"""ScamShield NLP — Hugging Face Space entry point.

A pure-inference Gradio app exposing the trained calibrated Linear SVM.

Two ways to use it:
1. Web UI   — type/paste a message and click "Analyze".
2. HTTP API — the same function is exposed through the Gradio API.
"""

import gradio as gr
import spaces

from predictor import ScamPredictor


# ---------------------------------------------------------
# Model
# ---------------------------------------------------------

predictor = ScamPredictor()


# ---------------------------------------------------------
# Configuration
# ---------------------------------------------------------

MAX_MESSAGE_LENGTH = 20000

TITLE = "ScamShield NLP"

DESCRIPTION = (
    "Classifies a text message as **SCAM** or **LEGITIMATE** using a "
    "**calibrated Linear SVM** trained on the ScamShield corpus "
    "(TF-IDF features, n-grams 1–2). Returns a calibrated scam probability "
    "(confidence) and a risk level."
)


# ---------------------------------------------------------
# Example messages
# ---------------------------------------------------------

EXAMPLES = [
    ["Congratulations! You have won a prize. Click this link now."],
    [
        "Dear customer, your account will be locked. "
        "Verify now: http://bit.ly/fake"
    ],
    [
        "URGENT: Your parcel could not be delivered. "
        "Pay $2 to reschedule: https://short.link/x"
    ],
    ["Hi, the meeting is moved to 4pm tomorrow. See you then."],
    [
        "Your OTP for verification is 482913. "
        "Do not share it with anyone."
    ],
    [
        "Reminder: your subscription renews on the 28th. "
        "Manage settings here."
    ],
]


# ---------------------------------------------------------
# Analyze function
# ---------------------------------------------------------

@spaces.GPU
def analyze(message: str) -> dict:
    """Validate input, then run inference with detailed error logging."""

    import traceback

    try:
        # ---------------------------------------------
        # Validate input
        # ---------------------------------------------

        if message is None or not str(message).strip():
            raise gr.Error("Message cannot be empty.")

        message = str(message)

        if len(message) > MAX_MESSAGE_LENGTH:
            raise gr.Error(
                f"Message too long (max {MAX_MESSAGE_LENGTH} characters)."
            )

        # ---------------------------------------------
        # Run model prediction
        # ---------------------------------------------

        print("=" * 60)
        print("ANALYZE REQUEST")
        print(f"Message: {message}")
        print("=" * 60)

        result = predictor.predict(message)

        # ---------------------------------------------
        # Success logging
        # ---------------------------------------------

        print("=" * 60)
        print(f"SUCCESS: {result}")
        print("=" * 60)

        return result

    except gr.Error:
        # Keep Gradio validation errors readable.
        raise

    except Exception as e:
        # ---------------------------------------------
        # Detailed diagnostic logging
        # ---------------------------------------------

        error_msg = (
            f"{type(e).__name__}: {e}\n"
            f"{traceback.format_exc()}"
        )

        print("=" * 60)
        print("ERROR DURING ANALYSIS")
        print(error_msg)
        print("=" * 60)

        # Show a shorter message to the UI while keeping
        # the complete traceback in Hugging Face logs.
        raise gr.Error(
            f"Analysis failed: {type(e).__name__}: {e}"
        )


# ---------------------------------------------------------
# Gradio UI
# ---------------------------------------------------------

with gr.Blocks(
    title=TITLE,
    theme=gr.themes.Soft()
) as demo:

    # ---------------------------------------------
    # Header
    # ---------------------------------------------

    gr.Markdown(f"# {TITLE}")

    gr.Markdown(DESCRIPTION)

    # ---------------------------------------------
    # Main layout
    # ---------------------------------------------

    with gr.Row():

        # -----------------------------------------
        # Input
        # -----------------------------------------

        with gr.Column(scale=3):

            message_input = gr.Textbox(
                label="Message",
                lines=5,
                placeholder=(
                    "Paste the SMS, email or chat message "
                    "to classify…"
                ),
            )

            analyze_button = gr.Button(
                "Analyze",
                variant="primary"
            )

        # -----------------------------------------
        # Output
        # -----------------------------------------

        with gr.Column(scale=2):

            result_output = gr.JSON(
                label="Result"
            )

    # ---------------------------------------------
    # Analyze button
    # ---------------------------------------------

    analyze_button.click(
        analyze,
        inputs=message_input,
        outputs=result_output,
        api_name="predict",
    )

    # ---------------------------------------------
    # Enter / Submit
    # ---------------------------------------------

    message_input.submit(
        analyze,
        inputs=message_input,
        outputs=result_output,
    )

    # ---------------------------------------------
    # Examples + Model information
    # ---------------------------------------------

    gr.Markdown(
        """
### Example messages

Click an example to load it, then press **Analyze**.

### Model

`CalibratedClassifierCV(LinearSVC)` over
`TfidfVectorizer` (10,000 features, n-gram 1–2)

**Held-out test set:**

- Accuracy: **96.91%**
- Precision: **96.64%**
- Recall: **94.99%**
- F1: **95.81%**
"""
    )

    # ---------------------------------------------
    # Example messages
    # ---------------------------------------------

    gr.Examples(
        examples=EXAMPLES,
        inputs=message_input,
        outputs=result_output,
    )


# ---------------------------------------------------------
# Launch
# ---------------------------------------------------------

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