{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# RoboX Egotask Dataset Explorer\n\nExplore the **RoboX-EgoTask-v1.7** bundle — egocentric dataset.  \nThis notebook is auto-generated for this specific campaign. Run all cells to get a quick overview."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\nimport json, sys, random\nfrom collections import Counter\n\nDATASET_ROOT = Path(\"..\")  # bundle root\nCAMPAIGN_TYPE = \"ego_task\"\n\nif not DATASET_ROOT.exists():\n    sys.exit(f\"Dataset not found: {DATASET_ROOT}\")\n\nclips = [json.loads(l) for l in (DATASET_ROOT / 'metadata' / 'clips.jsonl').read_text().splitlines() if l.strip()]\nsplits = {}\nfor c in clips:\n    splits.setdefault(c.get('split', 'unknown'), []).append(c)\n\nstats = json.loads((DATASET_ROOT / 'stats.json').read_text())\nmanifest = json.loads((DATASET_ROOT / 'manifest.json').read_text())\n\nprint(f\"Dataset     : {manifest['dataset_name']} v{manifest['schema_version']}\")\nprint(f\"Campaign    : {manifest['campaign']}\")\nprint(f\"Profile     : {manifest['dataset_profile']}\")\nprint(f\"Clips       : {stats['total_clips']}  |  Recordings: {stats['total_recordings']}\")\nprint(f\"Duration    : {stats['total_duration_sec']:.1f}s\")\nprint(f\"Contributors: {stats['unique_contributors']}\")\nprint(f\"Splits      : { {k: len(v) for k, v in splits.items()} }\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sensor Tier Distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tiers = Counter(c.get('sensor_tier') for c in clips)\nprint(\"Sensor tier distribution:\")\nfor tier, count in sorted(tiers.items(), key=lambda x: -x[1]):\n    pct = 100 * count / len(clips)\n    bar = '█' * count\n    print(f\"  {tier:15s}: {bar} {count} clips ({pct:.1f}%)\")\n\ntier_cov = stats.get('annotation_coverage_by_sensor_tier', {})\nif tier_cov:\n    print()\n    print(\"Annotation coverage by tier:\")\n    for tier, values in tier_cov.items():\n        cov_summary = ', '.join(f\"{k}={v['percent']:.0f}%\" for k, v in values['coverage'].items())\n        print(f\"  [{tier}] ({values['total']} clips): {cov_summary}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Object / Domain Distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "obj_counts = Counter(\n    c['labels'].get('object') for c in clips if c.get('labels', {}).get('object')\n)\ndom_counts = Counter(\n    c['labels'].get('object_domain') for c in clips if c.get('labels', {}).get('object_domain')\n)\nif obj_counts:\n    print('Object distribution:')\n    for label, count in obj_counts.most_common():\n        print(f'  {label:25s}: {count}')\nif dom_counts:\n    print()\n    print('Domain distribution:')\n    for dom, count in dom_counts.most_common():\n        print(f'  {dom:25s}: {count}')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Clip Duration Distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "dd = stats.get('duration_distribution', {})\nprint(\"Duration stats:\")\nprint(f\"  min    : {dd.get('min_sec')} s\")\nprint(f\"  max    : {dd.get('max_sec')} s\")\nprint(f\"  mean   : {dd.get('mean_sec')} s\")\nprint(f\"  median : {dd.get('median_sec')} s\")\n\nbuckets = [0, 2, 4, 6, 10, 20, 999]\nbucket_labels = ['<2s', '2-4s', '4-6s', '6-10s', '10-20s', '>20s']\nbucket_counts = [0] * 6\nfor d in [c['duration_sec'] for c in clips]:\n    for i in range(len(buckets) - 1):\n        if buckets[i] <= d < buckets[i + 1]:\n            bucket_counts[i] += 1\n            break\nprint()\nprint(\"Distribution:\")\nfor label, count in zip(bucket_labels, bucket_counts):\n    bar = '█' * count\n    print(f\"  {label:6s}: {bar} ({count})\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Object Tracks — temporal density and in_hand"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "track_dir = DATASET_ROOT / 'annotations' / 'object_tracks'\ntrack_files = sorted(track_dir.glob('*.jsonl'))\ncandidates = []\nfor tf in track_files:\n    lines = [l for l in tf.read_text().splitlines() if l.strip()]\n    candidates.append((len(lines), tf))\ncandidates.sort(reverse=True)\nprint(\"Top 5 clips by object track density:\")\nfor count, path in candidates[:5]:\n    print(f\"  {path.stem}: {count} observations\")\nif candidates:\n    best = candidates[0][1]\n    lines = [json.loads(l) for l in best.read_text().splitlines() if l.strip()]\n    print(f\"\\nFirst 3 observations from {best.stem}:\")\n    for obs in lines[:3]:\n        for o in obs.get('objects', []):\n            print(f\"  frame={obs.get('frame')} ts={obs.get('timestamp_sec'):.3f}s track_id={o.get('track_id')} label={o.get('class_label')} in_hand={o.get('in_hand')}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sample Narrations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample = [c for c in clips if c.get('narration')]\nrandom.shuffle(sample)\nprint(\"5 random clips with narration:\")\nfor c in sample[:5]:\n    gs = c.get('grip_summary')\n    print(f\"  [{c['split']}] {c['clip_id']}  tier={c['sensor_tier']}\")\n    print(f\"    {c['narration'][:100]}\")\n    if gs:\n        print(f\"    grip: {gs['dominant_grip']} hand={gs['grip_hand']}\")\n    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Showcase Clips"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "showcase_path = DATASET_ROOT / 'showcase' / 'best_clips.json'\nif showcase_path.exists():\n    showcase = json.loads(showcase_path.read_text())\n    print(f\"Showcase: {showcase.get('dataset')}\")\n    for clip in showcase.get('clips', []):\n        print(f\"  [{clip['split']}] {clip['clip_id']}  tier={clip['sensor_tier']}\")\n        if clip.get('labels'):\n            print(f\"    labels: {clip['labels']}\")\n        if clip.get('narration'):\n            print(f\"    {clip['narration'][:80]}\")\n        print()\nelse:\n    print(\"No showcase file found.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Validation & Sanitization Status"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "val = json.loads((DATASET_ROOT / 'VALIDATION_REPORT.json').read_text())\nsan = json.loads((DATASET_ROOT / 'SANITIZATION_REPORT.json').read_text())\nprint(\"Validation:\", '✅ PASSED' if val['passed'] else '❌ FAILED')\nprint(f\"  errors  : {val['summary']['error_count']}\")\nprint(f\"  warnings: {val['summary']['warning_count']}\")\nprint()\nprint(\"Sanitization:\", '✅ PASSED' if san['passed'] else '❌ FAILED')\nprint(f\"  findings: {san['summary']['finding_count']}\")\nprint()\nprint(\"Privacy review:\", '✅' if manifest.get('privacy_review_passed') else '❌')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.10.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}