app.py

import pandas as pd
import numpy as np
from scipy.stats import ttest_rel
import seaborn as sns
import matplotlib.pyplot as plt
import gradio as gr

# Load the results CSV
df_results = pd.read_csv("model_results.csv")

# Extract available models and datasets
ALL_MODELS = sorted(df_results['model'].unique().tolist())
AVAILABLE_DATASETS = sorted(df_results['dataset'].unique().tolist())

# Dataset categorization
DATASET_CATEGORIES = {
    "Medical & Healthcare": {
        "D1": "Heart Disease (Comprehensive)",
        "D2": "Heart attack possibility",
        "D3": "Heart Disease Dataset",
        "D4": "Liver Disorders",
        "D5": "Diabetes Prediction",
        "D9": "Chronic Kidney Disease",
        "D10": "Breast Cancer Prediction",
        "D11": "Stroke Prediction",
        "D12": "Lung Cancer Prediction",
        "D13": "Hepatitis",
        "D15": "Thyroid Disease",
        "D16": "Heart Failure Prediction",
        "D17": "Parkinson's",
        "D18": "Indian Liver Patient",
        "D19": "COVID-19 Effect on Liver Cancer",
        "D20": "Liver Dataset",
        "D21": "Specht Heart",
        "D22": "Early-stage Diabetes",
        "D23": "Diabetic Retinopathy",
        "D24": "Breast Cancer Coimbra",
        "D25": "Chronic Kidney Disease",
        "D26": "Kidney Stone",
        "D28": "Echocardiogram",
        "D29": "Bladder Cancer Recurrence",
        "D31": "Prostate Cancer",
        "D46": "Real Breast Cancer Data",
        "D47": "Breast Cancer (Royston)",
        "D48": "Lung Cancer Dataset",
        "D52": "Cervical Cancer Risk",
        "D53": "Breast Cancer Wisconsin",
        "D61": "Breast Cancer Prediction",
        "D62": "Thyroid Disease",
        "D68": "Lung Cancer",
        "D69": "Cancer Patients Data",
        "D70": "Labor Relations",
        "D71": "Glioma Grading",
        "D74": "Post-Operative Patient",
        "D80": "Heart Rate Stress Monitoring",
        "D82": "Diabetes 2019",
        "D87": "Personal Heart Disease Indicators",
        "D92": "Heart Disease (Logistic)",
        "D95": "Diabetes Prediction",
        "D97": "Cardiovascular Disease",
        "D98": "Diabetes 130 US Hospitals",
        "D99": "Heart Disease Dataset",
        "D181": "HCV Data",
        "D184": "Cardiotocography",
        "D189": "Mammographic Mass",
        "D199": "Easiest Diabetes",
        "D200": "Monkey-Pox Patients",
        "D54": "Breast Cancer Wisconsin",
        "D63": "Sick-euthyroid",
        "D64": "Ann-test",
        "D65": "Ann-train",
        "D66": "Hypothyroid",
        "D67": "New-thyroid",
        "D72": "Glioma Grading",
    },

    "Gaming & Sports": {
        "D27": "Chess King-Rook",
        "D36": "Tic-Tac-Toe",
        "D40": "IPL 2022 Matches",
        "D41": "League of Legends",
        "D55": "League of Legends Diamond",
        "D56": "Chess Game Dataset",
        "D57": "Game of Thrones",
        "D73": "Connect-4",
        "D75": "FIFA 2018",
        "D76": "Dota 2 Matches",
        "D77": "IPL Match Analysis",
        "D78": "CS:GO Professional",
        "D79": "IPL 2008-2022",
        "D114": "Video Games",
        "D115": "Video Games Sales",
        "D117": "Sacred Games",
        "D118": "PC Games Sales",
        "D119": "Popular Video Games",
        "D120": "Olympic Games 2021",
        "D121": "Video Games ESRB",
        "D122": "Top Play Store Games",
        "D123": "Steam Games",
        "D124": "PS4 Games",
        "D116": "Video Games Sales",
    },

    "Education & Students": {
        "D43": "Student Marks",
        "D44": "Student 2nd Year Result",
        "D45": "Student Mat Pass/Fail",
        "D103": "Academic Performance",
        "D104": "Student Academic Analysis",
        "D105": "Student Dropout Prediction",
        "D106": "Electronic Gadgets Impact",
        "D107": "Campus Recruitment",
        "D108": "End-Semester Performance",
        "D109": "Fitbits and Grades",
        "D110": "Student Time Management",
        "D111": "Student Feedback",
        "D112": "Depression & Performance",
        "D113": "University Rankings",
        "D126": "University Ranking CWUR",
        "D127": "University Ranking CWUR 2013-2014",
        "D128": "University Ranking CWUR 2014-2015",
        "D129": "University Ranking CWUR 2015-2016",
        "D130": "University Ranking CWUR 2016-2017",
        "D131": "University Ranking CWUR 2017-2018",
        "D132": "University Ranking CWUR 2018-2019",
        "D133": "University Ranking CWUR 2019-2020",
        "D134": "University Ranking CWUR 2020-2021",
        "D135": "University Ranking CWUR 2021-2022",
        "D136": "University Ranking CWUR 2022-2023",
        "D137": "University Ranking GM 2016",
        "D138": "University Ranking GM 2017",
        "D139": "University Ranking GM 2018",
        "D140": "University Ranking GM 2019",
        "D141": "University Ranking GM 2020",
        "D142": "University Ranking GM 2021",
        "D143": "University Ranking GM 2022",
        "D144": "University Ranking Webometric 2012",
        "D145": "University Ranking Webometric 2013",
        "D146": "University Ranking Webometric 2014",
        "D147": "University Ranking Webometric 2015",
        "D148": "University Ranking Webometric 2016",
        "D149": "University Ranking Webometric 2017",
        "D150": "University Ranking Webometric 2018",
        "D151": "University Ranking Webometric 2019",
        "D152": "University Ranking Webometric 2020",
        "D153": "University Ranking Webometric 2021",
        "D154": "University Ranking Webometric 2022",
        "D155": "University Ranking Webometric 2023",
        "D156": "University Ranking URAP 2018-2019",
        "D157": "University Ranking URAP 2019-2020",
        "D158": "University Ranking URAP 2020-2021",
        "D159": "University Ranking URAP 2021-2022",
        "D160": "University Ranking URAP 2022-2023",
        "D161": "University Ranking THE 2011",
        "D162": "University Ranking THE 2012",
        "D163": "University Ranking THE 2013",
        "D164": "University Ranking THE 2014",
        "D165": "University Ranking THE 2015",
        "D166": "University Ranking THE 2016",
        "D167": "University Ranking THE 2017",
        "D168": "University Ranking THE 2018",
        "D169": "University Ranking THE 2019",
        "D170": "University Ranking THE 2020",
        "D171": "University Ranking THE 2021",
        "D172": "University Ranking THE 2022",
        "D173": "University Ranking THE 2023",
        "D174": "University Ranking QS 2022",
        "D190": "Student Academics Performance"
    },

    "Banking & Finance": {
        "D6": "Bank Marketing 1",
        "D7": "Bank Marketing 2",
        "D30": "Adult Income",
        "D32": "Telco Customer Churn",
        "D35": "Credit Approval",
        "D50": "Term Deposit Prediction",
        "D96": "Credit Card Fraud",
        "D188": "South German Credit",
        "D193": "Credit Risk Classification",
        "D195": "Credit Score Classification",
        "D196": "Banking Classification"
    },

    "Science & Engineering": {
        "D8": "Mushroom",
        "D14": "Ionosphere",
        "D33": "EEG Eye State",
        "D37": "Steel Plates Faults",
        "D39": "Fertility",
        "D51": "Darwin",
        "D58": "EEG Emotions",
        "D81": "Predictive Maintenance",
        "D84": "Oranges vs Grapefruit",
        "D90": "Crystal System Li-ion",
        "D183": "Drug Consumption",
        "D49": "Air Pressure System Failures",
        "D93": "Air Pressure System Failures",
        "D185": "Toxicity",
        "D186": "Toxicity",
    },

    "Social & Lifestyle": {
        "D38": "Online Shoppers",
        "D59": "Red Wine Quality",
        "D60": "White Wine Quality",
        "D88": "Airline Passenger Satisfaction",
        "D94": "Go Emotions Google",
        "D100": "Spotify East Asian",
        "D125": "Suicide Rates",
        "D182": "Obesity Levels",
        "D187": "Blood Transfusion",
        "D191": "Obesity Classification",
        "D192": "Gender Classification",
        "D194": "Happiness Classification",
        "D42": "Airline customer Holiday Booking dataset"
    },

    "ML Benchmarks & Synthetic": {
        "D34": "Spambase",
        "D85": "Synthetic Binary",
        "D89": "Naive Bayes Data",
        "D175": "Monk's Problems 1",
        "D176": "Monk's Problems 2",
        "D177": "Monk's Problems 3",
        "D178": "Monk's Problems 4",
        "D179": "Monk's Problems 5",
        "D180": "Monk's Problems 6"
    },

    "Other": {
        "D83": "Paris Housing",
        "D91": "Fake Bills",
        "D197": "Star Classification"
    }
}


def compute_metrics(datasets_list, selected_models, metric_for_comparison):
    """Compute metrics and stats for selected datasets and models"""
    
    # Handle "All models" selection
    if "All models" in selected_models:
        selected_models = ALL_MODELS
    
    # Filter the dataframe for selected datasets and models
    df = df_results[
        (df_results['dataset'].isin(datasets_list)) & 
        (df_results['model'].isin(selected_models))
    ].copy()
    
    if df.empty:
        return df, pd.DataFrame(), None
    
    # Statistical comparisons
    stat_records = []
    models_list = df['model'].unique().tolist()
    
    for i, m1 in enumerate(models_list):
        for m2 in models_list[i+1:]:
            m1_vals = df[df['model'] == m1].set_index('dataset')[metric_for_comparison]
            m2_vals = df[df['model'] == m2].set_index('dataset')[metric_for_comparison]
            
            combined = pd.concat([m1_vals, m2_vals], axis=1, keys=['m1', 'm2']).dropna()
            
            if len(combined) < 2:
                continue
            
            t_stat, p_val = ttest_rel(combined['m1'], combined['m2'])
            
            stat_records.append({
                "model1": m1,
                "model2": m2,
                "mean_diff": combined['m1'].mean() - combined['m2'].mean(),
                "t_stat": t_stat,
                "p_value": p_val,
                "significant": "Yes" if p_val < 0.05 else "No"
            })
    
    stat_df = pd.DataFrame(stat_records)
    
    # Create visualization
    fig = create_heatmap(df, metric_for_comparison)
    
    return df, stat_df, fig

def create_heatmap(df, metric):
    """Create metric by dataset heatmap"""
    
    # Create heatmap of metric by dataset and model
    pivot = df.pivot_table(values=metric, index='dataset', columns='model')
    
    fig, ax = plt.subplots(figsize=(12, max(8, len(pivot) * 0.4)))
    sns.heatmap(pivot, annot=True, fmt='.3f', cmap='viridis', ax=ax, 
                cbar_kws={'label': metric.capitalize()})
    ax.set_title(f'{metric.capitalize()} by Dataset and Model', 
                 fontsize=14, fontweight='bold')
    ax.set_xlabel('Model', fontsize=12)
    ax.set_ylabel('Dataset', fontsize=12)
    
    plt.tight_layout()
    return fig

def run_evaluation(selected_datasets, selected_models, metric_comparison):
    """Main evaluation function"""
    
    if not selected_datasets:
        empty = gr.update(value=None, visible=False)
        return "Please select datasets", empty, empty, empty, empty
    
    if not selected_models:
        selected_models = ["All models"]
    
    # Ensure metric_comparison is a list
    if isinstance(metric_comparison, str):
        metric_comparison = [metric_comparison]
    
    if not metric_comparison:
        empty = gr.update(value=None, visible=False)
        return "Please select at least one metric", empty, empty, empty, empty
    
    # Compute metrics once
    df, _, _ = compute_metrics(selected_datasets, selected_models, metric_comparison[0])
    
    if df.empty:
        empty = gr.update(value=None, visible=False)
        return "No results found", empty, empty, empty, empty
    
    # Create stats and figures for EACH selected metric
    all_stats_html = ""
    outputs = []
    
    for i, metric in enumerate(metric_comparison):
        if i >= 4:
            break
        
        _, stat_df, fig = compute_metrics(selected_datasets, selected_models, metric)
        
        if not stat_df.empty:
            stats_html = f"""
            <h3>Statistical Tests ({metric})</h3>
            <p>Paired t-tests comparing model performance (* = significant at p < 0.05)</p>
            {stat_df.to_html(index=False, float_format='%.4f')}
            <hr>
            """
            all_stats_html += stats_html
        
        outputs.append(gr.update(value=fig, visible=True))
    
    # Fill remaining slots with hidden empty plots
    while len(outputs) < 4:
        outputs.append(gr.update(value=None, visible=False))
    
    if not all_stats_html:
        all_stats_html = "<p>Not enough data for statistical comparisons</p>"
    
    return all_stats_html, outputs[0], outputs[1], outputs[2], outputs[3]

# Build Gradio Interface
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("""
    # 🎯 Model Evaluation Platform
    ### Compare model performance across different datasets
    """)
    
    selected_datasets = gr.State([])
    
    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("### 📊 Select Datasets")
            
            # Create dropdowns for each category
            dropdowns = []
            for category, datasets in DATASET_CATEGORIES.items():
                choices = [f"{did}: {name}" for did, name in datasets.items() 
                          if did in AVAILABLE_DATASETS]
                if choices:
                    dd = gr.Dropdown(
                        choices=choices,
                        label=f"{category} ({len(choices)})",
                        multiselect=True,
                        value=[]
                    )
                    dropdowns.append(dd)
        
        with gr.Column(scale=1):
            gr.Markdown("### ⚙️ Evaluation Settings")
            
            summary = gr.Markdown("**0 datasets selected**")
            
            model_input = gr.Dropdown(
                choices=["All models"] + ALL_MODELS,
                label="Models",
                value=["All models"],
                multiselect=True
            )
            
            metric_comparison = gr.Dropdown(
                choices=["accuracy", "precision", "recall", "f1_score"],
                label="Metrics to Compare",
                value=["accuracy"],
                multiselect=True
            )
            
            run_btn = gr.Button("🚀 Run Evaluation", variant="primary", size="lg")
    
    def update_selection(*dropdown_values):
        ids = []
        for vals in dropdown_values:
            if vals:
                ids.extend([v.split(":")[0] for v in vals])
        ids = sorted(list(set(ids)))
        
        if ids:
            summary_text = f"**✓ {len(ids)} dataset{'s' if len(ids) != 1 else ''} selected:** {', '.join(ids)}"
        else:
            summary_text = "**No datasets selected**"
        
        return summary_text, ids
    
    for dd in dropdowns:
        dd.change(update_selection, inputs=dropdowns, outputs=[summary, selected_datasets])
    
    gr.Markdown("---")
    gr.Markdown("## 📈 Evaluation Results")
    
    output_stats = gr.HTML(label="Statistical Tests")
    
    with gr.Column():
        heatmap_output_1 = gr.Plot(label="Heatmap 1", visible=True)
        heatmap_output_2 = gr.Plot(label="Heatmap 2", visible=False)
        heatmap_output_3 = gr.Plot(label="Heatmap 3", visible=False)
        heatmap_output_4 = gr.Plot(label="Heatmap 4", visible=False)
    
    run_btn.click(
        run_evaluation,
        inputs=[selected_datasets, model_input, metric_comparison],
        outputs=[output_stats, heatmap_output_1, heatmap_output_2, 
                heatmap_output_3, heatmap_output_4]
    )

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