Spaces:
Sleeping
Sleeping
File size: 9,539 Bytes
6cc8ae1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 | {
"cells": [
{
"cell_type": "markdown",
"id": "228b68f5",
"metadata": {},
"source": [
"# 05 \u2014 Model Comparison & Best Model Selection\n",
"\n",
"Compare all 4 trained models using a weighted composite score:\n",
"\n",
"| Metric | Weight |\n",
"|--------|--------|\n",
"| Macro F1 | 50% |\n",
"| Top-1 Accuracy | 20% |\n",
"| Inference Latency | 15% |\n",
"| Model Size | 10% |\n",
"| Calibration | 5% |\n",
"\n",
"### Expected Output\n",
"- Radar chart comparing all models\n",
"- Grouped bar chart of key metrics\n",
"- Comparison table with rankings\n",
"- Best model recommendation"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b7c96a7a",
"metadata": {},
"outputs": [],
"source": [
"import sys, os, json\n",
"from pathlib import Path\n",
"\n",
"PROJECT_ROOT = Path(os.getcwd()).resolve()\n",
"if 'notebooks' in str(PROJECT_ROOT):\n",
" PROJECT_ROOT = PROJECT_ROOT.parent.parent\n",
"sys.path.insert(0, str(PROJECT_ROOT))\n",
"print(f'Project root: {PROJECT_ROOT}')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "17733cd5",
"metadata": {},
"outputs": [],
"source": [
"import pandas as pd\n",
"import matplotlib.pyplot as plt\n",
"%matplotlib inline\n",
"\n",
"from ml.src.evaluation.compare_models import (\n",
" compute_weighted_scores,\n",
" plot_comparison_radar,\n",
" plot_comparison_bar,\n",
" generate_comparison_report,\n",
")"
]
},
{
"cell_type": "markdown",
"id": "c29af067",
"metadata": {},
"source": [
"## 1. Load All Model Reports"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cea1417d",
"metadata": {},
"outputs": [],
"source": [
"reports_dir = PROJECT_ROOT / 'ml' / 'artifacts' / 'reports'\n",
"model_names = ['mlp', 'cnn', 'resnet', 'vit']\n",
"\n",
"model_reports = []\n",
"for name in model_names:\n",
" report_path = reports_dir / f'{name}_report.json'\n",
" if report_path.exists():\n",
" with open(report_path) as f:\n",
" report = json.load(f)\n",
" model_reports.append(report)\n",
" print(f'\u2713 Loaded {name} report')\n",
" else:\n",
" print(f'\u2717 Missing {name} report \u2014 run notebook {model_names.index(name)+1:02d} first')\n",
"\n",
"print(f'\\nLoaded {len(model_reports)} / {len(model_names)} model reports')"
]
},
{
"cell_type": "markdown",
"id": "570fe372",
"metadata": {},
"source": [
"## 2. Summary Table"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "18f372b6",
"metadata": {},
"outputs": [],
"source": [
"rows = []\n",
"for r in model_reports:\n",
" metrics = r.get('metrics', {})\n",
" latency = r.get('latency', {})\n",
" rows.append({\n",
" 'Model': r['model_name'],\n",
" 'Accuracy': f\"{metrics.get('accuracy', 0):.4f}\",\n",
" 'Macro F1': f\"{metrics.get('macro_f1', 0):.4f}\",\n",
" 'Macro Precision': f\"{metrics.get('macro_precision', 0):.4f}\",\n",
" 'Macro Recall': f\"{metrics.get('macro_recall', 0):.4f}\",\n",
" 'Latency (ms)': f\"{latency.get('avg_ms', 0):.1f}\",\n",
" 'Size (MB)': f\"{r.get('model_size_mb', 0):.1f}\",\n",
" 'Parameters': f\"{r.get('num_parameters', 0):,}\",\n",
" })\n",
"\n",
"df_summary = pd.DataFrame(rows)\n",
"df_summary.style.set_caption('Model Comparison Summary')"
]
},
{
"cell_type": "markdown",
"id": "1823a6bb",
"metadata": {},
"source": [
"## 3. Weighted Scoring & Ranking"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3e4addc9",
"metadata": {},
"outputs": [],
"source": [
"figures_dir = PROJECT_ROOT / 'ml' / 'artifacts' / 'figures' / 'comparison'\n",
"figures_dir.mkdir(parents=True, exist_ok=True)\n",
"\n",
"comparison_report = generate_comparison_report(\n",
" model_reports,\n",
" output_dir=figures_dir,\n",
")\n",
"\n",
"print('\\n=== Rankings ===')\n",
"for r in comparison_report['rankings']:\n",
" print(f\" #{r['rank']} {r['model_name']:>8s} \"\n",
" f\"composite={r['composite_score']:.4f} \"\n",
" f\"F1={r['raw_metrics']['macro_f1']:.4f} \"\n",
" f\"Acc={r['raw_metrics']['accuracy']:.4f} \"\n",
" f\"Lat={r['raw_metrics']['latency_ms']:.1f}ms \"\n",
" f\"Size={r['raw_metrics']['model_size_mb']:.1f}MB\")"
]
},
{
"cell_type": "markdown",
"id": "45637652",
"metadata": {},
"source": [
"## 4. Radar Chart"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "672dd90d",
"metadata": {},
"outputs": [],
"source": [
"from IPython.display import Image as IPImage, display\n",
"radar_path = figures_dir / 'comparison_radar.png'\n",
"if radar_path.exists():\n",
" display(IPImage(filename=str(radar_path), width=600))"
]
},
{
"cell_type": "markdown",
"id": "cb4d3040",
"metadata": {},
"source": [
"## 5. Bar Chart"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0101ced1",
"metadata": {},
"outputs": [],
"source": [
"bar_path = figures_dir / 'comparison_bar.png'\n",
"if bar_path.exists():\n",
" display(IPImage(filename=str(bar_path), width=800))"
]
},
{
"cell_type": "markdown",
"id": "5fd19a50",
"metadata": {},
"source": [
"## 6. Per-Model Confusion Matrices"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c62b87c5",
"metadata": {},
"outputs": [],
"source": [
"fig, axes = plt.subplots(1, len(model_reports), figsize=(6*len(model_reports), 5))\n",
"if len(model_reports) == 1:\n",
" axes = [axes]\n",
"\n",
"for idx, report in enumerate(model_reports):\n",
" cm_path = PROJECT_ROOT / 'ml' / 'artifacts' / 'figures' / report['model_name'] / 'confusion_matrix.png'\n",
" if cm_path.exists():\n",
" img = plt.imread(str(cm_path))\n",
" axes[idx].imshow(img)\n",
" axes[idx].set_title(report['model_name'], fontweight='bold')\n",
" axes[idx].axis('off')\n",
"\n",
"plt.tight_layout()\n",
"plt.show()"
]
},
{
"cell_type": "markdown",
"id": "b6adb115",
"metadata": {},
"source": [
"## 7. Best Model Recommendation"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ca574420",
"metadata": {},
"outputs": [],
"source": [
"best = comparison_report['rankings'][0]\n",
"print('=' * 60)\n",
"print(f'\ud83c\udfc6 RECOMMENDED MODEL: {best[\"model_name\"].upper()}')\n",
"print('=' * 60)\n",
"print(f' Composite Score: {best[\"composite_score\"]:.4f}')\n",
"print(f' Macro F1: {best[\"raw_metrics\"][\"macro_f1\"]:.4f}')\n",
"print(f' Accuracy: {best[\"raw_metrics\"][\"accuracy\"]:.4f}')\n",
"print(f' Latency: {best[\"raw_metrics\"][\"latency_ms\"]:.1f} ms')\n",
"print(f' Model Size: {best[\"raw_metrics\"][\"model_size_mb\"]:.1f} MB')\n",
"print('\\nCheckpoint path:')\n",
"print(f' ml/artifacts/checkpoints/{best[\"model_name\"]}_best.pth')\n",
"print('\\nTo deploy this model, update the backend environment variable:')\n",
"print(f' MODEL_PATH=ml/artifacts/checkpoints/{best[\"model_name\"]}_best.pth')\n",
"print(f' MODEL_NAME={best[\"model_name\"]}')"
]
},
{
"cell_type": "markdown",
"id": "9c83b31c",
"metadata": {},
"source": [
"## 8. Export for Deployment\n",
"\n",
"Copy the best checkpoint to the `models/` directory for the backend to load."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2c1084ec",
"metadata": {},
"outputs": [],
"source": [
"import shutil\n",
"\n",
"best_name = best['model_name']\n",
"src = PROJECT_ROOT / 'ml' / 'artifacts' / 'checkpoints' / f'{best_name}_best.pth'\n",
"dst = PROJECT_ROOT / 'models' / f'{best_name}_best.pth'\n",
"\n",
"if src.exists():\n",
" shutil.copy2(src, dst)\n",
" print(f'\u2713 Copied {src.name} \u2192 {dst}')\n",
"\n",
" # Also save class names for the backend\n",
" import torch\n",
" ckpt = torch.load(src, map_location='cpu', weights_only=False)\n",
" if 'class_names' in ckpt:\n",
" classes_path = PROJECT_ROOT / 'models' / 'classes.txt'\n",
" with open(classes_path, 'w') as f:\n",
" f.write(','.join(ckpt['class_names']))\n",
" print(f'\u2713 Updated {classes_path}')\n",
"else:\n",
" print(f'\u2717 Checkpoint not found: {src}')"
]
},
{
"cell_type": "markdown",
"id": "ebda2f85",
"metadata": {},
"source": [
"---\n",
"**\u2705 Model comparison complete!**\n",
"\n",
"### Next Steps\n",
"1. Start the backend: `PYTHONPATH=. uvicorn backend.app.main:app --reload`\n",
"2. Start the frontend: `cd frontend && npm run dev`\n",
"3. Upload images and verify predictions"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Cattle Classifier (Python 3.12)",
"language": "python",
"name": "cattle-classifier"
}
},
"nbformat": 4,
"nbformat_minor": 5
} |