{
  "id": 78422,
  "title": "Single, double, and triple phase faults",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/78422",
  "author_name": "Paul Nussbaum, PhD",
  "post_date": "2019-01-23T16:17:23.168000",
  "votes": 7,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Since you can only have two submissions per day, it's nice to be able to have a \"feel\" for whether or not your latest notebook is on the right track. In my most recent public notebook I include a metric showing how many times faults are found on all three phases, just two of the phases, or just one of the three phases. </p>\n\n<p>Of the total training set of 8712 snapshots, there are 525 total fault examples (just over 6%)\nOf those, faults occur on 3, 2, or only one of the phases as follows:\n - Triples 156 (89% of the time) \n - Doubles 19 (7% of the time)\n - Singles 19 (4% of the time)</p>\n\n<p>Although I can't be certain, it would be make sense that the TEST results would have similar ratios, so you can look for that in your notebook before submitting.</p>\n\n<p>Here is a code snippet used to calculate that (basically counting by threes through the data):</p>\n\n<p>triples = 0\ndoubles = 0\nsingles = 0\nfor i in range(0,num_results,3) : \n    if (meta_train.target[i] and meta_train.target[i+1] and meta_train.target[i+2] ):\n        triples = triples + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 2):\n        doubles = doubles + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 1):\n        singles = singles + 1</p>\n\n<p>print('triples', triples, 'doubles', doubles, 'singles', singles)\nprint('sanity check: ', 'total faults', meta_train.target[0:int(num_results)].sum(), ' sum of above ', 3 * triples + 2 * doubles + singles)</p>\n\n<p>Here is the public notebook using that metric...\n<a href=\"https://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09\">https://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09</a> </p>",
  "messages": [
    {
      "id": 460401,
      "postDate": "2019-01-23T16:17:23.170Z",
      "content": "<p>Since you can only have two submissions per day, it's nice to be able to have a \"feel\" for whether or not your latest notebook is on the right track. In my most recent public notebook I include a metric showing how many times faults are found on all three phases, just two of the phases, or just one of the three phases. </p>\n\n<p>Of the total training set of 8712 snapshots, there are 525 total fault examples (just over 6%)\nOf those, faults occur on 3, 2, or only one of the phases as follows:\n - Triples 156 (89% of the time) \n - Doubles 19 (7% of the time)\n - Singles 19 (4% of the time)</p>\n\n<p>Although I can't be certain, it would be make sense that the TEST results would have similar ratios, so you can look for that in your notebook before submitting.</p>\n\n<p>Here is a code snippet used to calculate that (basically counting by threes through the data):</p>\n\n<p>triples = 0\ndoubles = 0\nsingles = 0\nfor i in range(0,num_results,3) : \n    if (meta_train.target[i] and meta_train.target[i+1] and meta_train.target[i+2] ):\n        triples = triples + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 2):\n        doubles = doubles + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 1):\n        singles = singles + 1</p>\n\n<p>print('triples', triples, 'doubles', doubles, 'singles', singles)\nprint('sanity check: ', 'total faults', meta_train.target[0:int(num_results)].sum(), ' sum of above ', 3 * triples + 2 * doubles + singles)</p>\n\n<p>Here is the public notebook using that metric...\n<a href=\"https://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09\">https://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09</a> </p>",
      "rawMarkdown": "Since you can only have two submissions per day, it's nice to be able to have a \"feel\" for whether or not your latest notebook is on the right track. In my most recent public notebook I include a metric showing how many times faults are found on all three phases, just two of the phases, or just one of the three phases. \n\nOf the total training set of 8712 snapshots, there are 525 total fault examples (just over 6%)\nOf those, faults occur on 3, 2, or only one of the phases as follows:\n - Triples 156 (89% of the time) \n - Doubles 19 (7% of the time)\n - Singles 19 (4% of the time)\n\nAlthough I can't be certain, it would be make sense that the TEST results would have similar ratios, so you can look for that in your notebook before submitting.\n \nHere is a code snippet used to calculate that (basically counting by threes through the data):\n\ntriples = 0\ndoubles = 0\nsingles = 0\nfor i in range(0,num_results,3) : \n    if (meta_train.target[i] and meta_train.target[i+1] and meta_train.target[i+2] ):\n        triples = triples + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 2):\n        doubles = doubles + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 1):\n        singles = singles + 1\n\nprint('triples', triples, 'doubles', doubles, 'singles', singles)\nprint('sanity check: ', 'total faults', meta_train.target[0:int(num_results)].sum(), ' sum of above ', 3 * triples + 2 * doubles + singles)\n\nHere is the public notebook using that metric...\nhttps://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09 ",
      "votes": 7
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "460401": "Since you can only have two submissions per day, it's nice to be able to have a \"feel\" for whether or not your latest notebook is on the right track. In my most recent public notebook I include a metric showing how many times faults are found on all three phases, just two of the phases, or just one of the three phases. \n\nOf the total training set of 8712 snapshots, there are 525 total fault examples (just over 6%)\nOf those, faults occur on 3, 2, or only one of the phases as follows:\n - Triples 156 (89% of the time) \n - Doubles 19 (7% of the time)\n - Singles 19 (4% of the time)\n\nAlthough I can't be certain, it would be make sense that the TEST results would have similar ratios, so you can look for that in your notebook before submitting.\n \nHere is a code snippet used to calculate that (basically counting by threes through the data):\n\ntriples = 0\ndoubles = 0\nsingles = 0\nfor i in range(0,num_results,3) : \n    if (meta_train.target[i] and meta_train.target[i+1] and meta_train.target[i+2] ):\n        triples = triples + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 2):\n        doubles = doubles + 1\n    elif (meta_train.target[i] + meta_train.target[i+1] + meta_train.target[i+2] == 1):\n        singles = singles + 1\n\nprint('triples', triples, 'doubles', doubles, 'singles', singles)\nprint('sanity check: ', 'total faults', meta_train.target[0:int(num_results)].sum(), ' sum of above ', 3 * triples + 2 * doubles + singles)\n\nHere is the public notebook using that metric...\nhttps://www.kaggle.com/pnussbaum/vsb-power-using-autoencoding-v09 "
  }
}