app.py

import gradio as gr

from bertopic import BERTopic
from datasets import load_dataset
from functools import lru_cache


def prep_dataset():
    dataset = load_dataset("OpenAssistant/oasst1", split="train")
    assistant_ds = dataset.filter(lambda x: x["role"] == "assistant")
    assistant_ds_en = assistant_ds.filter(lambda x: x["lang"] == "en")
    return assistant_ds_en["text"]


topic_model = BERTopic.load("davanstrien/chat_topics")

fig = topic_model.visualize_topics()


def plot_docs():
    docs = prep_dataset()
    return topic_model.visualize_documents(docs)


def search_topic(text):
    similar_topics, _ = topic_model.find_topics(text, top_n=5)
    topic_info = topic_model.get_topic_info()
    return topic_info[topic_info["Topic"].isin(similar_topics)]


def plot_topic_words(num_topics=9, n_words=5):
    return topic_model.visualize_barchart(top_n_topics=num_topics, n_words=n_words)


with gr.Blocks() as demo:
    with gr.Tab("Topic words"):
        topic_number = gr.Slider(
            minimum=3, maximum=20, value=9, step=1, label="Number of topics"
        )
        plot = gr.Plot(plot_topic_words())
        topic_number.change(plot_topic_words, [topic_number], plot)
    with gr.Tab("Topic search"):
        text = gr.Textbox(lines=1, label="Search text")
        df = gr.DataFrame()
        text.change(search_topic, [text], df)
    with gr.Tab("Topic distribution"):
        gr.Plot(fig)

    # with gr.Tab("Doc visualization"):
    #     gr.Plot(plot_docs())

demo.launch(debug=True)