Spaces:
Running
Running
from typing import Optional | |
import spacy | |
from spacy import displacy | |
from spacy.language import Language | |
import streamlit as st | |
from spacy_streamlit import visualize_parser | |
from spacy_streamlit import visualize_tokens | |
import base64 | |
from PIL import Image | |
import deplacy | |
#import graphviz | |
st.set_page_config(layout="wide") | |
st.image("logo.png", use_column_width=False, width=150) | |
st.title("Diogenet's Ancient Greek Syntax Analyzer") | |
st.markdown("Here you'll find four spaCy models for processing ancient Greek. They have been trained with the Universal Dependencies datasets *Perseus* and *Proiel*. We provide two types of models for each dataset. The '_lg' models were built with tok2vec pretrained embeddings and fasttext vectors, while the '_tfr' models have a transfomers layer. You can choose among models to compare their performance. More information about the models can be found in the [Huggingface Models Hub] (https://huggingface.co/Jacobo).") | |
st.header("Select a model:") | |
spacy_model = st.selectbox("", ["grc_ud_perseus_lg", "grc_ud_proiel_lg","grc_ud_perseus_trf"]) | |
st.header("Enter text:") | |
text = st.text_area("", "Πλάτων ὁ Περικτιόνης τὸ γένος ἀνέφερεν εἰς Σόλωνα.") | |
nlp = spacy.load(spacy_model) | |
doc = nlp(text) | |
def get_html(html: str): | |
"""Convert HTML so it can be rendered.""" | |
WRAPPER = """<div style="overflow-x: auto; border: 1px solid #e6e9ef; border-radius: 0.25rem; padding: 1rem; margin-bottom: 2.5rem">{}</div>""" | |
# Newlines seem to mess with the rendering | |
html = html.replace("\n", " ") | |
return WRAPPER.format(html) | |
def get_svg(svg: str, style: str = "", wrap: bool = True): | |
"""Convert an SVG to a base64-encoded image.""" | |
b64 = base64.b64encode(svg.encode("utf-8")).decode("utf-8") | |
html = f'<img src="data:image/svg+xml;base64,{b64}" style="{style}"/>' | |
return get_html(html) if wrap else html | |
def visualize_parser( | |
doc: spacy.tokens.Doc, | |
*, | |
title: Optional[str] = "Dependency parse & part of speech:", | |
key: Optional[str] = None, | |
) -> None: | |
"""Visualizer for dependency parses.""" | |
if title: | |
st.header(title) | |
cols = st.columns(4) | |
split_sents = cols[0].checkbox( | |
"Split sentences", value=True, key=f"{key}_parser_split_sents" | |
) | |
options = { | |
"collapse_punct": cols[1].checkbox( | |
"Collapse punct", value=True, key=f"{key}_parser_collapse_punct" | |
), | |
"compact": cols[3].checkbox("Compact mode", value=False, key=f"{key}_parser_compact"), | |
} | |
docs = [span.as_doc() for span in doc.sents] if split_sents else [doc] | |
for sent in docs: | |
html = displacy.render(sent, options=options, style="dep") | |
# Double newlines seem to mess with the rendering | |
html = html.replace("\n\n", "\n") | |
if split_sents and len(docs) > 1: | |
st.markdown(f"> {sent.text}") | |
st.write(get_svg(html), unsafe_allow_html=True) | |
visualize_parser(doc) | |
st.header("Tree View:") | |
graph_dot = deplacy.dot(doc) | |
st.graphviz_chart(graph_dot) | |
visualize_tokens(doc, attrs=["text", "lemma_", "pos_", "dep_"], title="Table View:") | |