app.py

import json
from collections import defaultdict, Counter

import matplotlib.pyplot as plt
import gradio as gr
import pandas as pd
from transformers import pipeline, AutoTokenizer, AutoModelForTokenClassification

MODELS = ["d4data/biomedical-ner-all"]

current_model = MODELS[0]

tokenizer = AutoTokenizer.from_pretrained(current_model)
model = AutoModelForTokenClassification.from_pretrained(current_model)

plt.switch_backend("Agg")

examples = []
with open("examples.json", "r") as f:
    content = json.load(f)
    examples = [f"{x['label']}: {x['text']}" for x in content]

pipe = pipeline("ner", model=model, tokenizer=tokenizer, aggregation_strategy="simple")


def plot_to_figure(grouped):
    fig = plt.figure()
    plt.bar(x=list(grouped.keys()), height=list(grouped.values()))
    plt.margins(0.2)
    plt.subplots_adjust(bottom=0.4)
    plt.xticks(rotation=90)
    return fig


def run_ner(text):
    raw = pipe(text)
    ner_content = {
        "text": text,
        "entities": [
            {
                "entity": x["entity_group"],
                "word": x["word"],
                "score": x["score"],
                "start": x["start"],
                "end": x["end"],
            }
            for x in raw
        ],
    }
    grouped = Counter((x["entity_group"] for x in raw))
    rows = [[k, v] for k, v in grouped.items()]
    figure = plot_to_figure(grouped)
    return ner_content, rows, figure


with gr.Blocks() as demo:
    note = gr.Textbox(label="Note text")
    submit = gr.Button("Submit")
    # with gr.Accordion("Examples", open=False):
    example_dropdown = gr.Dropdown(label="Examples", choices=examples)
    example_dropdown.change(
        lambda x: gr.Textbox.update(value=x), inputs=example_dropdown, outputs=note
    )
    highlight = gr.HighlightedText(label="NER", combine_adjacent=True)
    table = gr.Dataframe(headers=["Entity", "Count"])
    plot = gr.Plot(label="Bar")
    submit.click(run_ner, [note], [highlight, table, plot])
    note.submit(run_ner, [note], [highlight, table, plot])

demo.launch()