{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "d9953ed2",
   "metadata": {},
   "source": [
    "# Brain Tumor Detection Experiment Tracking\n",
    "This notebook helps track dataset layout, model performance, and evaluation metrics for the brain tumor detection project."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "51214413",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from PIL import Image\n",
    "from src.utils import load_history_npz, load_metrics_json\n",
    "\n",
    "DATASET_ROOT = Path('dataset')\n",
    "ARTIFACTS_ROOT = Path('artifacts')\n",
    "MODEL_TYPES = ['cnn', 'transfer', 'vit']\n",
    "\n",
    "\n",
    "def inspect_dataset_structure(dataset_root):\n",
    "    root = Path(dataset_root)\n",
    "    layout = {}\n",
    "    for split in ['train', 'val', 'test']:\n",
    "        split_dir = root / split\n",
    "        if not split_dir.exists():\n",
    "            layout[split] = None\n",
    "            continue\n",
    "        counts = {}\n",
    "        for class_dir in sorted(split_dir.iterdir()):\n",
    "            if class_dir.is_dir():\n",
    "                counts[class_dir.name] = sum(1 for _ in class_dir.glob('*') if _.is_file())\n",
    "        layout[split] = counts\n",
    "    return layout\n",
    "\n",
    "\n",
    "def build_comparison_frame(artifacts_root):\n",
    "    rows = []\n",
    "    for model_type in MODEL_TYPES:\n",
    "        metrics_file = artifacts_root / f'{model_type}_evaluation_metrics.json'\n",
    "        if metrics_file.exists():\n",
    "            metrics = load_metrics_json(metrics_file)\n",
    "            report = metrics.get('classification_report', {})\n",
    "            weighted = report.get('weighted avg', report.get('weighted_avg', {})) if isinstance(report, dict) else {}\n",
    "            rows.append({\n",
    "                'model': model_type,\n",
    "                'accuracy': metrics.get('accuracy', report.get('accuracy')),\n",
    "                'precision': weighted.get('precision'),\n",
    "                'recall': weighted.get('recall'),\n",
    "                'f1_score': weighted.get('f1-score', weighted.get('f1_score')),\n",
    "                'roc_auc': metrics.get('roc_auc'),\n",
    "            })\n",
    "    return pd.DataFrame(rows)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9e9a0986",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Dataset layout inspection\n",
    "layout = inspect_dataset_structure(DATASET_ROOT)\n",
    "print('Dataset root:', DATASET_ROOT.resolve())\n",
    "for split, classes in layout.items():\n",
    "    if classes is None:\n",
    "        print(f'{split}: MISSING')\n",
    "    elif not classes:\n",
    "        print(f'{split}: EMPTY')\n",
    "    else:\n",
    "        print(f'{split}:')\n",
    "        for class_name, count in classes.items():\n",
    "            print(f'  - {class_name}: {count}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5e00b8a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Compare performance across models (if metrics exist)\n",
    "comparison_df = build_comparison_frame(ARTIFACTS_ROOT)\n",
    "if comparison_df.empty:\n",
    "    print('No evaluation metrics found in artifacts directory yet.')\n",
    "else:\n",
    "    display(comparison_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6b065548",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load a training history file if available\n",
    "history_files = sorted((ARTIFACTS_ROOT / 'cnn').glob('history_*.npz'))\n",
    "if history_files:\n",
    "    history = load_history_npz(history_files[-1])\n",
    "    print('Loaded history file:', history_files[-1])\n",
    "    for metric_name, values in history.items():\n",
    "        print(metric_name, len(values))\n",
    "else:\n",
    "    print('No history file found for CNN.')"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}