Python library centre

How to Train and Fine-Tune NLP Models Using SpaCy: A Step-by-Step Guide

SpaCy is a powerful Python library for Natural Language Processing (NLP). It is used to work with human language data easily. Common real-world uses are…

⬇ Download notebook Open in Colab ↗

CoursePython library centre
Video19 min
FormatJupyter notebook · 21 code cells

spaCyregex (re)

Data

No separate download needed — the notebook creates or downloads everything it uses.

📓 Full notebook

Download .ipynb

Introduction to SpaCy#

SpaCy is a powerful Python library for Natural Language Processing (NLP).
It is used to work with human language data easily.
Common real-world uses are automatic text classification, extracting information, and analyzing documents.
SpaCy makes it easy to process text, split text into words, and understand language structure.
In this lesson, we will learn the basics of SpaCy step-by-step.

import warnings; warnings.filterwarnings("ignore")
 
# To install: open your command prompt and run:
# pip install spacy
 
# You will also need to download a model:
# python -m spacy download en_core_web_sm
 
import spacy
from spacy.language import Language

Core Objects in SpaCy#

The main object is the 'nlp' object. It processes your text.
When you pass text to the nlp object, you get a 'Doc' object.
A Doc object holds your text, words, and linguistic information.
Tokens represent words or punctuation.
The library provides access to Parts of Speech, Lemmas, Entities, and more.

nlp = spacy.load("en_core_web_sm")
 
text = "SpaCy is an awesome library for NLP!"
doc = nlp(text)
 
print(doc)

SpaCy is an awesome library for NLP!

for token in doc:
    print(token.text)

SpaCy
is
an
awesome
library
for
NLP
!

for token in doc:
    print(f"{token.text}: {token.pos_}")

SpaCy: PROPN
is: AUX
an: DET
awesome: ADJ
library: NOUN
for: ADP
NLP: PROPN
!: PUNCT

Example: Tokenization#

Tokenization means splitting text into words and punctuation.
SpaCy handles tokenization automatically.
Each token is an object with helpful properties.

sentence = "Python is great for data science!"
doc2 = nlp(sentence)
 
for token in doc2:
    print(token.text, token.idx)

Python 0
is 7
great 10
for 16
data 20
science 25
! 32

print("Has vector:", doc2.has_vector)
print("Is tagged:", doc2.is_tagged)
print("Is parsed:", doc2.is_parsed)

Has vector: True
Is tagged: True
Is parsed: True

Example: Lemmatization#

Lemmatization reduces words to their base forms.
For example, 'running' becomes 'run'.
SpaCy makes it easy to get the lemma for each word.

lemmas_example = "I am running and I ran every day."
doc3 = nlp(lemmas_example)
 
for token in doc3:
    print(f"{token.text} -> {token.lemma_}")

I -> I
am -> be
running -> run
and -> and
I -> I
ran -> run
every -> every
day -> day
. -> .

Example: Named Entity Recognition (NER)#

NER finds names, places, dates, and organizations in text.
SpaCy can identify named entities easily.

ner_sentence = "Apple was founded by Steve Jobs in California in 1976."
doc4 = nlp(ner_sentence)
for ent in doc4.ents:
    print(ent.text, ent.label_)

Apple ORG
Steve Jobs PERSON
California GPE
1976 DATE

Example: Stop Words#

Stop words are common words like 'the', 'is', 'and'.
Often, these words are ignored in text processing.
SpaCy provides a list of stop words.

for token in doc4:
    if token.is_stop:
        print(token.text)

was
by
in
in

Intermediate: Dependency Parsing#

SpaCy understands how words are linked (syntax).
You can see the relation between words using dependencies.

sent = "The quick brown fox jumps over the lazy dog."
doc5 = nlp(sent)
for token in doc5:
    print(token.text, "<-", token.dep_, "-", token.head.text)

The <- det - fox
quick <- amod - fox
brown <- amod - fox
fox <- nsubj - jumps
jumps <- ROOT - jumps
over <- prep - jumps
the <- det - dog
lazy <- amod - dog
dog <- pobj - over
. <- punct - jumps

for chunk in doc5.noun_chunks:
    print(chunk.text)

The quick brown fox
the lazy dog

Intermediate: Sentence Segmentation#

SpaCy splits text into sentences automatically.
These sentences can be processed separately.

multi_sent = "SpaCy is easy. It is fast. It is accurate."
doc6 = nlp(multi_sent)
for sent in doc6.sents:
    print(sent.text)

SpaCy is easy.
It is fast.
It is accurate.

Intermediate: Custom Stop Words#

You can add your own stop words to SpaCy.
This can help handle domain-specific needs.

nlp.Defaults.stop_words.add("awesome")
doc7 = nlp("SpaCy is awesome and useful.")
for token in doc7:
    if token.is_stop:
        print(token.text)

is
and

Advanced: Create Custom Pipeline Component#

You can add your own processing step into SpaCy's NLP pipeline.
This allows you to run custom functions on all texts.

@Language.component("custom_component")
def custom_component(doc):
    print("Custom component ran!")
    return doc
 
nlp.add_pipe("custom_component", last=True)
doc8 = nlp("Testing component.")

Custom component ran!

Advanced: Save and Load SpaCy Models#

You can save your trained model or modified pipeline.
Later, you can load it back for use.

nlp.to_disk("my_spacy_pipeline")  # Save pipeline to folder
loaded_nlp = spacy.load("my_spacy_pipeline")
doc9 = loaded_nlp("Model loaded!")
print(doc9.text)

Custom component ran!
Model loaded!

Error Handling in SpaCy#

SpaCy will raise errors if you use a model that is not downloaded.
Let us handle some typical errors with try / except.

try:
    faulty_nlp = spacy.load("nonexistent_model")
except Exception as e:
    print("Error:", e)

Error: [E050] Can't find model 'nonexistent_model'. It doesn't seem to be a Python package or a valid path to a data directory.

try:
    blank_nlp = spacy.blank("zz")
except Exception as e:
    print("Error initializing blank model:", e)

Error initializing blank model: [E048] Can't import language zz or any matching language from spacy.lang: No module named 'spacy.lang.zz'

Best Practices with SpaCy#

Always use the smallest model possible for faster results.
Use pip requirements.txt to manage packages.
Save and load pipelines to reuse your setup.
Handle exceptions to make programs robust.
Read the SpaCy documentation for updates and new features.

def is_entity(text):
    doc = nlp(text)
    if len(doc.ents) > 0:
        print("Entities detected:")
        for ent in doc.ents:
            print(ent.text, ent.label_)
    else:
        print("No entities found.")

is_entity("Barack Obama was President of the United States.")
is_entity("Nothing special here.")

Custom component ran!
Entities detected:
Barack Obama PERSON
the United States GPE
Custom component ran!
No entities found.

Mini Project: Extract Emails from Text#

Let us build a function to extract email addresses using SpaCy.
This is useful for contact scraping and document analysis.

import re
def extract_emails(text):
    pattern = r"[\w.-]+@[\w.-]+\.[a-zA-Z]{2,}"
    emails = re.findall(pattern, text)
    if emails:
        print("Emails found:", emails)
    else:
        print("No emails found.")

sample = "Contact us at info@example.com or support@mail.com."
extract_emails(sample)

Emails found: ['info@example.com', 'support@mail.com']

YouTube Call to Action#

Please subscribe to our channel for more tutorials!
Like this video if you found it helpful.

Found this useful?

All lessons, notebooks and datasets here are free. If they helped you, a coffee keeps new lessons coming.

☕ Buy me a coffee Other ways to help