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

tokenizer = AutoTokenizer.from_pretrained("d4data/biomedical-ner-all")
model = AutoModelForTokenClassification.from_pretrained("d4data/biomedical-ner-all")

plt.switch_backend("Agg")

examples = {}
with open("examples.json", "r") as f:
    content = json.load(f)
    examples = {x["text"]: x["label"] 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")
    with gr.Accordion("Examples", open=False):
        examples = gr.Examples(examples=list(examples.keys()), inputs=note)
    with gr.Tab("NER"):
        highlight = gr.HighlightedText(label="NER", combine_adjacent=True)
    with gr.Tab("Bar"):
        plot = gr.Plot(label="Bar")
    with gr.Tab("Table"):
        table = gr.Dataframe(headers=["Entity", "Count"])
    note.submit(run_ner, [note], [highlight, table, plot])

demo.launch()