Repository files navigation

Linux	Mac

Spacy-cpp is a C++ wrapper library for the NLP libraryspaCy. This project is not affiliated with spaCy, it ishowever distributed under the same type of license (MIT).

The goal of spacy-cpp is to expose the functionality of spaCy to C++applications, and to provide an API that is similar to that of spaCy,enabling rapid development in Python and simple porting to C++.

Spacy-cpp is under development and does not yet support all API's of spaCy,refer to the API Documentation section below.

Simple POS tagging example using spacy-cpp:

Spacy::Spacy spacy;auto nlp = spacy.load("en_core_web_sm");auto doc = nlp.parse("This is a sentence.");for (auto& token : doc.tokens())    std::cout << token.text() <<" [" << token.pos_() <<"]\n";

For reference - doing the same using the spaCy API in Python:

importspacynlp=spacy.load("en_core_web_sm")doc=nlp(u"This is a sentence.")fortokenindoc:print(token.text+" ["+token.pos_+"]")

Spacy-cpp is implemented using C++11 with the intention of being portable.Current version has been tested on:

• macOS Ventura 13.1
• Ubuntu 22.04 LTS

Spacy-cpp requires python development library, pip, spaCy and typically aspaCy model.

Install build dependencies:

brew install cmake python

Install spaCy and an English model:

pip3 install -U spacypython3 -m spacy download en_core_web_sm

Install build dependencies:

sudo apt install cmake python3-pip libpython3-dev

Install spaCy and an English model:

pip3 install -U spacypython3 -m spacy download en_core_web_sm

Spacy-cpp can be used either as a shared library or as a header-only library.

Build and install spacy-cpp:

mkdir -p build && cd build && cmake .. && make && sudo make install

Link library:

-lspacy

Include header (convenience header including all modules):

#include <spacy/spacy>

Copy the src/spacy directory to the source directory of your project. Thendefine SPACY_HEADER_ONLY and include headers needed (spacy/spacy includesall headers):

#define SPACY_HEADER_ONLY#include <spacy/spacy>

The source tree includes two CMake project examples:

• Using spacy-cpp as header-only library
• Using spacy-cpp as shared library

If a system has more than one Python installation, each of the installationswill have its own set of pip-installed Python packages. One must ensure thatspacy is installed for the Python version used by spacy-cpp (alternativelypoint spacy-cpp to the desired Python installation). When building spacy-cppusing CMake (example:./make.sh tests) the Python version used will beoutput, for examplePYTHON_EXECUTABLE="/usr/local/bin/python3.11". Use thisinterpreter to ensure spacy works correctly in Python, example:/usr/local/bin/python3.11 ./examples/python-spacy-usage.py. If not working,use this Python version to install spacy and a language model:

/usr/local/bin/python3.11 -m pip install -U spacy/usr/local/bin/python3.11 -m spacy download en_core_web_sm

Spacy-cpp is under development and does not support the complete spaCy API yet.

• Attrs
• Doc
• MorphAnalysis
• Nlp
• Spacy
• Span
• StringStore
• Token
• Vocab

Attrs supports all attribute constants.

Doc supports the following methods / attributes:

• count_by()
• ents()
• has_vector()
• is_parsed()
• is_tagged()
• noun_chunks()
• sentiment()
• sents()
• similarity()
• text()
• text_with_ws()
• tokens()
• vector_norm()

MorphAnalysis supports the following methods / attributes:

• get()
• str()
• to_dict()

Nlp supports the following methods / attributes:

• parse()
• vocab()

Spacy supports the following methods / attributes:

• load()
• attrs()

Span supports the following methods / attributes:

• doc()
• label()
• label_()
• lemma_()
• orth_()
• root()
• sentiment()
• text()
• text_with_ws()
• tokens()
• vector_norm()

StringStore supports the following methods / attributes:

• add()

Token supports the following methods / attributes:

• check_flag()
• children()
• cluster()
• dep()
• dep_()
• ent_iob_()
• has_vector()
• head()
• i()
• idx()
• is_alpha()
• is_ascii()
• is_bracket()
• is_digit()
• is_left_punct()
• is_lower()
• is_oov()
• is_punct()
• is_quote()
• is_right_punct()
• is_space()
• is_stop()
• is_title()
• is_upper()
• lang()
• lang_()
• lemma()
• lemma_()
• like_email()
• like_num()
• like_url()
• lower()
• lower_()
• morph()
• nbor()
• norm()
• norm_()
• orth()
• orth_()
• pos()
• pos_()
• prob()
• rank()
• sentiment()
• shape()
• shape_()
• tag()
• tag_()
• text()
• text_with_ws()
• whitespace_()

Vocab supports the following methods / attributes:

• strings()

1. In spacy-cpp Nlp cannot be called as a method in order to perform parsing.Instead one need to use Nlp::parse().
2. In spacy-cpp Doc is not an iterable, instead one need to use Doc::token()to get a std::vector of the tokens in the Doc. Likewise for Span.
3. In spacy-cpp non-ASCII strings must be UTF-8 encoded in order to becorrectly processed.

Spacy-cpp uses cmake for its tests. Commands to build and execute the testsuite:

mkdir -p build && cd build && cmake -DSPACYCPP_BUILD_TESTS=ON .. && make && ctest --output-on-failure ; cd -

Spacy-cpp is distributed under the MIT license.SeeLICENSE file.

Bugs, PRs, etc are welcome on the GitHub project pagehttps://github.com/d99kris/spacy-cpp

c++, c++11, natural language processing, nlp, spacy.