Upload 78 files

.gitattributes

@@ -33,3 +33,33 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+data/Authorship.mat filter=lfs diff=lfs merge=lfs -text
+data/Dermatology.mat filter=lfs diff=lfs merge=lfs -text
+data/dna.mat filter=lfs diff=lfs merge=lfs -text
+data/Factors.mat filter=lfs diff=lfs merge=lfs -text
+data/madelon.mat filter=lfs diff=lfs merge=lfs -text
+data/Movement_libras.mat filter=lfs diff=lfs merge=lfs -text
+data/Musk1.mat filter=lfs diff=lfs merge=lfs -text
+data/results.db filter=lfs diff=lfs merge=lfs -text
+data/spambase.mat filter=lfs diff=lfs merge=lfs -text
+data/splice.mat filter=lfs diff=lfs merge=lfs -text
+data/Synthetic_control.mat filter=lfs diff=lfs merge=lfs -text
+data/Waveform.mat filter=lfs diff=lfs merge=lfs -text
+data/Wdbc.mat filter=lfs diff=lfs merge=lfs -text
+pdf/CFR.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/CIFE.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/CMIFS.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/CMIM.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/CSMDCCMR.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/CSMI.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/DCSF.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/DISR.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/DWFS.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/IWFS.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/JMI.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/JMIM.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/MIM.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/MRI.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/MRMD.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/MRMR.pdf filter=lfs diff=lfs merge=lfs -text
+pdf/UCRFS.pdf filter=lfs diff=lfs merge=lfs -text

Webapp/app.py

@@ -2,7 +2,8 @@ import os
 import sys
 import pickle
 import json
-from flask import Flask, jsonify, request, render_template
+import datetime
+from flask import Flask, jsonify, request, render_template, send_from_directory
 
 # Add project root to sys.path to import leaderboard
 sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
@@ -13,6 +14,7 @@ from leaderboard import rank_results
 PROJECT_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 RESULT_DIR = os.path.join(PROJECT_ROOT, "results")
 DATASET_DIR = os.path.join(PROJECT_ROOT, "datasets")
+PDF_DIR = os.path.join(PROJECT_ROOT, "pdf")
 
 os.makedirs(RESULT_DIR, exist_ok=True)
 os.makedirs(DATASET_DIR, exist_ok=True)
@@ -179,7 +181,26 @@ def get_results():
 # ===============================
 @app.route("/api/datasets")
 def api_datasets():
-    return jsonify(list_available_datasets())
+    try:
+        datasets = []
+        ds_names = list_available_datasets()
+        for name in ds_names:
+            # Get modification time of the result file
+            result_path = os.path.join(RESULT_DIR, f"{name}.json")
+            last_updated = "Unknown"
+            if os.path.exists(result_path):
+                mtime = os.path.getmtime(result_path)
+                last_updated = datetime.datetime.fromtimestamp(mtime).strftime('%Y-%m-%d')
+            
+            datasets.append({
+                "name": name,
+                "last_updated": last_updated
+            })
+        return jsonify(datasets)
+    except Exception as e:
+        import traceback
+        traceback.print_exc()
+        return jsonify({"error": str(e)}), 500
 
 
 @app.route("/api/global_stats")
@@ -256,6 +277,11 @@ def get_global_stats():
     return jsonify(final_list)
 
 
+@app.route("/pdfs/<path:filename>")
+def serve_pdf(filename):
+    return send_from_directory(PDF_DIR, filename)
+
+
 if __name__ == "__main__":
     port = int(os.environ.get("PORT", 7860))
     app.run(host="0.0.0.0", port=port, debug=False)

Webapp/templates/global.html

@@ -60,6 +60,24 @@
             background-color: #34495e;
         }
 
+        .description-box {
+            background-color: #e8f4fd;
+            border-left: 4px solid #3498db;
+            padding: 15px;
+            margin-bottom: 20px;
+            border-radius: 4px;
+        }
+
+        .description-box h3 {
+            margin-top: 0;
+            color: #2980b9;
+        }
+
+        .description-box p {
+            margin: 5px 0;
+            line-height: 1.5;
+        }
+
         .weights-control {
             background-color: #f1f1f1;
             padding: 15px;
@@ -89,16 +107,21 @@
         
         .slider-group label {
             font-weight: bold;
-            min-width: 80px;
+            min-width: 60px;
         }
 
         input[type="number"] {
-            width: 70px;
+            width: 80px;
             padding: 5px;
             border: 1px solid #ccc;
             border-radius: 4px;
         }
 
+        input[readonly] {
+            background-color: #e9ecef;
+            color: #666;
+        }
+
         button.recalc-btn {
             background-color: var(--primary-color);
             color: white;
@@ -162,11 +185,16 @@
             background-color: var(--primary-color);
         }
         
-        .warning-text {
-            color: #e74c3c;
-            font-size: 0.9em;
+        .time-detail {
+            font-size: 0.8em; 
+            color: #666;
+            margin-top: 2px;
+        }
+        
+        .version-tag {
+            font-size: 0.8em;
+            color: #7f8c8d;
             margin-top: 5px;
-            display: none;
         }
     </style>
 </head>
@@ -174,36 +202,46 @@
 
 <div class="container">
     <header>
-        <h1>🌍 Global Algorithm Rankings</h1>
+        <div>
+            <h1>🌍 Global Algorithm Rankings</h1>
+            <div id="last-updated" class="version-tag">Data Last Updated: Loading...</div>
+        </div>
         <a href="/" class="nav-link">← Back to Dataset View</a>
     </header>
 
+    <div class="description-box">
+        <h3>About Global Rankings</h3>
+        <p>
+            This page provides a comprehensive evaluation of feature selection algorithms across all available datasets.
+            Algorithms are ranked based on a weighted score combining <strong>Accuracy (F1)</strong>, <strong>Robustness (AUC)</strong>, and <strong>Efficiency (Time)</strong>.
+            You can adjust the importance of each factor below to customize the ranking criteria.
+        </p>
+    </div>
+
     <div class="weights-control">
-        <h3>🏆 Scoring Formula: S = a·F1 + b·AUC + c·TimeScore</h3>
+        <h3>🏆 Scoring Formula: S = α·F1 + β·AUC + γ·TimeScore</h3>
         <p style="font-size: 0.9em; color: #666; margin-bottom: 10px;">
-            Note: TimeScore is calculated as normalized efficiency (1 = fastest, 0 = slowest). 
-            <br>Constraint: a + b + c = 1.
+            Constraint: α + β + γ = 1. TimeScore is normalized (1 = fastest).
         </p>
         
         <div class="sliders-container">
             <div class="slider-group">
-                <label for="weight-a">a (F1):</label>
-                <input type="number" id="weight-a" value="0.4" step="0.1" min="0" max="1">
+                <label for="weight-a">F1 (α):</label>
+                <input type="number" id="weight-a" value="0.4" step="0.05" min="0" max="1">
             </div>
             
             <div class="slider-group">
-                <label for="weight-b">b (AUC):</label>
-                <input type="number" id="weight-b" value="0.4" step="0.1" min="0" max="1">
+                <label for="weight-b">AUC (β):</label>
+                <input type="number" id="weight-b" value="0.4" step="0.05" min="0" max="1">
             </div>
             
             <div class="slider-group">
-                <label for="weight-c">c (Time):</label>
-                <input type="number" id="weight-c" value="0.2" step="0.1" min="0" max="1">
+                <label for="weight-c">Time (γ):</label>
+                <input type="number" id="weight-c" value="0.2" readonly title="Auto-calculated: 1 - α - β">
             </div>
 
             <button class="recalc-btn" onclick="calculateAndRender()">Recalculate Rankings</button>
         </div>
-        <div id="weight-warning" class="warning-text">⚠️ Weights must sum to 1.0</div>
     </div>
 
     <div id="loading-indicator" style="text-align: center; color: #666;">Loading global stats...</div>
@@ -215,7 +253,7 @@
                 <th data-key="algorithm">Algorithm <span class="arrow">↕</span></th>
                 <th data-key="mean_f1_global">Global F1 <span class="arrow">↕</span></th>
                 <th data-key="mean_auc_global">Global AUC <span class="arrow">↕</span></th>
-                <th data-key="mean_time_global">Avg Time (s) <span class="arrow">↕</span></th>
+                <th data-key="mean_time_global">Efficiency (Time) <span class="arrow">↕</span></th>
                 <th data-key="final_score">Final Score <span class="arrow">↕</span></th>
             </tr>
         </thead>
@@ -235,7 +273,27 @@
     const weightA = document.getElementById("weight-a");
     const weightB = document.getElementById("weight-b");
     const weightC = document.getElementById("weight-c");
-    const warningText = document.getElementById("weight-warning");
+    const lastUpdatedDiv = document.getElementById("last-updated");
+
+    // Fetch datasets info to get latest date
+    function fetchLastUpdated() {
+        fetch("/api/datasets")
+            .then(res => res.json())
+            .then(data => {
+                if (data.length > 0) {
+                    // Sort by date to find latest? Or just take one? 
+                    // Usually we want the overall latest date.
+                    // Let's just pick the first one's date or find the max date if needed.
+                    // For simplicity, just showing "Latest"
+                    const dates = data.map(d => d.last_updated).filter(d => d !== 'Unknown').sort().reverse();
+                    if (dates.length > 0) {
+                        lastUpdatedDiv.textContent = `Data Last Updated: ${dates[0]}`;
+                    } else {
+                        lastUpdatedDiv.textContent = `Data Last Updated: Unknown`;
+                    }
+                }
+            });
+    }
 
     function fetchGlobalStats() {
         fetch("/api/global_stats")
@@ -251,20 +309,50 @@
             });
     }
 
+    // Weight auto-adjustment logic
+    function updateWeights(changedInput) {
+        let a = parseFloat(weightA.value) || 0;
+        let b = parseFloat(weightB.value) || 0;
+        
+        // Clamp inputs to 0-1
+        if (a < 0) a = 0; if (a > 1) a = 1;
+        if (b < 0) b = 0; if (b > 1) b = 1;
+
+        if (changedInput === 'a') {
+            // If a changes, we try to adjust c first (c = 1 - a - b)
+            // If 1 - a - b < 0, it means a + b > 1, so we must reduce b
+            let c = 1 - a - b;
+            if (c < 0) {
+                b = Math.max(0, 1 - a); // Reduce b
+                c = 0; // c becomes 0
+            }
+            // Update UI
+            weightA.value = parseFloat(a.toFixed(2));
+            weightB.value = parseFloat(b.toFixed(2));
+            weightC.value = parseFloat(c.toFixed(2));
+        } else if (changedInput === 'b') {
+            // If b changes, we try to adjust c first
+            // If 1 - a - b < 0, we must reduce a
+            let c = 1 - a - b;
+            if (c < 0) {
+                a = Math.max(0, 1 - b); // Reduce a
+                c = 0;
+            }
+            // Update UI
+            weightA.value = parseFloat(a.toFixed(2));
+            weightB.value = parseFloat(b.toFixed(2));
+            weightC.value = parseFloat(c.toFixed(2));
+        }
+    }
+
+    weightA.addEventListener('input', () => updateWeights('a'));
+    weightB.addEventListener('input', () => updateWeights('b'));
+
     function calculateAndRender() {
         const a = parseFloat(weightA.value) || 0;
         const b = parseFloat(weightB.value) || 0;
         const c = parseFloat(weightC.value) || 0;
 
-        // Validation
-        const sum = a + b + c;
-        if (Math.abs(sum - 1.0) > 0.01) {
-            warningText.style.display = 'block';
-            warningText.textContent = `⚠️ Weights sum to ${sum.toFixed(2)}. They should sum to 1.0.`;
-        } else {
-            warningText.style.display = 'none';
-        }
-
         // Find min/max time for normalization
         let minTime = Infinity;
         let maxTime = -Infinity;
@@ -273,14 +361,11 @@
             if (d.mean_time_global > maxTime) maxTime = d.mean_time_global;
         });
         
-        // Prevent division by zero if all times are same
         const timeRange = maxTime - minTime;
 
         // Process data
         processedData = rawData.map(d => {
             // Time Score: 1 if fast, 0 if slow
-            // Formula: 1 - (time - min) / (max - min)
-            // If range is 0, score is 1 (all same speed)
             let timeScore = 1.0;
             if (timeRange > 0.0001) {
                 timeScore = 1.0 - ((d.mean_time_global - minTime) / timeRange);
@@ -338,8 +423,8 @@
                 <td>${safeFixed(row.mean_f1_global)}</td>
                 <td>${safeFixed(row.mean_auc_global)}</td>
                 <td>
-                    ${safeFixed(row.mean_time_global, 2)}s
-                    <div style="font-size:0.7em; color:#888;">Score: ${safeFixed(row.time_score_norm, 2)}</div>
+                    <strong>${safeFixed(row.time_score_norm)}</strong>
+                    <div class="time-detail">${safeFixed(row.mean_time_global, 2)}s</div>
                 </td>
                 <td>
                     <strong>${safeFixed(row.final_score)}</strong>
@@ -358,7 +443,6 @@
          if (activeHeader) activeHeader.textContent = sortDirection === 1 ? '↑' : '↓';
     }
 
-    // Sort handlers
     document.querySelectorAll('th[data-key]').forEach(th => {
         th.addEventListener('click', () => {
             const key = th.dataset.key;
@@ -368,19 +452,19 @@
                 sortKey = key;
                 sortDirection = (key === 'rank' || key === 'mean_time_global') ? 1 : -1;
             }
-            // For rank, it's just index, but let's assume we sort by score desc if rank is clicked
             if (key === 'rank') {
                 sortKey = 'final_score';
                 sortDirection = -1;
             }
-            
             sortData();
             renderTable();
         });
     });
 
-    // Initial load
-    document.addEventListener("DOMContentLoaded", fetchGlobalStats);
+    document.addEventListener("DOMContentLoaded", () => {
+        fetchLastUpdated();
+        fetchGlobalStats();
+    });
 
 </script>
 

Webapp/templates/index.html

@@ -13,6 +13,7 @@
             --text-color: #333;
             --border-color: #dee2e6;
             --hover-color: #f1f1f1;
+            --accent-color: #e67e22;
         }
 
         body {
@@ -59,6 +60,76 @@
             font-size: 14px;
         }
 
+        /* Info Boxes */
+        .info-section {
+            display: flex;
+            gap: 20px;
+            margin-bottom: 20px;
+            flex-wrap: wrap;
+        }
+
+        .description-box, .metadata-box {
+            flex: 1;
+            background-color: #e8f4fd;
+            border-left: 4px solid #3498db;
+            padding: 15px;
+            border-radius: 4px;
+            min-width: 300px;
+        }
+
+        .metadata-box {
+            background-color: #fef9e7;
+            border-left-color: #f1c40f;
+        }
+
+        h3 {
+            margin-top: 0;
+            margin-bottom: 10px;
+            font-size: 1.1em;
+            color: var(--secondary-color);
+        }
+
+        p {
+            margin: 5px 0;
+            line-height: 1.5;
+            font-size: 0.95em;
+        }
+
+        .version-tag {
+            font-size: 0.8em;
+            color: #7f8c8d;
+            margin-top: 5px;
+        }
+
+        /* Filters */
+        .filters-box {
+            background-color: #f1f1f1;
+            padding: 15px;
+            border-radius: 8px;
+            margin-bottom: 20px;
+            border: 1px solid #ddd;
+            display: flex;
+            gap: 20px;
+            align-items: center;
+            flex-wrap: wrap;
+        }
+
+        .filter-group {
+            display: flex;
+            align-items: center;
+            gap: 10px;
+        }
+
+        input[type="range"] {
+            width: 120px;
+        }
+
+        .filter-val {
+            font-weight: bold;
+            min-width: 40px;
+        }
+
+        /* Table */
         table {
             width: 100%;
             border-collapse: collapse;
@@ -122,6 +193,25 @@
             color: var(--primary-color);
         }
 
+        /* Charts */
+        .charts-section {
+            margin-top: 30px;
+            display: flex;
+            flex-direction: column;
+            gap: 20px;
+        }
+
+        .chart-container {
+            background: white;
+            padding: 15px;
+            border-radius: 8px;
+            box-shadow: 0 2px 4px rgba(0,0,0,0.1);
+            border: 1px solid #eee;
+            position: relative;
+            height: 400px;
+            width: 100%;
+        }
+
         /* Modal styles */
         .modal {
             display: none;
@@ -171,13 +261,64 @@
             padding: 20px;
             color: #666;
         }
+
+        /* Sidebar for PDF */
+        .pdf-sidebar {
+            position: fixed;
+            top: 0;
+            right: -50%; /* Hidden by default */
+            width: 50%;
+            height: 100%;
+            background: white;
+            box-shadow: -2px 0 5px rgba(0,0,0,0.2);
+            z-index: 2000;
+            transition: right 0.3s ease-in-out;
+            display: flex;
+            flex-direction: column;
+        }
+
+        .pdf-sidebar.open {
+            right: 0;
+        }
+
+        .sidebar-header {
+            padding: 10px 20px;
+            background: var(--primary-color);
+            color: white;
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+        }
+
+        .sidebar-content {
+            flex: 1;
+            padding: 0;
+        }
+
+        .sidebar-content iframe {
+            width: 100%;
+            height: 100%;
+            border: none;
+        }
+
+        .algo-link {
+            color: var(--primary-color);
+            cursor: pointer;
+            font-weight: bold;
+        }
+        .algo-link:hover {
+            text-decoration: underline;
+        }
     </style>
 </head>
 <body>
 
 <div class="container">
     <header>
-        <h1>🏆 AutoFS Leaderboard</h1>
+        <div>
+            <h1>🏆 AutoFS Leaderboard</h1>
+            <div id="last-updated" class="version-tag">Data Last Updated: Loading...</div>
+        </div>
         <div style="display:flex; gap:15px; align-items:center;">
             <a href="/global" style="text-decoration:none; color:white; background-color:#8e44ad; padding:8px 15px; border-radius:4px; font-size:0.9em;">🌍 Global Rankings</a>
             <div class="controls">
@@ -189,29 +330,51 @@
         </div>
     </header>
 
-    <div id="loading-indicator" class="loading" style="display: none;">Loading data...</div>
+    <div class="info-section">
+        <div class="description-box">
+            <h3>About This Dataset</h3>
+            <p>
+                This dashboard displays the performance of various feature selection algorithms on the 
+                <strong><span id="desc-dataset-name">Selected</span></strong> dataset. 
+                Compare algorithms based on accuracy (F1), stability (AUC), and computational efficiency.
+            </p>
+        </div>
+        <div class="metadata-box">
+            <h3>Dataset Metadata</h3>
+            <p><strong>Name:</strong> <span id="meta-name">-</span></p>
+            <p><strong>Last Updated:</strong> <span id="meta-updated">-</span></p>
+            <!-- Placeholder for future metadata -->
+            <p style="color:#888; font-size:0.8em;">(Additional metadata like samples/features not available)</p>
+        </div>
+    </div>
 
-    <div class="chart-controls" style="text-align:center; margin-top: 20px; margin-bottom: 15px;">
-        <label style="margin-right:15px; font-weight:bold;">View Mode:</label>
-        <input type="radio" id="view-overall" name="chart-view" value="overall" checked onchange="updateView()">
-        <label for="view-overall" style="margin-right:10px;">Overall (Mean)</label>
-        
-        <input type="radio" id="view-classifiers-f1" name="chart-view" value="classifiers-f1" onchange="updateView()">
-        <label for="view-classifiers-f1" style="margin-right:10px;">F1 by Classifier</label>
+    <div class="filters-box">
+        <h3>🔍 Filters</h3>
         
-        <input type="radio" id="view-classifiers-auc" name="chart-view" value="classifiers-auc" onchange="updateView()">
-        <label for="view-classifiers-auc">AUC by Classifier</label>
-    </div>
+        <div class="filter-group">
+            <label>Min F1 Score:</label>
+            <input type="range" id="filter-f1" min="0" max="1" step="0.05" value="0">
+            <span id="val-f1" class="filter-val">0.00</span>
+        </div>
 
-    <div class="charts-container" style="display: flex; gap: 20px; margin-bottom: 20px;">
-        <div style="flex: 1; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);">
-            <canvas id="scoreChart"></canvas>
+        <div class="filter-group">
+            <label>Max Time (s):</label>
+            <input type="range" id="filter-time" min="1" max="500" step="10" value="500">
+            <span id="val-time" class="filter-val">500+</span>
         </div>
-        <div style="flex: 1; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);">
-            <canvas id="timeChart"></canvas>
+
+        <div style="margin-left: auto;">
+            <label style="margin-right:10px; font-weight:bold;">Chart View:</label>
+            <select id="chart-view-mode" onchange="updateView()">
+                <option value="overall">Overall (Mean)</option>
+                <option value="classifiers-f1">F1 by Classifier</option>
+                <option value="classifiers-auc">AUC by Classifier</option>
+            </select>
         </div>
     </div>
 
+    <div id="loading-indicator" class="loading" style="display: none;">Loading data...</div>
+
     <table id="result-table">
         <thead>
             <!-- Headers generated dynamically -->
@@ -220,6 +383,19 @@
             <!-- Data rows will be populated here -->
         </tbody>
     </table>
+
+    <div class="charts-section">
+        <div class="chart-container">
+            <h3>📊 Performance Comparison</h3>
+            <canvas id="scoreChart"></canvas>
+        </div>
+        
+        <div class="chart-container">
+            <h3>📉 Pareto Frontier (Trade-off)</h3>
+            <p style="font-size:0.9em; color:#666; margin-top:-10px;">X: Number of Selected Features (Lower is better) vs Y: F1 Score (Higher is better). Optimal: Top-Left.</p>
+            <canvas id="paretoChart"></canvas>
+        </div>
+    </div>
 </div>
 
 <!-- Modal for details -->
@@ -231,11 +407,30 @@
     </div>
 </div>
 
+<!-- PDF Sidebar -->
+<div id="pdf-sidebar" class="pdf-sidebar">
+    <div class="sidebar-header">
+        <h3 id="sidebar-title" style="margin:0; color:white;">Paper Preview</h3>
+        <span class="close" onclick="closeSidebar()" style="color:white; opacity:0.8; font-size: 28px; cursor: pointer;">&times;</span>
+    </div>
+    <div class="sidebar-content">
+        <iframe id="pdf-frame" src=""></iframe>
+    </div>
+</div>
+
 <script>
     let currentResults = [];
+    let filteredResults = [];
+    let allDatasets = [];
     let sortDirection = 1; // 1 for asc, -1 for desc
     let lastSortKey = '';
 
+    // Filter Elements
+    const filterF1 = document.getElementById('filter-f1');
+    const filterTime = document.getElementById('filter-time');
+    const valF1 = document.getElementById('val-f1');
+    const valTime = document.getElementById('val-time');
+
     const VIEW_CONFIG = {
         'overall': [
             { key: 'mean_f1', label: 'Mean F1' },
@@ -259,6 +454,12 @@
     const loadingIndicator = document.getElementById("loading-indicator");
     const modal = document.getElementById("details-modal");
     const closeModal = document.querySelector(".close");
+    
+    // Metadata elements
+    const metaName = document.getElementById('meta-name');
+    const metaUpdated = document.getElementById('meta-updated');
+    const descName = document.getElementById('desc-dataset-name');
+    const globalUpdated = document.getElementById('last-updated');
 
     // Close modal
     closeModal.onclick = () => modal.style.display = "none";
@@ -266,217 +467,185 @@
         if (event.target == modal) modal.style.display = "none";
     }
 
-    // Global chart instances
+    // Chart instances
     let scoreChartInstance = null;
-    let timeChartInstance = null;
+    let paretoChartInstance = null;
 
-    function updateCharts(results) {
-        if (!Array.isArray(results) || results.length === 0) return;
+    // Filter Logic
+    function applyFilters() {
+        const minF1 = parseFloat(filterF1.value);
+        const maxTime = parseFloat(filterTime.value);
+        
+        valF1.textContent = minF1.toFixed(2);
+        valTime.textContent = maxTime >= 500 ? "500+" : maxTime + "s";
+
+        filteredResults = currentResults.filter(r => {
+            const f1 = r.mean_f1 || 0;
+            const time = r.time || 0;
+            return f1 >= minF1 && (maxTime >= 500 || time <= maxTime);
+        });
 
-        // Limit to top 15 for readability
-        const topResults = results.slice(0, 15);
+        renderTable(filteredResults);
+        updateCharts(filteredResults);
+    }
+
+    filterF1.addEventListener('input', applyFilters);
+    filterTime.addEventListener('input', applyFilters);
+
+    function updateCharts(results) {
+        if (!Array.isArray(results)) return;
+        
+        // Use filtered results for charts too
+        // Limit to top 20 for bar chart readability
+        const topResults = results.slice(0, 20); 
         const labels = topResults.map(r => r.algorithm || 'Unknown');
-        const times = topResults.map(r => r.time || 0);
 
-        const viewMode = document.querySelector('input[name="chart-view"]:checked').value;
+        const viewMode = document.getElementById('chart-view-mode').value;
         let datasets = [];
 
         if (viewMode === 'overall') {
-            const f1Scores = topResults.map(r => r.mean_f1 || 0);
-            const aucScores = topResults.map(r => r.mean_auc || 0);
             datasets = [
                 {
                     label: 'Mean F1',
-                    data: f1Scores,
+                    data: topResults.map(r => r.mean_f1 || 0),
                     backgroundColor: 'rgba(52, 152, 219, 0.7)',
                     borderColor: 'rgba(52, 152, 219, 1)',
                     borderWidth: 1
                 },
                 {
                     label: 'Mean AUC',
-                    data: aucScores,
+                    data: topResults.map(r => r.mean_auc || 0),
                     backgroundColor: 'rgba(46, 204, 113, 0.7)',
                     borderColor: 'rgba(46, 204, 113, 1)',
                     borderWidth: 1
                 }
             ];
         } else if (viewMode === 'classifiers-f1') {
-            const classifiers = ['nb', 'svm', 'rf'];
-            const colors = ['rgba(255, 206, 86, 0.5)', 'rgba(75, 192, 192, 0.5)', 'rgba(153, 102, 255, 0.5)'];
-            const borderColors = ['rgba(255, 206, 86, 1)', 'rgba(75, 192, 192, 1)', 'rgba(153, 102, 255, 1)'];
-            
-            datasets = classifiers.map((cls, idx) => ({
-                label: cls.toUpperCase() + ' F1',
-                data: topResults.map(r => (r.metrics && r.metrics[cls]) ? r.metrics[cls].f1 : 0),
-                backgroundColor: colors[idx],
-                borderColor: borderColors[idx],
+            datasets = ['nb', 'svm', 'rf'].map((clf, i) => ({
+                label: clf.toUpperCase() + ' F1',
+                data: topResults.map(r => r.metrics?.[clf]?.f1 || 0),
+                backgroundColor: `hsla(${200 + i*40}, 70%, 60%, 0.7)`,
+                borderColor: `hsla(${200 + i*40}, 70%, 60%, 1)`,
                 borderWidth: 1
             }));
-        } else if (viewMode === 'classifiers-auc') {
-            const classifiers = ['nb', 'svm', 'rf'];
-            const colors = ['rgba(255, 206, 86, 0.5)', 'rgba(75, 192, 192, 0.5)', 'rgba(153, 102, 255, 0.5)'];
-            const borderColors = ['rgba(255, 206, 86, 1)', 'rgba(75, 192, 192, 1)', 'rgba(153, 102, 255, 1)'];
-            
-            datasets = classifiers.map((cls, idx) => ({
-                label: cls.toUpperCase() + ' AUC',
-                data: topResults.map(r => (r.metrics && r.metrics[cls]) ? r.metrics[cls].auc : 0),
-                backgroundColor: colors[idx],
-                borderColor: borderColors[idx],
+        } else {
+            datasets = ['nb', 'svm', 'rf'].map((clf, i) => ({
+                label: clf.toUpperCase() + ' AUC',
+                data: topResults.map(r => r.metrics?.[clf]?.auc || 0),
+                backgroundColor: `hsla(${30 + i*40}, 70%, 60%, 0.7)`,
+                borderColor: `hsla(${30 + i*40}, 70%, 60%, 1)`,
                 borderWidth: 1
             }));
         }
 
-        // Score Chart
-        const scoreCtx = document.getElementById('scoreChart').getContext('2d');
+        // 1. Performance Chart (Horizontal Bar)
+        const ctxScore = document.getElementById('scoreChart').getContext('2d');
         if (scoreChartInstance) scoreChartInstance.destroy();
         
-        scoreChartInstance = new Chart(scoreCtx, {
+        scoreChartInstance = new Chart(ctxScore, {
             type: 'bar',
-            data: {
-                labels: labels,
-                datasets: datasets
-            },
+            data: { labels: labels, datasets: datasets },
             options: {
+                indexAxis: 'y', // Horizontal
                 responsive: true,
                 maintainAspectRatio: false,
-                plugins: {
-                    title: {
-                        display: true,
-                        text: viewMode === 'overall' ? 'Top Algorithms Performance (Mean)' : 
-                              (viewMode === 'classifiers-f1' ? 'F1-Score by Classifier' : 'AUC by Classifier')
-                    }
-                },
                 scales: {
-                    y: {
-                        beginAtZero: false,
-                        // min: 0.8
-                    }
+                    x: { beginAtZero: true, max: 1.0 },
+                    y: { ticks: { autoSkip: false } }
                 }
             }
         });
 
-        // Time Chart
-        const timeCtx = document.getElementById('timeChart').getContext('2d');
-        if (timeChartInstance) timeChartInstance.destroy();
+        // 2. Pareto Frontier Chart (Scatter)
+        // X: Num Selected Features, Y: Mean F1
+        const paretoData = results.map(r => ({
+            x: r.num_features || (r.selected_features ? r.selected_features.length : 0),
+            y: r.mean_f1 || 0,
+            algorithm: r.algorithm
+        }));
 
-        timeChartInstance = new Chart(timeCtx, {
-            type: 'line',
+        const ctxPareto = document.getElementById('paretoChart').getContext('2d');
+        if (paretoChartInstance) paretoChartInstance.destroy();
+
+        paretoChartInstance = new Chart(ctxPareto, {
+            type: 'scatter',
             data: {
-                labels: labels,
                 datasets: [{
-                    label: 'Time (s)',
-                    data: times,
-                    backgroundColor: 'rgba(231, 76, 60, 0.2)',
-                    borderColor: 'rgba(231, 76, 60, 1)',
-                    borderWidth: 2,
-                    tension: 0.3,
-                    fill: true
+                    label: 'Algorithm Performance',
+                    data: paretoData,
+                    backgroundColor: 'rgba(230, 126, 34, 0.7)', // Orange accent
+                    borderColor: 'rgba(230, 126, 34, 1)',
+                    pointRadius: 6,
+                    pointHoverRadius: 8
                 }]
             },
             options: {
                 responsive: true,
                 maintainAspectRatio: false,
-                plugins: {
-                    title: { display: true, text: 'Execution Time' }
-                },
                 scales: {
-                    y: { beginAtZero: true }
+                    x: { 
+                        type: 'linear', 
+                        position: 'bottom',
+                        title: { display: true, text: 'Number of Selected Features' }
+                    },
+                    y: { 
+                        title: { display: true, text: 'Mean F1 Score' },
+                        min: 0, max: 1
+                    }
+                },
+                plugins: {
+                    tooltip: {
+                        callbacks: {
+                            label: function(context) {
+                                const pt = context.raw;
+                                return `${pt.algorithm}: F1=${pt.y.toFixed(4)}, Feats=${pt.x}`;
+                            }
+                        }
+                    }
                 }
             }
         });
     }
 
-    function showDetails(result) {
-        const title = document.getElementById("modal-title");
-        const body = document.getElementById("modal-body");
-        
-        title.textContent = `${result.algorithm} Details`;
-        
-        let featuresHtml = result.selected_features.map(f => 
-            `<span class="feature-tag">${f}</span>`
-        ).join('');
-
-        let metricsHtml = '<div style="margin-top: 15px;"><h3>Metrics Breakdown</h3>';
-        for (const [clf, m] of Object.entries(result.metrics || {})) {
-            metricsHtml += `
-                <div style="margin-bottom: 10px;">
-                    <strong>${clf.toUpperCase()}:</strong> 
-                    F1: ${m.f1.toFixed(4)}, AUC: ${m.auc.toFixed(4)}
-                </div>`;
-        }
-        metricsHtml += '</div>';
-
-        body.innerHTML = `
-            <p><strong>Time:</strong> ${result.time.toFixed(4)}s</p>
-            <p><strong>Num Features:</strong> ${result.num_features}</p>
-            <p><strong>Selected Features (${result.selected_features.length}):</strong></p>
-            <div>${featuresHtml}</div>
-            ${metricsHtml}
-        `;
-        
-        modal.style.display = "block";
-    }
-
-    function getValue(obj, path) {
-        if (!path) return undefined;
-        return path.split('.').reduce((acc, part) => (acc && acc[part] !== undefined) ? acc[part] : undefined, obj);
-    }
-
-    function safeFixed(value, digits=4) {
-        if (value === undefined || value === null) return 'N/A';
-        return Number(value).toFixed(digits);
+    function updateView() {
+        renderTableHeader();
+        renderTable(filteredResults);
+        updateCharts(filteredResults);
     }
 
     function renderTableHeader() {
-        const viewMode = document.querySelector('input[name="chart-view"]:checked').value;
-        const dynamicCols = VIEW_CONFIG[viewMode] || VIEW_CONFIG['overall'];
-
-        let html = '<tr>';
-        html += '<th data-key="rank" style="width: 60px;">#</th>';
-        html += '<th data-key="algorithm">Algorithm <span class="arrow">↕</span></th>';
+        const viewMode = document.getElementById('chart-view-mode').value;
+        const config = VIEW_CONFIG[viewMode];
         
-        dynamicCols.forEach(col => {
-            html += `<th data-key="${col.key}">${col.label} <span class="arrow">↕</span></th>`;
-        });
-
-        html += '<th data-key="time">Time (s) <span class="arrow">↕</span></th>';
-        html += '<th data-key="selected_features">Selected Features</th>';
-        html += '</tr>';
-
-        tableHead.innerHTML = html;
-
-        // Re-attach sort listeners
-        tableHead.querySelectorAll('th[data-key]').forEach(th => {
-            th.addEventListener('click', () => sortTable(th.dataset.key));
+        let headerHTML = `
+            <tr>
+                <th>Rank</th>
+                <th onclick="sortTable('algorithm')">Algorithm <span class="arrow"></span></th>
+        `;
+        
+        config.forEach(col => {
+            headerHTML += `<th onclick="sortTable('${col.key}')">${col.label} <span class="arrow"></span></th>`;
         });
+        
+        headerHTML += `
+                <th onclick="sortTable('time')">Time (s) <span class="arrow"></span></th>
+                <th onclick="sortTable('selected_features')">Selected Features <span class="arrow"></span></th>
+            </tr>
+        `;
+        
+        tableHead.innerHTML = headerHTML;
     }
 
-    function updateTable(results) {
+    function renderTable(results) {
         tableBody.innerHTML = "";
         
         // Robust data handling
-        if (!results) {
-            results = [];
-        } else if (!Array.isArray(results)) {
-            // Try to handle wrapped data or single object
-            if (results.data && Array.isArray(results.data)) {
-                results = results.data;
-            } else if (results.results && Array.isArray(results.results)) {
-                results = results.results;
-            } else if (typeof results === 'object') {
-                // Assume it's a single record? Or convert object values to array?
-                // For now, wrap in array if it looks like a record (has algorithm)
-                if (results.algorithm) {
-                    results = [results];
-                } else {
-                    console.error("Invalid data format:", results);
-                    tableBody.innerHTML = '<tr><td colspan="10" style="text-align:center; color:red;">Error: Invalid data format. Check console for details.</td></tr>';
-                    return;
-                }
-            } else {
-                tableBody.innerHTML = '<tr><td colspan="10" style="text-align:center; color:red;">Error: Invalid data format</td></tr>';
-                return;
-            }
+        if (!results) { results = []; }
+        else if (!Array.isArray(results)) {
+            if (results.data && Array.isArray(results.data)) results = results.data;
+            else if (results.results && Array.isArray(results.results)) results = results.results;
+            else if (results.algorithm) results = [results];
+            else results = [];
         }
 
         if (results.length === 0) {
@@ -484,39 +653,47 @@
             return;
         }
 
-        const viewMode = document.querySelector('input[name="chart-view"]:checked').value;
-        const dynamicCols = VIEW_CONFIG[viewMode] || VIEW_CONFIG['overall'];
+        const viewMode = document.getElementById('chart-view-mode').value;
+        const config = VIEW_CONFIG[viewMode];
 
-        results.forEach((r, idx) => {
-            const row = document.createElement("tr");
-            
-            // Format features for preview
-            const featurePreview = (r.selected_features && Array.isArray(r.selected_features)) 
-                ? r.selected_features.slice(0, 5).join(', ') + (r.selected_features.length > 5 ? '...' : '')
-                : 'N/A';
+        results.forEach((row, index) => {
+            const tr = document.createElement("tr");
             
-            let html = `<td>${idx + 1}</td>`;
-            html += `<td><strong>${r.algorithm || 'Unknown'}</strong></td>`;
-
-            dynamicCols.forEach(col => {
-                const val = getValue(r, col.key);
-                const score = val !== undefined ? val : 0;
-                html += `
-                    <td>
-                        ${safeFixed(val)}
-                        <div class="score-bar"><div class="score-fill" style="width: ${Math.min(score * 100, 100)}%"></div></div>
-                    </td>`;
+            // Helper to get nested property safely
+            const getVal = (obj, path) => {
+                return path.split('.').reduce((acc, part) => acc && acc[part], obj);
+            };
+
+            let metricsHTML = '';
+            config.forEach(col => {
+                const val = getVal(row, col.key);
+                const numVal = (val !== undefined && val !== null) ? Number(val).toFixed(4) : 'N/A';
+                metricsHTML += `<td>${numVal}</td>`;
             });
-
-            const time = r.time || 0;
-            html += `<td>${safeFixed(time, 2)}</td>`;
-            html += `
-                <td class="features-cell" onclick="showDetails(currentResults[${idx}])" title="Click for details">
-                    ${featurePreview} <span style="font-size:0.8em; color:#999;">(Click for details)</span>
-                </td>`;
             
-            row.innerHTML = html;
-            tableBody.appendChild(row);
+            // Features
+            let featCount = row.num_features;
+            if (featCount === undefined && row.selected_features) featCount = row.selected_features.length;
+            
+            let featText = "";
+            if (Array.isArray(row.selected_features)) {
+                featText = row.selected_features.join(", ");
+            } else {
+                featText = "N/A";
+            }
+
+            const rank = index + 1;
+
+            tr.innerHTML = `
+                <td>${rank}</td>
+                <td class="algo-link" onclick="openPdf('${row.algorithm}')" title="Click to view paper">${row.algorithm || 'Unknown'}</td>
+                ${metricsHTML}
+                <td>${row.time ? Number(row.time).toFixed(4) : 'N/A'}</td>
+                <td class="features-cell" onclick="showDetails('${row.algorithm}', '${featText}')" title="${featText}">
+                    ${featText}
+                </td>
+            `;
+            tableBody.appendChild(tr);
         });
     }
 
@@ -524,21 +701,22 @@
         if (lastSortKey === key) {
             sortDirection *= -1;
         } else {
-            sortDirection = key === 'time' || key === 'rank' ? 1 : -1;
+            sortDirection = 1;
             lastSortKey = key;
         }
 
-        // We don't call renderTableHeader here because it resets the sort indicators if we rebuild entirely.
-        // Instead, we just update the arrows.
-        document.querySelectorAll('th .arrow').forEach(span => span.textContent = '↕');
-        const activeHeader = document.querySelector(`th[data-key="${key}"] .arrow`);
-        if (activeHeader) activeHeader.textContent = sortDirection === 1 ? '↑' : '↓';
+        // Helper to get nested value
+        const getVal = (obj, path) => path.split('.').reduce((acc, part) => acc && acc[part], obj);
 
-        const sorted = [...currentResults].sort((a, b) => {
-            let valA = getValue(a, key);
-            let valB = getValue(b, key);
+        filteredResults.sort((a, b) => {
+            let valA = getVal(a, key);
+            let valB = getVal(b, key);
             
-            if (key === 'rank') return 0;
+            // Handle array length for selected_features sort
+            if (key === 'selected_features') {
+                valA = Array.isArray(valA) ? valA.length : 0;
+                valB = Array.isArray(valB) ? valB.length : 0;
+            }
 
             if (valA === undefined) valA = -Infinity;
             if (valB === undefined) valB = -Infinity;
@@ -547,79 +725,127 @@
             if (valA > valB) return 1 * sortDirection;
             return 0;
         });
-        
-        // Don't update currentResults global if it breaks things, but here it's fine.
-        // Actually, let's keep currentResults as the master list? 
-        // No, currentResults should be the sorted list for consistent subsequent sorts.
-        currentResults = sorted; 
-        updateTable(sorted);
-    }
 
-    function updateView() {
-        renderTableHeader();
-        updateTable(currentResults);
-        updateCharts(currentResults);
+        renderTable(filteredResults);
+        updateSortArrows(key);
     }
-
-    function fetchResults(dataset) {
-        loadingIndicator.style.display = 'block';
-        tableBody.innerHTML = '';
-        
-        console.log("Fetching results for:", dataset);
-        fetch(`/api/results?dataset=${dataset}`)
-            .then(res => {
-                if (!res.ok) throw new Error("Network response was not ok");
-                return res.json();
-            })
-            .then(data => {
-                console.log("Data received:", data);
-                currentResults = data;
-                updateView();
-                loadingIndicator.style.display = 'none';
-            })
-            .catch(err => {
-                console.error("Error fetching results:", err);
-                loadingIndicator.textContent = "Error loading data. Make sure the server is running.";
-            });
+    
+    function updateSortArrows(activeKey) {
+        document.querySelectorAll('th .arrow').forEach(span => span.textContent = '↕');
+        // Find the th with onclick containing this key
+        const ths = document.querySelectorAll('th');
+        ths.forEach(th => {
+            if (th.getAttribute('onclick').includes(`'${activeKey}'`)) {
+                th.querySelector('.arrow').textContent = sortDirection === 1 ? '↑' : '↓';
+            }
+        });
     }
 
-    // Initialize
-    document.addEventListener("DOMContentLoaded", () => {
-        // Setup sort listeners
-        document.querySelectorAll('th[data-key]').forEach(th => {
-            th.addEventListener('click', () => sortTable(th.dataset.key));
-        });
+    function showDetails(algo, features) {
+        document.getElementById("modal-title").innerText = `${algo} - Selected Features`;
+        const featArray = features.split(", ");
+        const html = featArray.map(f => `<span class="feature-tag">${f}</span>`).join(" ");
+        document.getElementById("modal-body").innerHTML = `
+            <p><strong>Total Selected:</strong> ${featArray.length}</p>
+            <div style="margin-top:10px; line-height:1.6;">${html}</div>
+        `;
+        modal.style.display = "block";
+    }
 
-        // Load datasets
+    function fetchDatasets() {
         fetch("/api/datasets")
             .then(res => res.json())
-            .then(datasets => {
+            .then(data => {
+                allDatasets = data;
                 datasetSelect.innerHTML = "";
-                datasets.forEach(ds => {
+                
+                // Sort dates for global updated
+                const dates = data.map(d => d.last_updated).filter(d => d !== 'Unknown').sort().reverse();
+                if (dates.length > 0) {
+                    globalUpdated.textContent = `Data Last Updated: ${dates[0]}`;
+                } else {
+                    globalUpdated.textContent = `Data Last Updated: Unknown`;
+                }
+
+                data.forEach(ds => {
                     const option = document.createElement("option");
-                    option.value = ds;
-                    option.textContent = ds;
+                    option.value = ds.name;
+                    option.textContent = ds.name;
                     datasetSelect.appendChild(option);
                 });
 
-                if (datasets.includes("Authorship")) {
-                    datasetSelect.value = "Authorship";
-                    fetchResults("Authorship");
-                } else if (datasets.length > 0) {
-                    datasetSelect.value = datasets[0];
-                    fetchResults(datasets[0]);
+                // Default selection
+                if (data.length > 0) {
+                    loadDataset(data[0].name);
                 }
             })
             .catch(err => {
-                console.error("Error fetching datasets:", err);
-                datasetSelect.innerHTML = "<option>Error loading datasets</option>";
+                console.error("Error loading datasets:", err);
+                datasetSelect.innerHTML = '<option disabled>Error loading</option>';
             });
+    }
 
-        datasetSelect.addEventListener('change', (e) => {
-            fetchResults(e.target.value);
-        });
+    function loadDataset(name) {
+        datasetSelect.value = name;
+        loadingIndicator.style.display = "block";
+        tableBody.innerHTML = "";
+        
+        // Update metadata box
+        const dsInfo = allDatasets.find(d => d.name === name);
+        if (dsInfo) {
+            metaName.textContent = dsInfo.name;
+            metaUpdated.textContent = dsInfo.last_updated;
+            descName.textContent = dsInfo.name;
+        }
+
+        fetch(`/api/results?dataset=${name}`)
+            .then(res => res.json())
+            .then(data => {
+                loadingIndicator.style.display = "none";
+                currentResults = data;
+                
+                // Reset filters on new dataset? Or keep them? 
+                // Let's reset to show all data first, or apply current?
+                // Applying current is better UX
+                applyFilters();
+                renderTableHeader(); // Ensure headers match view mode
+            })
+            .catch(err => {
+                loadingIndicator.style.display = "none";
+                console.error("Error:", err);
+                tableBody.innerHTML = '<tr><td colspan="10" style="color:red; text-align:center;">Error loading results</td></tr>';
+            });
+    }
+
+    datasetSelect.addEventListener("change", (e) => {
+        loadDataset(e.target.value);
     });
+
+    // PDF Sidebar Logic
+    function openPdf(algoName) {
+        if (!algoName) return;
+        const sidebar = document.getElementById('pdf-sidebar');
+        const frame = document.getElementById('pdf-frame');
+        
+        // Use upper case as observed in file system
+        const filename = algoName.toUpperCase() + ".pdf";
+        
+        frame.src = `/pdfs/${filename}`;
+        sidebar.classList.add('open');
+    }
+
+    function closeSidebar() {
+        const sidebar = document.getElementById('pdf-sidebar');
+        sidebar.classList.remove('open');
+        // Clear src after transition to avoid flicker or keep memory usage low
+        setTimeout(() => {
+             document.getElementById('pdf-frame').src = "";
+        }, 300);
+    }
+
+    document.addEventListener("DOMContentLoaded", fetchDatasets);
+
 </script>
 
 </body>
-</html>
+</html>

check_datasets_api.py

@@ -0,0 +1,20 @@
+import urllib.request
+import json
+
+try:
+    url = "http://127.0.0.1:7861/api/datasets"
+    with urllib.request.urlopen(url) as response:
+        data = json.loads(response.read().decode())
+        print(f"Status Code: {response.getcode()}")
+        print(f"Type: {type(data)}")
+        if isinstance(data, list) and len(data) > 0:
+            print(f"First item: {data[0]}")
+            if "last_updated" in data[0]:
+                print("SUCCESS: last_updated field found.")
+            else:
+                print("FAILURE: last_updated field MISSING.")
+        else:
+            print("Data is empty or not a list.")
+            print(data)
+except Exception as e:
+    print(f"Error: {e}")

config.py

@@ -0,0 +1,10 @@
+# MODEL_NAME = "/home/fangsensen/.cache/huggingface/hub/models--deepseek-ai--DeepSeek-R1-Distill-Qwen-1.5B/snapshots/ad9f0ae0864d7fbcd1cd905e3c6c5b069cc8b562"
+MODEL_NAME = "/data1/fangsensen/deepseek-math-7b-rl"
+
+DB_PATH = "data/results.db"
+DATASET_PATH = "data/Authorship.mat"
+EXPR_SEED_PATH = "data/expression_seed.json"
+ITERATIONS = 10
+TOP_K = 10
+CV_FOLDS = 5
+GPU = True

data/Authorship.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d735be2a10e0e6560fe0534f3915a1ca4adc6ec65848d795b53f80623c3355a1
+size 3345720

data/Dermatology.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b85673218727df5d9fd309b382cc8619d8efca653b9fa4b001f5614dea53eeb
+size 700024

data/Factors.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3e8714849b4647d74752aa75a8813c48c1c621de2cef4fa7da54e57a919cfbb
+size 26030704

data/Movement_libras.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:843da25828f4d6195c95f745dfeba9c197ca7dba2d3f527f655d68c37147d104
+size 262320

data/Musk1.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0242a3fdaf222db642b60fa9fd8224be3d94958f17a4bdce3c8062c48a11e6f
+size 636176

data/Synthetic_control.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50157294b6ccc640fe3b3791ea3c784ab414b1a05864de2f931931983c5b1f56
+size 2270504

data/Waveform.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd453b528573aadab4f3385f5d47a2b66d2ac68563757170b3cc66a5290abb19
+size 12820752

data/Wdbc.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6ccf0ff7747481ebe864f935c6cc3d4c342377b2419ab607fec97a91b2351be
+size 139080

data/analyzor.txt

@@ -0,0 +1,57 @@
+[
+    {
+        "expression": "I(X;Y;Z)",
+        "operators": [
+            "Interaction Information"
+        ],
+        "dependency": "Conditional Independence",
+        "theoretical_advantage": "Interaction Information is a measure of the unique information shared between three random variables X, Y, and Z. It can be used to detect conditional independence relationships, which can be useful for feature selection and model simplification.",
+        "complexity": "medium"
+    }
+],
+[
+    {
+        "expression": "I(X;Y)",
+        "operators": [
+            "Mutual Information"
+        ],
+        "dependency": "measures the mutual dependence between two random variables X and Y",
+        "theoretical_advantage": "Mutual Information can be used for feature selection in machine learning, as it measures the dependence between features and the target variable.",
+        "complexity": "low"
+    }
+],
+[
+    {
+        "expression": "I(X;Y|Z)",
+        "operators": [
+            "MI"
+        ],
+        "dependency": "Conditional Independence",
+        "theoretical_advantage": "It can help identify relevant features for classification or regression tasks.",
+        "complexity": "low"
+    }
+],
+[
+    {
+        "expression": "I(X;Y) - I(X;Z)",
+        "operators": [
+            "MI",
+            "MI"
+        ],
+        "dependency": "X and Y are independent of Z",
+        "theoretical_advantage": "It can be used to identify features that are relevant to the target variable but not related to each other.",
+        "complexity": "low"
+    }
+],
+[
+    {
+        "expression": "I(X;Y|Z) - I(X;Y)",
+        "operators": [
+            "Mutual Information",
+            "Conditional Mutual Information"
+        ],
+        "dependency": "Conditional Independence",
+        "theoretical_advantage": "It can help identify features that are relevant to the target variable given a set of other features.",
+        "complexity": "medium"
+    }
+],

data/dna.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0d9a32fd59c16b059be5236efb2f398229f5a41e0034afb27bedc8c126c6f1d
+size 4613568

data/expressions.py

@@ -0,0 +1,9 @@
+EXPRESSIONS = [
+    "I(X;Y)",
+    "I(X;Y) / (H(X) + 1e-6)",
+    "I(X;Y) - 0.1 * I(X;X_other)",
+    "I(X;Y|Z)",
+    "I(X;Y) / (H(X) + H(Y))"
+]
+
+

data/madelon.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c9a047aaa7511f8ca965e05ca275155e0fc7d6a3fab682ecd30665c2c5a9915
+size 8016240

data/results.db

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:531c443d140ed2b80d26ec6ec7131a74567641c6213d32e977c1b885115da4d4
+size 331776

data/spambase.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d49f4a633f845513ff04e29ed6d9aa8fc0cc01eabddaaa682e10bdd33d4d3361
+size 2135104

data/splice.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c3caf785bab5285ebeca87ddff74807c18a57eef9331caf60eb4780466e3a31
+size 778600

download_model.py

@@ -0,0 +1,11 @@
+from huggingface_hub import snapshot_download
+
+MODEL_ID = "deepseek/DeepSeek-R1-Distill"   # 替换成你实际使用的模型
+OUT_DIR = "/home/fangsensen/AutoFS/models/DeepSeek-R1-R1-1.5B"
+
+snapshot_download(
+    repo_id=MODEL_ID,
+    local_dir=OUT_DIR,
+    local_dir_use_symlinks=False,
+)
+print("Done! Model saved at", OUT_DIR)

extract_metadata.py

@@ -0,0 +1,77 @@
+import os
+import json
+import scipy.io
+import numpy as np
+
+DATA_DIR = "data"
+OUTPUT_FILE = "Webapp/dataset_metadata.json"
+
+def get_metadata():
+    metadata = {}
+    if not os.path.exists(DATA_DIR):
+        print(f"Data directory {DATA_DIR} not found.")
+        return metadata
+
+    for filename in os.listdir(DATA_DIR):
+        if filename.endswith(".mat"):
+            name = filename[:-4] # Remove .mat
+            path = os.path.join(DATA_DIR, filename)
+            try:
+                mat = scipy.io.loadmat(path)
+                # Usually X is data, Y is label, or similar keys
+                # We need to find the data key.
+                # Common keys in such datasets: 'X', 'data', 'features'
+                
+                n_samples = 0
+                n_features = 0
+                n_classes = 0
+                
+                # Heuristic to find data
+                keys = [k for k in mat.keys() if not k.startswith('__')]
+                
+                # Assume the largest array is the data
+                max_size = 0
+                data_key = None
+                
+                for k in keys:
+                    if isinstance(mat[k], np.ndarray):
+                        if mat[k].size > max_size:
+                            max_size = mat[k].size
+                            data_key = k
+                
+                if data_key:
+                    data = mat[data_key]
+                    if len(data.shape) == 2:
+                        n_samples, n_features = data.shape
+                    
+                    # Try to find labels to count classes
+                    # Usually the other array or 'Y'
+                    label_key = None
+                    for k in keys:
+                        if k != data_key and isinstance(mat[k], np.ndarray):
+                            # Labels usually have same length as samples
+                            if mat[k].shape[0] == n_samples or (len(mat[k].shape) > 1 and mat[k].shape[1] == n_samples):
+                                label_key = k
+                                break
+                    
+                    if label_key:
+                        labels = mat[label_key]
+                        n_classes = len(np.unique(labels))
+                    
+                metadata[name] = {
+                    "n_samples": int(n_samples),
+                    "n_features": int(n_features),
+                    "n_classes": int(n_classes)
+                }
+                print(f"Processed {name}: {n_samples}x{n_features}, {n_classes} classes")
+                
+            except Exception as e:
+                print(f"Error processing {filename}: {e}")
+                
+    return metadata
+
+if __name__ == "__main__":
+    meta = get_metadata()
+    with open(OUTPUT_FILE, "w") as f:
+        json.dump(meta, f, indent=2)
+    print(f"Metadata saved to {OUTPUT_FILE}")

main.py

@@ -0,0 +1,244 @@
+# import json, datetime
+# from config import *
+# import pandas as pd
+# import scipy.io as scio
+# from modules.expression_pool import init_db, add_expr, top_exprs
+# from agents.generator_agent import GeneratorAgent
+# from agents.evaluator_agent import evaluate_expression
+# from modules.utils import load_mat_as_numeric
+# # prepare
+# conn = init_db(DB_PATH)
+# # df = pd.read_csv(DATASET_PATH)
+# # df = scio.loadmat(DATASET_PATH)  # 读取数据文件
+# X, y = load_mat_as_numeric(DATASET_PATH)
+# X_df = pd.DataFrame(X)
+# # print(df)
+# # X_df = pd.DataFrame(df['X'])  # 读取训练数据
+# # print(df['Y'])
+# # y0 = pd.DataFrame(df['Y'])  # 读取标签
+# # X_df = df.drop(columns=['label'])
+# # y = y0.values
+# # print("y type:", type(y), "dtype:", getattr(y, "dtype", None))
+# # print("y example:", y[:10])
+
+
+# # load seed
+# with open(EXPR_SEED_PATH) as f:
+#     seeds = json.load(f)
+# # evaluate seeds first
+# for s in seeds:
+#     score, fvals, top_idx = evaluate_expression(s['expression'], X_df, y, TOP_K, CV_FOLDS)
+#     add_expr(conn, s['expression'], score, s.get('explanation',''), str(s.get('complexity','')))
+
+# # init generator
+# gen = GeneratorAgent(MODEL_NAME)
+
+# # iterative loop
+# for it in range(ITERATIONS):
+#     print("Iteration", it+1)
+#     refs = top_exprs(conn, k=TOP_K)
+#     # build prompt_text with refs + feature stats
+#     # prompt = "Given top expressions: " + str(refs) + "\nGenerate expressions in format: Expression: ... Rationale: ..."
+#     top_expressions = []  # List[(expr, score)]
+#     top_expressions.append((refs, score))
+#     top_expressions = sorted(
+#         top_expressions,
+#         key=lambda x: -x[1]
+#     )[:5]
+#     new_text = gen.generate_candidates(top_expressions)
+#     for out in new_text:
+#         # extract Expression line
+#         expr_line = None
+#         for line in out.splitlines():
+#             if line.strip().lower().startswith("expression"):
+#                 expr_line = line.split(":",1)[1].strip()
+#                 break
+#         if not expr_line: expr_line = out.strip()
+#         score, fvals, top_idx = evaluate_expression(expr_line, X_df, y, TOP_K, CV_FOLDS)
+#         add_expr(conn, expr_line, score, out, "")
+#         print(f"Candidate {expr_line} -> score {score:.4f}")
+
+
+
+# results = []
+
+# for expr in EXPRESSIONS:
+#     exec_out = executor.run(expr, X, y)
+#     analysis = analyzer.analyze(expr, exec_out["cv_score"])
+
+#     results.append({
+#         "expression": expr,
+#         "score": exec_out["cv_score"],
+#         "analysis": analysis
+#     })
+
+# ranking = judge.rank(results)
+#-----------------------------------------------------------------------2.0---------------
+
+# from agents.analyzer_agent import AnalyzerAgent
+
+# MODEL_PATH = "/data1/fangsensen/deepseek-math-7b-rl"
+
+# agent = AnalyzerAgent(
+#     name="AnalyzerAgent",
+#     model_path=MODEL_PATH
+# )
+
+# expressions = [
+#     "I(X;Y)",
+#     "I(X;Y|Z)",
+#     "I(X;Y) - I(X;Z)",
+#     "I(X;Y|Z) - I(X;Y)",
+#     "I(X;Y;Z)"
+# ]
+# # expressions = [
+# #     "I(X;Y|Z) - I(X;Y)",
+# # ]
+# for expr in expressions:
+#     print("=" * 80)
+#     result = agent.analyze_expression(expr)
+#     print(result)
+#-----------------------------------------------------------------------路由---------------
+import numpy as np
+from agents.router_agent import FSRouterAgent
+
+import scipy.io as scio
+import pandas as pd
+from sklearn.preprocessing import LabelEncoder
+
+def load_mat_dataset(
+    file_path,
+    feature_keys=("X", "data", "fea"),
+    label_keys=("Y", "y", "label"),
+):
+    """
+    通用 .mat 数据集读取函数（FSExecutor / Agent 兼容）
+
+    Parameters
+    ----------
+    file_path : str
+        .mat 文件路径
+    feature_keys : tuple
+        特征矩阵可能的 key
+    label_keys : tuple
+        标签可能的 key
+
+    Returns
+    -------
+    X : np.ndarray, shape (n_samples, n_features)
+    y : np.ndarray, shape (n_samples,)
+    meta : dict
+        元信息（类别数、样本数等）
+    """
+
+    data = scio.loadmat(file_path)
+
+    # ---------- 1. 读取 X ----------
+    X = None
+    for key in feature_keys:
+        if key in data:
+            X = data[key]
+            break
+    if X is None:
+        raise KeyError(f"Cannot find feature matrix in {file_path}")
+
+    X = np.asarray(X)
+
+    if X.dtype == object:
+        X = np.array(
+            [[float(v[0]) if isinstance(v, (list, np.ndarray)) else float(v)
+            for v in row]
+            for row in X]
+        )
+    else:
+        X = X.astype(float)
+
+
+    # ---------- 2. 读取 y ----------
+    y = None
+    for key in label_keys:
+        if key in data:
+            y = data[key]
+            break
+    if y is None:
+        raise KeyError(f"Cannot find label vector in {file_path}")
+
+    # y 常见是 (n,1)
+    y = np.asarray(y).reshape(-1)
+
+    # ---------- 3. 标签清洗 & 编码 ----------
+    # 处理 object / string / 混合类型
+    if y.dtype == object:
+        y = pd.Series(y).apply(lambda x: x[0] if isinstance(x, (list, np.ndarray)) else x)
+
+    label_encoder = LabelEncoder()
+    y = label_encoder.fit_transform(y)
+
+    # ---------- 4. 元信息 ----------
+    meta = {
+        "n_samples": X.shape[0],
+        "n_features": X.shape[1],
+        "n_classes": len(np.unique(y)),
+        "classes": np.unique(y),
+        "label_encoder": label_encoder,
+    }
+
+    return X, y, meta
+
+base_url = "/home/fangsensen/AutoFS/data/"
+datanames = ['dna','Factors','madelon','Movement_libras','Musk1','spambase','splice','Synthetic_control',  'Waveform','Wdbc',]
+# dataname = 'Authorship'
+def main(dataname):
+    X, y, meta = load_mat_dataset(
+    base_url + dataname + ".mat"
+)
+    # X = data.data
+    # y = data.target
+    #
+
+    task = {
+        "X": X,
+        "y": y,
+        "algorithms": ["JMIM","CFR","DCSF","IWFS","MRI","MRMD","UCRFS","CSMDCCMR",],
+        "n_selected_features": 5,
+        "class_specific": False,
+        "classifiers": ["nb", "svm", "rf"],
+        "cv": 10,
+        "random_state": 19,
+        "params":{"n_selected_features":15,},
+        "dataname":dataname,
+    }
+
+
+    router = FSRouterAgent()
+    leaderboard = router.run(task)
+    
+    for rank, res in enumerate(leaderboard, 1):
+        print(f"Rank {rank}: {res}")
+    return leaderboard
+
+if __name__ == "__main__":
+    for dataname in datanames:
+        main(dataname)
+
+
+
+
+# {'selected_features': [59, 50, 56, 4, 38, 9, 29, 23, 0, 20, 34, 36, 24, 26, 28], 
+#  'num_features': 15, 
+#  'metrics': {'nb': {'f1': 0.9181133571145461, 'auc': 0.9807805770573524}, 
+#              'svm': {'f1': 0.9282600079270711, 'auc': 0.980695564275392}, 
+#              'rf': {'f1': 0.9219976218787156, 'auc': 0.9768411621948705}}, 
+#  'time': 7.378173112869263, 
+#  'algorithm': 'JMIM'}, 
+
+
+# {'selected_features': [59, 50, 56, 4, 38, 0, 9, 29, 23, 20, 36, 34, 24, 28, 26], 
+#  'num_features': 15, 
+#  'metrics': {'nb': {'f1': 0.9163694015061433, 'auc': 0.9805189493459717}, 
+#              'svm': {'f1': 0.9265953230281413, 'auc': 0.98064247666047}, 
+#              'rf': {'f1': 0.9189853349187476, 'auc': 0.9769441217042379}}, 
+#  'time': 2.0774385929107666, 
+#  'algorithm': 'CFR'}
+
+

modules/expr_to_code.py

@@ -0,0 +1,163 @@
+import sympy as sp
+import numpy as np
+from sklearn.metrics import mutual_info_score
+# 符号
+import sympy as sp
+import pandas as pd
+
+# symbols
+X = sp.Symbol("X")
+Y = sp.Symbol("Y")
+Z = sp.Symbol("Z")
+
+class MI(sp.Function):
+    nargs = (2,)
+
+class CMI(sp.Function):
+    nargs = (3,)
+
+class II(sp.Function):
+    nargs = (3,)   # interaction information
+
+ALLOWED_LOCALS = {
+    "X": X,
+    "Y": Y,
+    "Z": Z,
+    "I": MI,      # I(X,Y)
+    "CI": CMI,    # I(X,Y|Z) 条件互信息
+    "II": II      # I(X；Y；Z)交互信息
+}
+
+def parse_expression(expr_str: str) -> sp.Expr:
+    """
+    String → SymPy Expression
+    """
+    expr = sp.sympify(expr_str, locals=ALLOWED_LOCALS)
+    return expr
+
+
+def entropy(x):#计算熵
+    _, cnt = np.unique(x, return_counts=True)
+    p = cnt / cnt.sum()
+    return -np.sum(p * np.log(p + 1e-12))
+
+def mi(x, y):#互信息
+    return mutual_info_score(x, y)
+
+def cmi(x, y, z):#条件互信息（通过熵的加减计算）
+    # I(X;Y|Z) = H(X,Z)+H(Y,Z)-H(Z)-H(X,Y,Z)
+    return (
+        entropy(np.c_[x, z].tolist())
+        + entropy(np.c_[y, z].tolist())
+        - entropy(z)
+        - entropy(np.c_[x, y, z].tolist())
+    )
+
+def interaction_info(x, y, z):#交互信息
+    # I(X;Y;Z) = I(X;Y) - I(X;Y|Z)
+    return mi(x, y) - cmi(x, y, z)
+
+
+def expr_to_callable(expr: sp.Expr):
+
+    def eval_node(node, ctx):
+        if isinstance(node, MI):
+            return mi(eval_node(node.args[0], ctx),
+                      eval_node(node.args[1], ctx))
+
+        if isinstance(node, CMI):
+            return cmi(eval_node(node.args[0], ctx),
+                       eval_node(node.args[1], ctx),
+                       eval_node(node.args[2], ctx))
+
+        if isinstance(node, II):
+            return interaction_info(
+                eval_node(node.args[0], ctx),
+                eval_node(node.args[1], ctx),
+                eval_node(node.args[2], ctx)
+            )
+
+        if node == X:
+            return ctx["X"]
+        if node == Y:
+            return ctx["Y"]
+        if node == Z:
+            return ctx["Z"]
+
+        if node.is_Number:
+            return float(node)
+
+        if node.is_Add:
+            return sum(eval_node(arg, ctx) for arg in node.args)
+
+        if node.is_Mul:
+            r = 1.0
+            for arg in node.args:
+                r *= eval_node(arg, ctx)
+            return r
+
+        if node.is_Pow:
+            base, exp = node.args
+            return eval_node(base, ctx) ** eval_node(exp, ctx)
+
+        raise ValueError(f"Unsupported node: {node}")
+
+    def f(X_arr, Y_arr, Z_arr=None):
+        ctx = {"X": X_arr, "Y": Y_arr}
+        if Z_arr is not None:
+            ctx["Z"] = Z_arr
+        return eval_node(expr, ctx)
+
+    return f
+
+
+
+from sklearn.preprocessing import LabelEncoder
+def changetosinge(x):
+    return float(x)
+# scores = f(X, y, X_other_list)
+def prepare_data(dataname, base_url):
+    url = os.path.join(base_url, dataname + '.mat')
+    data = scio.loadmat(url)
+    X0 = pd.DataFrame(data['X'])
+    y0 = pd.DataFrame(data['Y'])
+
+    if dataname == 'Dermatology':
+        Special = X0.iloc[:, -1]
+        a = np.array([item[0] for item in Special])
+        label_encoder = LabelEncoder()
+        a33 = label_encoder.fit_transform(a)
+        X0 = X0.iloc[:, :-1]
+        X0[33] = a33
+
+    X0 = X0.applymap(changetosinge)
+    y0 = y0.applymap(changetosinge)
+    label_encoder = LabelEncoder()
+    y_encoded = label_encoder.fit_transform(y0)
+    y = pd.DataFrame(y_encoded)
+    X = pd.DataFrame()
+
+    for col in X0.columns:
+        X[col] = pd.cut(X0[col], bins=5, labels=False)
+
+    new_columns = [str(i) for i in range(X.shape[1] + 1)]
+    X = X.rename(columns=dict(zip(X.columns, new_columns[:-1])))
+    y = y.rename(columns=dict(zip(y.columns, [new_columns[-1]])))
+    data_processed = pd.concat([X, y], axis=1)
+    # data_processed = pd.DataFrame(X)
+
+    return data_processed, list(set(y_encoded))
+
+import os
+import scipy.io as scio
+dataname = 'Authorship'
+base_url = '/home/fangsensen/AutoFS/data/'
+data_processed, class_set = prepare_data(dataname, base_url)
+# print(data_processed)
+# X_arr = data_processed['0']
+# y_arr = data_processed['69']
+print(111111,X_arr,2222222,y_arr)
+expr = parse_expression("I(X,Y)")
+f = expr_to_callable(expr)
+score = f(X_arr, y_arr)
+print(score)

modules/expression_pool.py

@@ -0,0 +1,24 @@
+import sqlite3, os
+def init_db(db_path="data/results.db"):
+    os.makedirs(os.path.dirname(db_path), exist_ok=True)
+    conn = sqlite3.connect(db_path)
+    conn.execute("""
+    CREATE TABLE IF NOT EXISTS exprs(
+        expr TEXT PRIMARY KEY,
+        score REAL,
+        rationale TEXT,
+        complexity TEXT,
+        created_at TEXT
+    )""")
+    conn.commit()
+    return conn
+
+def add_expr(conn, expr, score, rationale="", complexity=""):
+    conn.execute("INSERT OR REPLACE INTO exprs(expr, score, rationale, complexity, created_at) VALUES(?,?,?,?,datetime('now'))",
+                 (expr, score, rationale, complexity))
+    conn.commit()
+
+def top_exprs(conn, k=5, min_score=0.0):
+    cur = conn.cursor()
+    cur.execute("SELECT expr,score,rationale FROM exprs WHERE score>=? ORDER BY score DESC LIMIT ?", (min_score,k))
+    return cur.fetchall()

modules/modules/expr_to_code.py

@@ -0,0 +1,163 @@
+import sympy as sp
+import numpy as np
+from sklearn.metrics import mutual_info_score
+# 符号
+import sympy as sp
+import pandas as pd
+
+# symbols
+X = sp.Symbol("X")
+Y = sp.Symbol("Y")
+Z = sp.Symbol("Z")
+
+class MI(sp.Function):
+    nargs = (2,)
+
+class CMI(sp.Function):
+    nargs = (3,)
+
+class II(sp.Function):
+    nargs = (3,)   # interaction information
+
+ALLOWED_LOCALS = {
+    "X": X,
+    "Y": Y,
+    "Z": Z,
+    "I": MI,      # I(X,Y)
+    "CI": CMI,    # I(X,Y|Z) 条件互信息
+    "II": II      # I(X；Y；Z)交互信息
+}
+
+def parse_expression(expr_str: str) -> sp.Expr:
+    """
+    String → SymPy Expression
+    """
+    expr = sp.sympify(expr_str, locals=ALLOWED_LOCALS)
+    return expr
+
+
+def entropy(x):#计算熵
+    _, cnt = np.unique(x, return_counts=True)
+    p = cnt / cnt.sum()
+    return -np.sum(p * np.log(p + 1e-12))
+
+def mi(x, y):#互信息
+    return mutual_info_score(x, y)
+
+def cmi(x, y, z):#条件互信息（通过熵的加减计算）
+    # I(X;Y|Z) = H(X,Z)+H(Y,Z)-H(Z)-H(X,Y,Z)
+    return (
+        entropy(np.c_[x, z].tolist())
+        + entropy(np.c_[y, z].tolist())
+        - entropy(z)
+        - entropy(np.c_[x, y, z].tolist())
+    )
+
+def interaction_info(x, y, z):#交互信息
+    # I(X;Y;Z) = I(X;Y) - I(X;Y|Z)
+    return mi(x, y) - cmi(x, y, z)
+
+
+def expr_to_callable(expr: sp.Expr):
+
+    def eval_node(node, ctx):
+        if isinstance(node, MI):
+            return mi(eval_node(node.args[0], ctx),
+                      eval_node(node.args[1], ctx))
+
+        if isinstance(node, CMI):
+            return cmi(eval_node(node.args[0], ctx),
+                       eval_node(node.args[1], ctx),
+                       eval_node(node.args[2], ctx))
+
+        if isinstance(node, II):
+            return interaction_info(
+                eval_node(node.args[0], ctx),
+                eval_node(node.args[1], ctx),
+                eval_node(node.args[2], ctx)
+            )
+
+        if node == X:
+            return ctx["X"]
+        if node == Y:
+            return ctx["Y"]
+        if node == Z:
+            return ctx["Z"]
+
+        if node.is_Number:
+            return float(node)
+
+        if node.is_Add:
+            return sum(eval_node(arg, ctx) for arg in node.args)
+
+        if node.is_Mul:
+            r = 1.0
+            for arg in node.args:
+                r *= eval_node(arg, ctx)
+            return r
+
+        if node.is_Pow:
+            base, exp = node.args
+            return eval_node(base, ctx) ** eval_node(exp, ctx)
+
+        raise ValueError(f"Unsupported node: {node}")
+
+    def f(X_arr, Y_arr, Z_arr=None):
+        ctx = {"X": X_arr, "Y": Y_arr}
+        if Z_arr is not None:
+            ctx["Z"] = Z_arr
+        return eval_node(expr, ctx)
+
+    return f
+
+
+
+from sklearn.preprocessing import LabelEncoder
+def changetosinge(x):
+    return float(x)
+# scores = f(X, y, X_other_list)
+def prepare_data(dataname, base_url):
+    url = os.path.join(base_url, dataname + '.mat')
+    data = scio.loadmat(url)
+    X0 = pd.DataFrame(data['X'])
+    y0 = pd.DataFrame(data['Y'])
+
+    if dataname == 'Dermatology':
+        Special = X0.iloc[:, -1]
+        a = np.array([item[0] for item in Special])
+        label_encoder = LabelEncoder()
+        a33 = label_encoder.fit_transform(a)
+        X0 = X0.iloc[:, :-1]
+        X0[33] = a33
+
+    X0 = X0.applymap(changetosinge)
+    y0 = y0.applymap(changetosinge)
+    label_encoder = LabelEncoder()
+    y_encoded = label_encoder.fit_transform(y0)
+    y = pd.DataFrame(y_encoded)
+    X = pd.DataFrame()
+
+    for col in X0.columns:
+        X[col] = pd.cut(X0[col], bins=5, labels=False)
+
+    new_columns = [str(i) for i in range(X.shape[1] + 1)]
+    X = X.rename(columns=dict(zip(X.columns, new_columns[:-1])))
+    y = y.rename(columns=dict(zip(y.columns, [new_columns[-1]])))
+    data_processed = pd.concat([X, y], axis=1)
+    # data_processed = pd.DataFrame(X)
+
+    return data_processed, list(set(y_encoded))
+
+import os
+import scipy.io as scio
+dataname = 'Authorship'
+base_url = '/home/fangsensen/AutoFS/data/'
+data_processed, class_set = prepare_data(dataname, base_url)
+# print(data_processed)
+# X_arr = data_processed['0']
+# y_arr = data_processed['69']
+print(111111,X_arr,2222222,y_arr)
+expr = parse_expression("I(X,Y)")
+f = expr_to_callable(expr)
+score = f(X_arr, y_arr)
+print(score)

modules/modules/expression_pool.py

@@ -0,0 +1,24 @@
+import sqlite3, os
+def init_db(db_path="data/results.db"):
+    os.makedirs(os.path.dirname(db_path), exist_ok=True)
+    conn = sqlite3.connect(db_path)
+    conn.execute("""
+    CREATE TABLE IF NOT EXISTS exprs(
+        expr TEXT PRIMARY KEY,
+        score REAL,
+        rationale TEXT,
+        complexity TEXT,
+        created_at TEXT
+    )""")
+    conn.commit()
+    return conn
+
+def add_expr(conn, expr, score, rationale="", complexity=""):
+    conn.execute("INSERT OR REPLACE INTO exprs(expr, score, rationale, complexity, created_at) VALUES(?,?,?,?,datetime('now'))",
+                 (expr, score, rationale, complexity))
+    conn.commit()
+
+def top_exprs(conn, k=5, min_score=0.0):
+    cur = conn.cursor()
+    cur.execute("SELECT expr,score,rationale FROM exprs WHERE score>=? ORDER BY score DESC LIMIT ?", (min_score,k))
+    return cur.fetchall()

modules/modules/utils.py

@@ -0,0 +1,35 @@
+import numpy as np
+import scipy.io as sio
+
+def load_mat_as_numeric(path, x_key="X", y_key="Y"):
+    data = sio.loadmat(path)
+    X_raw = data[x_key]
+    y_raw = data[y_key]
+
+    # Step 1: flatten MATLAB cell array elements
+    def clean_cell_array(arr):
+        cleaned = []
+        for row in arr:
+            new_row = []
+            for elem in row:
+                # elem is usually array(['46.0'])
+                if isinstance(elem, np.ndarray):
+                    elem = elem[0]  # '46.0'
+                elem = elem.strip()
+                new_row.append(elem)
+            cleaned.append(new_row)
+        return np.array(cleaned)
+
+    X_str = clean_cell_array(X_raw)
+    y_str = clean_cell_array(y_raw).reshape(-1)
+
+    # Step 2: convert X to float
+    X = X_str.astype(float)
+
+    # Step 3: convert y to numeric or keep string
+    try:
+        y = y_str.astype(float)
+    except:
+        y = y_str.astype(str)
+
+    return X, y

modules/utils.py

@@ -0,0 +1,35 @@
+import numpy as np
+import scipy.io as sio
+
+def load_mat_as_numeric(path, x_key="X", y_key="Y"):
+    data = sio.loadmat(path)
+    X_raw = data[x_key]
+    y_raw = data[y_key]
+
+    # Step 1: flatten MATLAB cell array elements
+    def clean_cell_array(arr):
+        cleaned = []
+        for row in arr:
+            new_row = []
+            for elem in row:
+                # elem is usually array(['46.0'])
+                if isinstance(elem, np.ndarray):
+                    elem = elem[0]  # '46.0'
+                elem = elem.strip()
+                new_row.append(elem)
+            cleaned.append(new_row)
+        return np.array(cleaned)
+
+    X_str = clean_cell_array(X_raw)
+    y_str = clean_cell_array(y_raw).reshape(-1)
+
+    # Step 2: convert X to float
+    X = X_str.astype(float)
+
+    # Step 3: convert y to numeric or keep string
+    try:
+        y = y_str.astype(float)
+    except:
+        y = y_str.astype(str)
+
+    return X, y

pdf/CFR.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43040c5cd02372547ccfd0ff233c1f8db8492bb79cd6648471a6b261197a291b
+size 1011137

pdf/CIFE.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40f99e19316a356c47d3c228bb17ac7d4200a55e6d36da9adbc01d13ae978aab
+size 900327

pdf/CMIFS.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9afa507819700f14a66009e7ea7061b178427de7a0c4455d1a666c16b7d261e5
+size 782737

pdf/CMIM.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e59696e31620f0328f7a5aaf9c667180537c03892735ff68cbbbfa721bd72506
+size 192556

pdf/CSMDCCMR.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de200fbb840c74fcd142883c81ff10824a872d06292559abb178bb937ee0e541
+size 1210174

pdf/CSMI.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc3c25a21685f327822a6794f3986a8d89bee172f6fb413f063529464c608dba
+size 5412954

pdf/DCSF.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31b0c67c1eadb708396803b6991298c4b965b0ed827b455a77e02281c9505475
+size 1533397

pdf/DISR.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12163bcb61689372a0fa6c057b55614e70fd263c75cd0869e58fb1a8bc9ef85b
+size 215860

pdf/DWFS.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7d9f02c654e665591d365940f0b7239eb5d64f669210683eef7282aede6c378
+size 817732

pdf/IWFS.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f8387f8508dbe3f74abfb41acf1c68f3da99e93dc465b10e680051fadd7e091
+size 582942

pdf/JMI.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8491d05d11decc319e5d83f3867bc06bf9ccc984f6bd1854060684067bbd14c
+size 1442313

pdf/JMIM.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:880adb6df5fec2761b2aafbbf555dc9aebfd10fe9016dd92eb5b4ff481494dd9
+size 1064488

pdf/MIM.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:232e906e125fa5173af1ed66b446740d11e5c43b2dd911f3af729672141b4fbb
+size 506331

pdf/MRI.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ca2c265bb597e2886c33448b028845f0755ca0cd952eb319dc74f31929ad300
+size 1189309

pdf/MRMD.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d067d694f1c70da19da5455237a5f33601c213f11f452ab73ee0d7ce9ccca8a9
+size 2395695

pdf/MRMR.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb30d4b51eb630aff13a33cdd539d756b99021fa3446ad61cf82322cb5b97dee
+size 1295526

pdf/UCRFS.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fb819c4f606121529e3f7c4de93e652fb1d7d88173600adb87ea0188e8bd528
+size 1823221

requirements.txt

@@ -1,3 +1,11 @@
-Flask
-pandas
+bitsandbytes
+accelerate
+sympy
 numpy
+pandas
+scikit-learn
+sqlalchemy
+lark-parser
+autogen   
+tqdm
+python-dotenv

test.py

@@ -0,0 +1,7 @@
+import pickle
+
+
+path = "results/dna.pkl"
+with open(path, "rb") as f:
+    results = pickle.load(f)
+    print(results)