import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
from theme import custom_css, header


# --------------------------
#  Model setup
# --------------------------
MODEL_ID = "roncc13/autotrain-ixzm9-t6dbc"


tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)


label_names = ["fake", "real"]


def classify(text: str):
    if not text.strip():
        return {"fake": 0.0, "real": 0.0}
    inputs = tokenizer(
        text,
        return_tensors="pt",
        truncation=True,
        padding=True,
        max_length=256,
    )
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=-1)[0].tolist()
    return {label_names[i]: float(probs[i]) for i in range(len(label_names))}


# --------------------------
#  Helpers for rendering result HTML
# --------------------------
def _result_label_html(label: str, css_class: str) -> str:
    return f'<span class="{css_class}">{label}</span>'


def _confidence_text_html(conf_pct: int) -> str:
    return (
        "<div class='conf-text-row'>"
        f"<span class='conf-val'>{conf_pct}%</span>"
        "<span class='conf-caption'>confidence</span>"
        "</div>"
    )


def _confidence_bar_html(conf_pct: int) -> str:
    return (
        "<div class='conf-bar-bg'>"
        f"<div class='conf-bar-fill' style='width:{conf_pct}%;'></div>"
        "</div>"
    )


# Initial (empty) state values for the result card
INITIAL_LABEL_HTML = _result_label_html("—", "badge-pill badge-neutral")
INITIAL_CONF_HTML = _confidence_text_html(0)
INITIAL_BAR_HTML = _confidence_bar_html(0)


# --------------------------
#  UI with Tabs
# --------------------------
with gr.Blocks(fill_height=True) as demo:
    gr.HTML("<div style='height:12px;'></div>")

    # ===== Analyzer tab =====
    with gr.Tab("Analyzer"):
        header()

        gr.HTML(
            """
            <section class="hero-section">
              <div class="hero-title">
                Check Cebuano text for a misleading writing style.
              </div>
              <div class="hero-subtitle">
                This tool analyzes linguistic patterns and writing style in Cebuano text to detect potential
                misinformation. It does not verify factual correctness. The model returns a classification
                (Fake/Legit) and a confidence score based on writing patterns.
              </div>
            </section>
            """
        )

        with gr.Row(elem_classes=["two-col"], equal_height=True):
            # Left: input card
            with gr.Column(scale=3):
                with gr.Group(elem_classes=["glass-card"], elem_id="input-card"):
                    gr.Markdown(
                        "#### Text input\n"
                        "Cebuano only. This tool checks linguistic patterns; it does not verify facts."
                    )
                    gr.Markdown(
                        "> **Example**  \n"
                        "> \u201cNakadisubre og milagro nga tambal sa COVID\u201119 ang usa ka local doktor, "
                        "giingon nga walay side effects ug dili kinahanglan og bakuna.\u201d"
                    )
                    news_text = gr.Textbox(
                        lines=7,
                        label="",
                        placeholder="Paste Cebuano news text here...",
                        elem_id="news-textbox",
                    )
                    with gr.Row(elem_classes=["btn-row"]):
                        analyze_btn = gr.Button("Analyze", elem_classes=["btn-primary-custom"])
                        clear_btn = gr.Button("Clear", elem_classes=["btn-secondary-custom"])
                    gr.Markdown(
                        "<span class='helper-text'>"
                        "Tip: Keep inputs under 1,000 characters for faster results."
                        "</span>",
                        container=False,
                    )

            # Right: result card
            with gr.Column(scale=2):
                with gr.Group(elem_classes=["glass-card"], elem_id="result-card"):
                    gr.Markdown("#### Result")
                    result_label_html = gr.HTML(INITIAL_LABEL_HTML)
                    conf_text = gr.HTML(INITIAL_CONF_HTML)
                    conf_bar = gr.HTML(INITIAL_BAR_HTML)
                    gr.Markdown(
                        "<span class='helper-text'>"
                        "Model: CMD\u2011BERT (fine\u2011tuned BERT\u2011base). "
                        "Output: Label and confidence score for the submitted text."
                        "</span>",
                        container=False,
                    )

        def analyze_ui(text):
            if not text or not text.strip():
                return INITIAL_LABEL_HTML, INITIAL_CONF_HTML, INITIAL_BAR_HTML

            probs = classify(text)
            fake_p = probs.get("fake", 0.0)
            real_p = probs.get("real", 0.0)
            if fake_p >= real_p:
                label, css_class, conf = "FAKE", "badge-pill badge-fake", fake_p
            else:
                label, css_class, conf = "LEGIT", "badge-pill badge-real", real_p

            # Clamp and convert to integer percentage 0–100
            conf_pct = max(0, min(100, int(round(conf * 100))))

            return (
                _result_label_html(label, css_class),
                _confidence_text_html(conf_pct),
                _confidence_bar_html(conf_pct),
            )

        def clear_ui():
            return "", INITIAL_LABEL_HTML, INITIAL_CONF_HTML, INITIAL_BAR_HTML

        analyze_btn.click(
            fn=analyze_ui,
            inputs=news_text,
            outputs=[result_label_html, conf_text, conf_bar],
        )
        clear_btn.click(
            fn=clear_ui,
            inputs=None,
            outputs=[news_text, result_label_html, conf_text, conf_bar],
        )

    # ===== How it works tab =====
    with gr.Tab("How it works"):
        header()
        with gr.Group(elem_classes=["glass-card"], elem_id="hiw-intro-card"):
            gr.Markdown(
                "## How CMD\u2011BERT works\n"
                "CMD\u2011BERT is an AI\u2011augmented linguistic model that focuses on writing style, "
                "not literal truth. It looks for patterns such as exaggerated wording, "
                "over\u2011confident claims, and framing that often appear in misleading content."
            )
        with gr.Row(elem_classes=["card-row"]):
            with gr.Column():
                with gr.Group(elem_classes=["glass-card"], elem_id="hiw-step1-card"):
                    gr.Markdown(
                        "### 1. Input and preprocessing\n"
                        "- User pastes a Cebuano headline, post, or short article.\n"
                        "- The text is tokenized and trimmed to a safe maximum length.\n"
                        "- Inputs are processed in memory and not stored permanently."
                    )
            with gr.Column():
                with gr.Group(elem_classes=["glass-card"], elem_id="hiw-step2-card"):
                    gr.Markdown(
                        "### 2. CMD\u2011BERT analysis\n"
                        "- CMD\u2011BERT is a fine\u2011tuned BERT\u2011base model trained on Cebuano news.\n"
                        "- It computes probabilities for two classes: **Fake** and **Legit**.\n"
                        "- The highest\u2011probability class becomes the predicted label."
                    )
        with gr.Group(elem_classes=["glass-card"], elem_id="hiw-step3-card"):
            gr.Markdown(
                "### 3. Result and interpretation\n"
                "- The interface shows the predicted label and confidence bar.\n"
                "- Users are reminded that this is a screening tool only.\n"
                "- Final judgment should always involve human critical thinking."
            )

    # ===== About tab =====
    with gr.Tab("About"):
        header()
        with gr.Group(elem_classes=["glass-card"], elem_id="about-intro-card"):
            gr.Markdown(
                "## About CMD\u2011BERT\n"
                "**CMD\u2011BERT: An AI Augmented Linguistic Recognition Model for Cebuano Fake News Detection**\n\n"
                "CMD\u2011BERT is a thesis project in the Department of Computer Engineering at "
                "Cebu Technological University\u2013Main Campus. The tool aims to support Cebuano readers "
                "by highlighting potentially misleading writing patterns in online news and posts."
            )
        with gr.Group(elem_classes=["glass-card"], elem_id="about-thesis-card"):
            gr.Markdown(
                "### Thesis information\n"
                "_A Thesis Project presented to the Faculty of the Department of Computer Engineering_\n\n"
                "Cebu Technological University\u2013Main Campus  \n"
                "Cebu City, Philippines  \n\n"
                "_In partial fulfillment of the requirements for the degree_  \n"
                "**Bachelor of Science in Computer Engineering**\n\n"
                "**By:**  \n"
                "- Cabag, Ronilo Jose Jr. S.  \n"
                "- Libron, Andio Mart  \n"
                "- Omega, Noel  \n\n"
                "**Adviser:** Engr. Jueco, M.Eng.  \n"
                "January 2026"
            )

    # ===== Feedback tab =====
    with gr.Tab("Feedback"):
        header()
        with gr.Group(elem_classes=["glass-card"], elem_id="fb-intro-card"):
            gr.Markdown(
                "## Feedback and model improvement\n"
                "CMD\u2011BERT is experimental and continuously improving. Your feedback can help "
                "identify model mistakes, usability issues, and opportunities to refine the dataset."
            )
        with gr.Row(elem_classes=["card-row"]):
            with gr.Column():
                with gr.Group(elem_classes=["glass-card"], elem_id="fb-form-card"):
                    fb_type = gr.Dropdown(
                        ["Bug / technical issue", "Model mistake", "UI suggestion", "Other"],
                        label="Feedback type",
                    )
                    fb_text = gr.Textbox(
                        lines=6,
                        label="Your message or example text",
                        placeholder="Describe the issue or paste an example of text the model misclassified.",
                        elem_id="fb-textbox",
                    )
                    fb_email = gr.Textbox(
                        label="Email (optional, for follow\u2011up)",
                        placeholder="you@example.com",
                        elem_id="fb-email-textbox",
                    )
                    fb_checkbox = gr.Checkbox(
                        label="Allow us to use this text anonymously for future model improvements.",
                        value=True,
                    )
                    fb_submit = gr.Button("Submit feedback", elem_classes=["btn-primary-custom"])
            with gr.Column():
                with gr.Group(elem_classes=["glass-card"], elem_id="fb-faq-card"):
                    fb_status = gr.Markdown("No feedback submitted yet.")
                    gr.Markdown(
                        "### FAQ\n"
                        "**What happens to my feedback?**  \n"
                        "It is stored securely and reviewed by the CMD\u2011BERT thesis team.\n\n"
                        "**Will CMD\u2011BERT replace human fact\u2011checkers?**  \n"
                        "No. It is a support tool to encourage critical reading.\n\n"
                        "**Who maintains this tool?**  \n"
                        "The CMD\u2011BERT thesis team at Cebu Technological University\u2013Main Campus."
                    )

        def save_feedback(ftype, text, email, consent):
            if not text.strip():
                return "Please enter a message before submitting."
            return "Thank you for your feedback! It has been recorded."

        fb_submit.click(
            fn=save_feedback,
            inputs=[fb_type, fb_text, fb_email, fb_checkbox],
            outputs=fb_status,
        )


if __name__ == "__main__":
    demo.launch(css=custom_css, theme=gr.themes.Soft())