"""mundart-explorer — Swiss-German (gsw) language-ID + gsw→de alignment demo.

Paste a sentence; the Space says whether it reads as written Swiss-German or
Standard German (with a confidence) and surfaces the nearest aligned Standard-
German rendering from the gsw-eval probe set. All logic lives in
:mod:`lid_service`; this file is the thin Gradio UI only.
"""

from __future__ import annotations

import gradio as gr
import lid_service as svc

_HUB = "https://huggingface.co/datasets/mischeiwiller"
_CORPUS_URL = f"{_HUB}/swiss-german-text"
_EVAL_URL = f"{_HUB}/gsw-eval"

_EXAMPLES = [
    "Ich ha hüt es Brötli gässe und bi denn go poschte.",
    "Mir gönd am Samschtig go bärgsteige, wenn s Wätter schön isch.",
    "Chasch mir bitte säge, wo de Bahnhof isch?",
    "Was machsch du hüt z Aabig?",
    "Was machst du heute Abend?",
    "Guten Morgen!",
]


def analyze(text: str) -> tuple[str, str]:
    """Run LID + alignment lookup; return (verdict markdown, alignment markdown)."""
    text = (text or "").strip()
    if not text:
        return "_Paste a sentence above to begin._", ""

    res = svc.identify(text)
    pct = f"{res['confidence'] * 100:.0f}%"
    flag = "🇨🇭" if res["label"] == "gsw" else "🇩🇪"
    verdict = (
        f"### {flag} {res['label_long']}\n"
        f"**Confidence:** {pct}  \n"
        f"<sub>gsw {res['probs']['gsw'] * 100:.0f}% · de {res['probs']['de'] * 100:.0f}% "
        f"— char n-gram Naive Bayes baseline</sub>"
    )

    align = svc.nearest_alignment(text)
    if align is None:
        alignment = (
            "_No close Standard-German alignment in the probe set "
            f"({svc.probe_size()} gsw→de pairs)._"
        )
    else:
        sim = f"{align['similarity'] * 100:.0f}%"
        alignment = (
            f"**Nearest aligned pair** (similarity {sim}):\n\n"
            f"> 🇨🇭 {align['gsw']}\n>\n"
            f"> 🇩🇪 {align['de']}"
        )
    return verdict, alignment


with gr.Blocks(title="mundart-explorer", fill_width=True) as demo:
    gr.Markdown(
        "# 🇨🇭 mundart-explorer\n"
        "Is it written **Swiss-German (gsw)** or **Standard German (de)**? "
        "Paste a sentence — the demo also surfaces the nearest aligned German "
        f"rendering from the [`gsw-eval`]({_EVAL_URL}) probe set."
    )
    inp = gr.Textbox(
        label="Sentence",
        placeholder="Schrib öppis uf Schwiizerdütsch oder Hochdütsch …",
        lines=3,
        autofocus=True,
    )
    btn = gr.Button("Analyze", variant="primary")
    verdict = gr.Markdown()
    alignment = gr.Markdown()
    gr.Examples(examples=_EXAMPLES, inputs=inp)

    btn.click(analyze, inputs=inp, outputs=[verdict, alignment])
    inp.submit(analyze, inputs=inp, outputs=[verdict, alignment])

    gr.Markdown(
        "<sub>LID baseline: deterministic char n-gram multinomial Naive Bayes "
        "(macro-F1 0.63 on the held-out test split). Part of the **Mundart** project — "
        f"[`swiss-german-text`]({_CORPUS_URL}) corpus + "
        f"[`gsw-eval`]({_EVAL_URL}) benchmark.</sub>"
    )

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