/spacy-streamlit

👑 spaCy building blocks and visualizers for Streamlit apps

Primary LanguagePythonMIT LicenseMIT

spacy-streamlit: spaCy building blocks for Streamlit apps

This package contains utilities for visualizing spaCy models and building interactive spaCy-powered apps with Streamlit. It includes various building blocks you can use in your own Streamlit app, like visualizers for syntactic dependencies, named entities, text classification, semantic similarity via word vectors, token attributes, and more.

Current Release Version pypi Version

🚀 Quickstart

You can install spacy-streamlit from pip:

pip install spacy-streamlit

The package includes building blocks that call into Streamlit and set up all the required elements for you. You can either use the individual components directly and combine them with other elements in your app, or call the visualize function to embed the whole visualizer.

Download the English model from spaCy to get started.

python -m spacy download en_core_web_sm

Then put the following example code in a file.

# streamlit_app.py
import spacy_streamlit

models = ["en_core_web_sm", "en_core_web_md"]
default_text = "Sundar Pichai is the CEO of Google."
spacy_streamlit.visualize(models, default_text)

You can then run your app with streamlit run streamlit_app.py. The app should pop up in your web browser. 😀

📦 Example: 01_out-of-the-box.py

Use the embedded visualizer with custom settings out-of-the-box.

streamlit run https://raw.githubusercontent.com/explosion/spacy-streamlit/master/examples/01_out-of-the-box.py

👑 Example: 02_custom.py

Use individual components in your existing app.

streamlit run https://raw.githubusercontent.com/explosion/spacy-streamlit/master/examples/02_custom.py

🎛 API

Visualizer components

These functions can be used in your Streamlit app. They call into streamlit under the hood and set up the required elements.

function visualize

Embed the full visualizer with selected components.

import spacy_streamlit

models = ["en_core_web_sm", "/path/to/model"]
default_text = "Sundar Pichai is the CEO of Google."
visualizers = ["ner", "textcat"]
spacy_streamlit.visualize(models, default_text, visualizers)
Argument Type Description
models List[str] / Dict[str, str] Names of loadable spaCy models (paths or package names). The models become selectable via a dropdown. Can either be a list of names or the names mapped to descriptions to display in the dropdown.
default_text str Default text to analyze on load. Defaults to "".
default_model Optional[str] Optional name of default model. If not set, the first model in the list of models is used.
visualizers List[str] Names of visualizers to show. Defaults to ["parser", "ner", "textcat", "similarity", "tokens"].
ner_labels Optional[List[str]] NER labels to include. If not set, all labels present in the "ner" pipeline component will be used.
ner_attrs List[str] Span attributes shown in table of named entities. See visualizer.py for defaults.
token_attrs List[str] Token attributes to show in token visualizer. See visualizer.py for defaults.
similarity_texts Tuple[str, str] The default texts to compare in the similarity visualizer. Defaults to ("apple", "orange").
show_json_doc bool Show button to toggle JSON representation of the Doc. Defaults to True.
show_meta bool Show button to toggle meta.json of the current pipeline. Defaults to True.
show_config bool Show button to toggle config.cfg of the current pipeline. Defaults to True.
show_visualizer_select bool Show sidebar dropdown to select visualizers to display (based on enabled visualizers). Defaults to False.
sidebar_title Optional[str] Title shown in the sidebar. Defaults to None.
sidebar_description Optional[str] Description shown in the sidebar. Accepts Markdown-formatted text.
show_logo bool Show the spaCy logo in the sidebar. Defaults to True.
color Optional[str] Experimental: Primary color to use for some of the main UI elements (None to disable hack). Defaults to "#09A3D5".
get_default_text Callable[[Language], str] Optional callable that takes the currently loaded nlp object and returns the default text. Can be used to provide language-specific default texts. If the function returns None, the value of default_text is used, if available. Defaults to None.

function visualize_parser

Visualize the dependency parse and part-of-speech tags using spaCy's displacy visualizer.

import spacy
from spacy_streamlit import visualize_parser

nlp = spacy.load("en_core_web_sm")
doc = nlp("This is a text")
visualize_parser(doc)
Argument Type Description
doc Doc The spaCy Doc object to visualize.
keyword-only
title Optional[str] Title of the visualizer block.
sidebar_title Optional[str] Title of the config settings in the sidebar.

function visualize_ner

Visualize the named entities in a Doc using spaCy's displacy visualizer.

import spacy
from spacy_streamlit import visualize_ner

nlp = spacy.load("en_core_web_sm")
doc = nlp("Sundar Pichai is the CEO of Google.")
visualize_ner(doc, labels=nlp.get_pipe("ner").labels)
Argument Type Description
doc Doc The spaCy Doc object to visualize.
keyword-only
labels Sequence[str] The labels to show in the labels dropdown.
attrs List[str] The span attributes to show in entity table.
show_table bool Whether to show a table of entities and their attributes. Defaults to True.
title Optional[str] Title of the visualizer block.
sidebar_title Optional[str] Title of the config settings in the sidebar.
colors Dict[str,str] A dictionary mapping labels to display colors ({"LABEL": "COLOR"})

function visualize_textcat

Visualize text categories predicted by a trained text classifier.

import spacy
from spacy_streamlit import visualize_textcat

nlp = spacy.load("./my_textcat_model")
doc = nlp("This is a text about a topic")
visualize_textcat(doc)
Argument Type Description
doc Doc The spaCy Doc object to visualize.
keyword-only
title Optional[str] Title of the visualizer block.

visualize_similarity

Visualize semantic similarity using the model's word vectors. Will show a warning if no vectors are present in the model.

import spacy
from spacy_streamlit import visualize_similarity

nlp = spacy.load("en_core_web_lg")
visualize_similarity(nlp, ("pizza", "fries"))
Argument Type Description
nlp Language The loaded nlp object with vectors.
default_texts Tuple[str, str] The default texts to compare on load. Defaults to ("apple", "orange").
keyword-only
threshold float Threshold for what's considered "similar". If the similarity score is greater than the threshold, the result is shown as similar. Defaults to 0.5.
title Optional[str] Title of the visualizer block.

function visualize_tokens

Visualize the tokens in a Doc and their attributes.

import spacy
from spacy_streamlit import visualize_tokens

nlp = spacy.load("en_core_web_sm")
doc = nlp("This is a text")
visualize_tokens(doc, attrs=["text", "pos_", "dep_", "ent_type_"])
Argument Type Description
doc Doc The spaCy Doc object to visualize.
keyword-only
attrs List[str] The names of token attributes to use. See visualizer.py for defaults.
title Optional[str] Title of the visualizer block.

Cached helpers

These helpers attempt to cache loaded models and created Doc objects.

function process_text

Process a text with a model of a given name and create a Doc object. Calls into the load_model helper to load the model.

import streamlit as st
from spacy_streamlit import process_text

spacy_model = st.sidebar.selectbox("Model name", ["en_core_web_sm", "en_core_web_md"])
text = st.text_area("Text to analyze", "This is a text")
doc = process_text(spacy_model, text)
Argument Type Description
model_name str Loadable spaCy model name. Can be path or package name.
text str The text to process.
RETURNS Doc The processed document.

function load_model

Load a spaCy model from a path or installed package and return a loaded nlp object.

import streamlit as st
from spacy_streamlit import load_model

spacy_model = st.sidebar.selectbox("Model name", ["en_core_web_sm", "en_core_web_md"])
nlp = load_model(spacy_model)
Argument Type Description
name str Loadable spaCy model name. Can be path or package name.
RETURNS Language The loaded nlp object.