app.py

"""

Hugging Face Space — Indic Language Identifier

Supports 12 Indic languages + English.

Uses best available model: Ensemble > CharNgram > Unicode Heuristic fallback.

"""
from __future__ import annotations

import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).parent))

import gradio as gr
from src.utils.languages import LANGUAGE_REGISTRY, ALL_LANGUAGE_CODES
from src.data.preprocessor import IndicTextPreprocessor
from src.models.ensemble_model import EnsembleLIDModel, UnicodeHeuristicModel

# ── Model Loading ─────────────────────────────────────────────────────────────

preprocessor = IndicTextPreprocessor()
model = None

for mpath in ["./artifacts/models/ensemble.pkl", "./artifacts/models/ngram_model.pkl"]:
    if Path(mpath).exists():
        try:
            import pickle
            with open(mpath, "rb") as f:
                model = pickle.load(f)
            print(f"✓ Loaded: {mpath}")
            break
        except Exception as e:
            print(f"Could not load {mpath}: {e}")

if model is None:
    model = EnsembleLIDModel()
    print("⚠ No trained model found — using Unicode heuristic fallback")


# ── Inference ─────────────────────────────────────────────────────────────────

def predict(text: str) -> str:
    """Clean inference function: text in → language name out."""
    if not text or not text.strip():
        return "Please enter some text."
    processed = preprocessor.preprocess(text)
    if hasattr(model, "predict_with_confidence"):
        result = model.predict_with_confidence([processed])[0]
        lang = result.get("predicted_lang", "?")
        conf = result.get("confidence", 0.0)
    elif hasattr(model, "predict"):
        lang = model.predict([processed])[0]
        conf = 0.9
    else:
        lang, conf = "eng", 0.5

    info = LANGUAGE_REGISTRY.get(lang)
    name = info.name if info else lang
    native = info.native_name if info else ""
    script = info.script if info else "?"
    direction = info.direction if info else "LTR"
    confidence_icon = "🟢" if conf > 0.8 else "🟡" if conf > 0.6 else "🔴"

    return (
        f"**{name}** ({native})\n\n"
        f"Code: `{lang}` | Script: {script} | Direction: {direction}\n\n"
        f"Confidence: {conf:.2%} {confidence_icon}"
    )


def predict_detailed(text: str) -> tuple:
    """Returns (result_markdown, top5_table) for the full UI."""
    if not text or not text.strip():
        return "Please enter some text.", None

    processed = preprocessor.preprocess(text)

    if hasattr(model, "predict_with_confidence"):
        result = model.predict_with_confidence([processed])[0]
        lang = result.get("predicted_lang", "?")
        conf = result.get("confidence", 0.0)
        top5 = result.get("top_5", [])
    elif hasattr(model, "predict"):
        lang = model.predict([processed])[0]
        conf = 0.9
        top5 = [{"lang": lang, "probability": conf}]
    else:
        lang, conf = "eng", 0.5
        top5 = []

    info = LANGUAGE_REGISTRY.get(lang)
    confidence_icon = "🟢" if conf > 0.8 else "🟡" if conf > 0.6 else "🔴"

    result_md = f"""### 🎯 {info.name if info else lang} ({info.native_name if info else ''})



| | |

|---|---|

| Code | `{lang}` |

| Script | {info.script if info else '?'} |

| Direction | {info.direction if info else 'LTR'} |

| Family | {info.family if info else '?'} |

| Confidence | {conf:.2%} {confidence_icon} |

"""

    table = None
    if top5:
        import pandas as pd
        rows = []
        for t in top5:
            i = LANGUAGE_REGISTRY.get(t["lang"])
            rows.append({
                "Language": i.name if i else t["lang"],
                "Code": t["lang"],
                "Probability": f"{t['probability']:.2%}",
            })
        table = pd.DataFrame(rows)

    return result_md, table


# ── Sample Texts ──────────────────────────────────────────────────────────────

SAMPLES = [
    ["भारत एक विशाल देश है जहाँ अनेक भाषाएँ बोली जाती हैं।"],
    ["বাংলাদেশ এবং পশ্চিমবঙ্গে বাংলা ভাষা প্রচলিত।"],
    ["தமிழ் மொழி உலகின் மிகவும் பழமையான மொழிகளில் ஒன்றாகும்."],
    ["India is a diverse country with hundreds of languages."],
    ["اردو پاکستان کی قومی زبان ہے۔"],
    ["ಕನ್ನಡ ಭಾಷೆ ಕರ್ನಾಟಕ ರಾಜ್ಯದ ಅಧಿಕೃತ ಭಾಷೆ."],
]

# ── Gradio UI ─────────────────────────────────────────────────────────────────

with gr.Blocks(title="Indic Language Identifier") as demo:
    gr.Markdown("""

# 🔤 Indic Language Identifier

Detects language from text across **12 Indic languages + English**.

Powered by Unicode heuristics + CharNgram ensemble.

""")

    with gr.Row():
        with gr.Column(scale=2):
            text_input = gr.Textbox(
                label="Input Text",
                placeholder="Type or paste text in any supported language...",
                lines=5,
            )
            with gr.Row():
                clear_btn = gr.Button("Clear", variant="secondary")
                detect_btn = gr.Button("🔍 Detect Language", variant="primary")

            gr.Examples(examples=SAMPLES, inputs=text_input, label="Try a sample")

        with gr.Column(scale=2):
            result_md = gr.Markdown(label="Result")
            top5_table = gr.Dataframe(label="Top Predictions", visible=True)

    detect_btn.click(fn=predict_detailed, inputs=text_input, outputs=[result_md, top5_table])
    clear_btn.click(fn=lambda: ("", None), outputs=[result_md, top5_table])
    text_input.submit(fn=predict_detailed, inputs=text_input, outputs=[result_md, top5_table])

    gr.Markdown("""---

**Supported languages:** Bengali · English · Gujarati · Hindi · Kannada · Malayalam · Marathi · Odia · Punjabi · Tamil · Telugu · Urdu

""")

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