP

Cython compiles Python code to C, providing the speed boost behind libraries like NumPy and pandas. The author introduces the NeuralCoref v3.0 project, which runs roughly 100× faster than its predecessor while keeping Python compatibility.

Profiling Python Bottlenecks

Using cProfile and pstats helps locate slow loops in pure Python code, especially when processing large news articles or training data for deep learning frameworks.

import cProfile
import pstats
import myslowmodule
cProfile.run('myslowmodule.run()', 'restats')
p = pstats.Stats('restats')
p.sortstats('cumulative').printstats(30)

Designing a High‑Speed Cython Module

The article shows how to rewrite a Python function that checks rectangle areas into Cython, defining C structs for rectangles and using typed memoryviews for fast loops.

cdef struct Rectangle:
    float w
    float h
cdef int check_rectangles(Rectangle* rectangles, int n_rectangles, float threshold):
    cdef int n_out = 0
    for rectangle in rectangles[:n_rectangles]:
        if rectangle.w * rectangle.h > threshold:
            n_out += 1
    return n_out

Integrating Cython with spaCy

spaCy’s internal C structures (e.g., TokenC) and the StringStore allow fast access to token data via 64‑bit hash codes. By converting target strings to hashes, Cython loops can operate directly on C arrays without Python overhead.

%%cython -+
import numpy
from cymem.cymem cimport Pool
from spacy.tokens.doc cimport Doc
from spacy.typedefs cimport hash_t
from spacy.structs cimport TokenC
cdef struct DocElement:
    TokenC* c
    int length
cdef int fast_loop(DocElement* docs, int n_docs, hash_t word, hash_t tag):
    cdef int n_out = 0
    for doc in docs[:n_docs]:
        for c in doc.c[:doc.length]:
            if c.lex.lower == word and c.tag == tag:
                n_out += 1
    return n_out

Performance Results

Running the Cython‑accelerated NLP loop in a Jupyter Notebook processes about 17 million characters in ~20 ms, roughly 80× faster than the pure Python version, achieving up to 8 × 10⁷ characters per second.

Getting Started

Install Cython with pip install cython, then use %load_ext Cython or %%cython magic in notebooks. The article provides links to the full Jupyter Notebook and additional Cython tutorials.

For more details, see the original Medium post and the spaCy Cython conventions documentation.