app.py

"""Gradio demo: type a sentence in Amharic or Afaan Oromo, see the topic."""

from __future__ import annotations

import json
import sys
from pathlib import Path

import gradio as gr

ROOT = Path(__file__).resolve().parent
sys.path.insert(0, str(ROOT / "src"))

import predict  # noqa: E402

EXAMPLES_PATH = ROOT / "reports" / "demo_examples.json"

DESCRIPTION = """
News topic classification for **Amharic** and **Afaan Oromo**, two languages
with very little NLP tooling. TF-IDF over character n-grams plus logistic
regression, trained on [MasakhaNEWS](https://huggingface.co/datasets/masakhane/masakhanews).

The language is detected from the script (Ge'ez vs Latin) and routed to the
matching model, so just type — you do not need to pick one. The two languages
have different topic sets: Amharic covers business, health, politics and
sports; Afaan Oromo covers entertainment, health, politics, sports and
technology.

---

Built by **Bekan**.
"""


def load_examples() -> list[list[str]]:
    if not EXAMPLES_PATH.exists():
        return []
    data = json.loads(EXAMPLES_PATH.read_text(encoding="utf-8"))
    return [[item["text"]] for entries in data.values() for item in entries]


def classify(text: str) -> tuple[dict, str]:
    result = predict.predict(text)
    if not result["predictions"]:
        return {}, "Type a sentence to see a prediction."

    confidences = {p["label"]: p["confidence"] for p in result["predictions"]}
    top = result["predictions"][0]
    note = f"Detected **{result['language_name']}** from the script."
    if top["confidence"] < 0.5:
        note += " Confidence is low here — the model is genuinely unsure."
    return confidences, note


with gr.Blocks(title="Amharic / Afaan Oromo news topic classifier") as demo:
    gr.Markdown("# Amharic / Afaan Oromo news topic classifier")
    gr.Markdown(DESCRIPTION)

    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(
                label="Your sentence",
                placeholder="Barreeffama afaan Oromoo yookaan Amaariffaa asitti barreessi…",
                lines=3,
            )
            submit = gr.Button("Classify", variant="primary")
        with gr.Column():
            label_output = gr.Label(label="Predicted topic", num_top_classes=3)
            note_output = gr.Markdown()

    gr.Examples(
        examples=load_examples(),
        inputs=text_input,
        label="Held-out test headlines to try",
    )

    submit.click(classify, inputs=text_input, outputs=[label_output, note_output])
    text_input.submit(classify, inputs=text_input, outputs=[label_output, note_output])

if __name__ == "__main__":
    demo.launch(favicon_path=ROOT / "favicon.svg")