src/export_utils.py

import json
from io import BytesIO

import plotly.graph_objects as go
from reportlab.lib import colors
from reportlab.lib.pagesizes import letter
from reportlab.lib.styles import getSampleStyleSheet
from reportlab.platypus import Image, Paragraph, Preformatted, SimpleDocTemplate, Spacer, Table, TableStyle


def _figure_to_image(figure, width=640, height=300):
    """Convert a Plotly figure to a reportlab Image. Returns None if conversion fails."""
    try:
        image_bytes = figure.to_image(format="png", width=width, height=height, scale=2)
        return Image(BytesIO(image_bytes), width=width * 0.75, height=height * 0.75)
    except Exception:
        return None


def _metrics_table(rows):
    table = Table(rows, hAlign="LEFT")
    table.setStyle(
        TableStyle(
            [
                ("BACKGROUND", (0, 0), (-1, 0), colors.HexColor("#f3f4f6")),
                ("TEXTCOLOR", (0, 0), (-1, 0), colors.HexColor("#111827")),
                ("ALIGN", (0, 0), (-1, -1), "LEFT"),
                ("FONTNAME", (0, 0), (-1, 0), "Helvetica-Bold"),
                ("FONTSIZE", (0, 0), (-1, -1), 9),
                ("GRID", (0, 0), (-1, -1), 0.5, colors.HexColor("#d1d5db")),
                ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
            ]
        )
    )
    return table


def _single_model_section(story, model_name, model_payload, styles):
    prediction = "HATE SPEECH DETECTED" if model_payload["prediction"] == 1 else "NOT HATE SPEECH"
    prediction_color = "#c62828" if model_payload["prediction"] == 1 else "#2e7d32"

    story.append(Paragraph(f"<b>{model_name}</b>", styles["Heading3"]))
    story.append(Paragraph(f"<font color='{prediction_color}'><b>{prediction}</b></font>", styles["Normal"]))
    story.append(Spacer(1, 6))

    rows = [
        ["Metric", "Value"],
        ["Confidence", f"{model_payload['confidence']:.1%}"],
        ["Not Hate Speech", f"{model_payload['probabilities'][0]:.1%}"],
        ["Hate Speech", f"{model_payload['probabilities'][1]:.1%}"],
        ["Processing Time", f"{model_payload['processing_time']:.3f}s"],
    ]
    story.append(_metrics_table(rows))
    story.append(Spacer(1, 8))

    probability_chart = go.Figure(
        data=[
            go.Bar(
                x=["Not Hate Speech", "Hate Speech"],
                y=model_payload["probabilities"],
                marker_color=["#66bb6a", "#ef5350"],
                text=[f"{value:.1%}" for value in model_payload["probabilities"]],
                textposition="auto",
            )
        ]
    )
    probability_chart.update_layout(yaxis_range=[0, 1], height=300, showlegend=False, margin=dict(l=20, r=20, t=20, b=20))

    chart_image = _figure_to_image(probability_chart)
    if chart_image:
        story.append(Paragraph("Probability Distribution", styles["Italic"]))
        story.append(chart_image)
        story.append(Spacer(1, 8))

    token_rows = model_payload.get("token_importance") or []
    if token_rows:
        story.append(Paragraph("Top Important Tokens", styles["Italic"]))
        token_table_rows = [["Token", "Importance"]] + [
            [token_data["Token"], f"{token_data['Importance']:.4f}"] for token_data in token_rows
        ]
        story.append(_metrics_table(token_table_rows))
        story.append(Spacer(1, 8))

    technical_details = model_payload.get("technical_details")
    if technical_details:
        story.append(Paragraph("Technical Details", styles["Italic"]))
        story.append(Preformatted(json.dumps(technical_details, indent=2), styles["Code"]))

    story.append(Spacer(1, 12))


def _batch_model_section(story, model_name, model_payload, styles, colorscale):
    story.append(Paragraph(f"<b>{model_name}</b>", styles["Heading3"]))
    story.append(Spacer(1, 4))

    rows = [
        ["Metric", "Value"],
        ["F1 Score", f"{model_payload['f1_score']:.4f}"],
        ["Precision", f"{model_payload['precision']:.4f}"],
        ["Accuracy", f"{model_payload['accuracy']:.4f}"],
        ["Recall", f"{model_payload['recall']:.4f}"],
        ["Avg CPU", f"{model_payload['cpu_usage']:.2f}%"],
        ["Peak CPU", f"{model_payload['peak_cpu_usage']:.2f}%"],
        ["Avg Memory", f"{model_payload['memory_usage']:.2f} MB"],
        ["Peak Memory", f"{model_payload['peak_memory_usage']:.2f} MB"],
        ["Total Runtime", f"{model_payload['runtime']:.2f}s"],
        ["Avg Time/Sample", f"{model_payload['avg_time_per_sample']:.3f}s"],
    ]
    story.append(_metrics_table(rows))
    story.append(Spacer(1, 8))

    confusion_matrix = model_payload.get("confusion_matrix")
    if confusion_matrix is not None:
        confusion_fig = go.Figure(
            data=go.Heatmap(
                z=confusion_matrix,
                x=["Pred Not Hate", "Pred Hate"],
                y=["True Not Hate", "True Hate"],
                colorscale=colorscale,
                text=confusion_matrix,
                texttemplate="%{text}",
                textfont={"size": 14},
                showscale=False,
            )
        )
        confusion_fig.update_layout(height=300, margin=dict(l=20, r=20, t=20, b=20))
        confusion_image = _figure_to_image(confusion_fig)
        if confusion_image:
            story.append(Paragraph("Confusion Matrix", styles["Italic"]))
            story.append(confusion_image)

    story.append(Spacer(1, 12))


def generate_results_pdf(payload):
    """Generate a PDF from the current Streamlit results payload."""
    output_buffer = BytesIO()
    document = SimpleDocTemplate(
        output_buffer,
        pagesize=letter,
        leftMargin=36,
        rightMargin=36,
        topMargin=36,
        bottomMargin=36,
        title="Hate Speech Detection Results",
    )

    styles = getSampleStyleSheet()
    story = [
        Paragraph("Hate Speech Detection Results", styles["Title"]),
        Spacer(1, 8),
    ]

    mode = payload.get("mode", "single")
    if mode == "single":
        story.append(Paragraph("Single Text Analysis", styles["Heading2"]))
        story.append(Spacer(1, 6))
        story.append(Paragraph(f"<b>Input Text:</b> {payload.get('input_text', '')}", styles["Normal"]))
        rationale_text = payload.get("rationale")
        if rationale_text:
            story.append(Spacer(1, 4))
            story.append(Paragraph(f"<b>Rationale:</b> {rationale_text}", styles["Normal"]))
        story.append(Spacer(1, 10))

        models = payload.get("models", {})
        if "base" in models:
            _single_model_section(story, "Base Bert Ensemble Results", models["base"], styles)
        if "enhanced" in models:
            _single_model_section(story, "Enhanced Bert Ensemble Results", models["enhanced"], styles)

    elif mode == "batch":
        story.append(Paragraph("Batch File Analysis", styles["Heading2"]))
        story.append(Spacer(1, 6))
        story.append(Paragraph(f"<b>Filename:</b> {payload.get('filename', 'Unknown')}", styles["Normal"]))
        story.append(Paragraph(f"<b>Rows:</b> {payload.get('rows', 0)}", styles["Normal"]))
        story.append(Spacer(1, 10))

        models = payload.get("models", {})
        if "base" in models:
            _batch_model_section(story, "Base Bert Ensemble Results", models["base"], styles, "Blues")
        if "enhanced" in models:
            _batch_model_section(story, "Enhanced Bert Ensemble Results", models["enhanced"], styles, "Greens")

    else:
        story.append(Paragraph("No exportable results found.", styles["Normal"]))

    document.build(story)
    return output_buffer.getvalue()