app.py

from __future__ import annotations

import os
from pathlib import Path

import gradio as gr
from huggingface_hub import snapshot_download

from src.sentence_classifier.predict import DEFAULT_ARTIFACT_DIR, SentenceFunctionClassifier


def resolve_artifact_dir() -> Path:
    local_dir = Path(os.environ.get("MODEL_DIR", DEFAULT_ARTIFACT_DIR))
    if (local_dir / "classifier.joblib").exists():
        return local_dir

    model_repo_id = os.environ.get("MODEL_REPO_ID")
    if model_repo_id:
        return Path(
            snapshot_download(
                repo_id=model_repo_id,
                allow_patterns=["*.joblib", "metadata.json"],
            )
        )

    return local_dir


classifier: SentenceFunctionClassifier | None = None
load_error: str | None = None

try:
    classifier = SentenceFunctionClassifier(resolve_artifact_dir())
except Exception as exc:
    load_error = str(exc)


def classify(sentence: str) -> tuple[str, dict[str, float]]:
    if classifier is None:
        message = (
            "Model artifacts are not available yet. Train locally with "
            "`python -m src.sentence_classifier.train`, or set MODEL_REPO_ID "
            "in the Hugging Face Space."
        )
        if load_error:
            message = f"{message}\n\nLoad error: {load_error}"
        return message, {}

    try:
        prediction = classifier.predict(sentence)
    except ValueError as exc:
        return str(exc), {}

    label = prediction.label.replace("_", " ").title()
    summary = f"{label} ({prediction.confidence:.1%} confidence)"
    return summary, prediction.probabilities


examples = [
    "The lecture starts at nine.",
    "Please open the window.",
    "Where did you put the keys?",
    "What a remarkable answer!",
    "May your work be rewarded.",
]


demo = gr.Interface(
    fn=classify,
    inputs=gr.Textbox(label="Sentence", placeholder="Enter one sentence..."),
    outputs=[
        gr.Textbox(label="Prediction"),
        gr.Label(label="Class probabilities"),
    ],
    examples=examples,
    title="Sentence Function Classifier",
    description=(
        "Classifies sentences as declarative, imperative, interrogative, "
        "exclamatory, or optative."
    ),
)


if __name__ == "__main__":
    demo.launch()