spaCy: Industrial-strength Natural Language Processing

Warnings

• breaking Python version support changes frequently. For example, spaCy v3.8.8 dropped Python 3.9 support, v3.7 dropped 3.6, while later 3.8.x versions added 3.14 support. Always check `requires_python` for your specific spaCy version.
  Fix: Ensure your Python environment meets the `requires_python` specification for your installed spaCy version. Upgrade or downgrade Python if necessary.
• breaking Migration from Pydantic v1 to v2 caused issues with model loading in older v3.8.x releases (e.g., v3.8.12). This was due to changes in how `confection` and `Thinc` handled dependency validation. While patched in v3.8.13, similar deep dependency incompatibilities can recur.
  Fix: Always install the latest patch release (e.g., 3.8.13 for the Pydantic v2 issue). Pin major dependency versions (e.g., Pydantic) to avoid unexpected upgrades, or upgrade spaCy and all its dependencies in a clean virtual environment.
• breaking Upgrading from spaCy v2.x to v3.x involves significant API changes, including the configuration system, pipeline architecture, and how models are trained. Existing custom components and trained models will likely require migration.
  Fix: Consult the official spaCy v2.x to v3.x migration guide. Retrain custom models with the new spaCy version and update code to use the new API and configuration patterns.
• gotcha Trained models are separate Python packages that must be downloaded after the core spaCy library is installed. Forgetting to download a model (e.g., `en_core_web_sm`) will result in `OSError: [E050] Can't find model` errors when calling `spacy.load()`.
  Fix: After `pip install spacy`, always run `python -m spacy download [model_name]` for the models you intend to use (e.g., `en_core_web_sm`).
• gotcha Installed spaCy models must be compatible with your spaCy library version. Incompatible models can lead to unexpected errors or incorrect behavior.
  Fix: After upgrading spaCy, run `python -m spacy validate` to check compatibility of installed models and get recommendations for updates. Retrain any custom models.
• deprecated The `spacy project` functionality was moved into a new standalone library called `Weasel` in spaCy v3.7. While `spacy project` commands still work, some spaCy-specific configuration keys (`spacy_version`, `check_requirements`) are deprecated.
  Fix: For new projects, consider using `Weasel` directly if applicable. Update project configuration files to use `WEASEL_CONFIG_OVERRIDES` instead of `SPACY_CONFIG_OVERRIDES` for environment variables.

Install

• pip install spacy Install core library
• python -m spacy download en_core_web_sm Download a small English model

Imports

• spacy

  import spacy

  Standard import for the spaCy library.
• Language

  from spacy.language import Language

  Used for type hinting or direct instantiation of a blank NLP pipeline.
• Doc

  from spacy.tokens import Doc

  Type for processed text objects.

Quickstart

This quickstart demonstrates how to load a pre-trained English language model and use it to process text. It shows tokenization, lemmatization, part-of-speech tagging, dependency parsing, and named entity recognition. Remember that models must be downloaded separately after installing the spaCy library.

import spacy

# Load a pre-trained English pipeline
# Make sure to run `python -m spacy download en_core_web_sm` first
nlp = spacy.load("en_core_web_sm")

# Process a text
doc = nlp("Apple is looking at buying U.K. startup for $1 billion.")

# Iterate over tokens
for token in doc:
    print(f"{token.text:<15} {token.lemma_:<10} {token.pos_:<10} {token.dep_:<10} {token.ent_type_:<10}")

# Access named entities
print("\nNamed Entities:")
for ent in doc.ents:
    print(f"{ent.text} ({ent.label_})")