import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt

app_data_balanced = pd.read_csv("sample_deid_balanced_safe.csv")

def show_balanced_record(row_index):
    i = max(0, min(int(float(row_index)), len(app_data_balanced) - 1))
    
    # Convert 1-row DataFrame to a vertical 2-column DataFrame (field, value)
    row_data = app_data_balanced.drop(columns=["_k"]).loc[i]
    record = pd.DataFrame({"field": row_data.index, "value": row_data.values})
    
    k = int(app_data_balanced.loc[i, "_k"])

    verdict = f"k = {k}  ->  SAFE (Balanced Generalization, Strict k>=5)"
    max_k_cap = 15

    fig, ax = plt.subplots(figsize=(6, 4))
    kc = app_data_balanced["_k"].clip(upper=max_k_cap)
    bins = [b + 0.5 for b in range(4, max_k_cap + 2)]
    ax.hist(kc, bins=bins, color="#1b9e77", edgecolor="white")
    ax.axvline(5, color="k", ls="--", lw=1.5, label="k=5 safety bar")
    ax.axvline(min(k, max_k_cap), color="blue", ls="-", lw=3, label=f"Selected Patient (k={k})")

    ax.set_title("Balanced k-Anonymity Distribution")
    ax.set_xlabel(f"k-value (capped at {max_k_cap} for readability)")
    ax.set_ylabel("Number of records")
    ax.legend()
    plt.tight_layout()
    return record, verdict, fig

with gr.Blocks() as demo:
    gr.Markdown("# Balanced Privacy De-Identification Viewer\n"
                "This dashboard runs on the **balanced generalization** dataset (20-yr age bands, 2-digit ZIP). "
                "We preserved ~93.5% of the data while maintaining perfect `k>=5` anonymity!")
    with gr.Row():
        with gr.Column():
            idx = gr.Slider(minimum=0, maximum=len(app_data_balanced)-1, value=0, step=1, label="Search Record by Row Index")
            kmeter = gr.Textbox(label="Re-identification risk (k-anonymity)")
            out = gr.Dataframe(label="De-identified record")
        with gr.Column():
            plot_out = gr.Plot(label="Risk Distribution")

    idx.change(fn=show_balanced_record, inputs=[idx], outputs=[out, kmeter, plot_out])
    demo.load(fn=show_balanced_record, inputs=[idx], outputs=[out, kmeter, plot_out])

demo.launch()