{
  "id": 79444,
  "title": "What's your best single model?",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/79444",
  "author_name": "",
  "post_date": "2019-02-04T04:55:30.924156900Z",
  "votes": 18,
  "comment_count": 4,
  "views": 0,
  "content": "<p>My model is basic LSTM with 0.714.. I think I overfitted the LB. What's yours?</p>",
  "messages": [
    {
      "id": "465825",
      "postDate": "02/04/2019 04:55:30",
      "content": "<p>My model is basic LSTM with 0.714.. I think I overfitted the LB. What's yours?</p>",
      "rawMarkdown": "My model is basic LSTM with 0.714.. I think I overfitted the LB. What's yours?",
      "votes": null
    },
    {
      "id": "469987",
      "postDate": "02/12/2019 07:14:47",
      "content": "<p>I have tried some DTFT with LSTM, got score 0.519,  not very well</p>",
      "rawMarkdown": "I have tried some DTFT with LSTM, got score 0.519,  not very well",
      "votes": null
    },
    {
      "id": "478079",
      "postDate": "02/25/2019 17:43:23",
      "content": "<p>I am getting 0.699 LB with the same features as <a href=\"https://www.kaggle.com/braquino/5-fold-lstm-attention-fully-commented-0-694\">this kernel</a>, but with a completely different architecture. I have also engineered few extra signal features. My CV at around 0.754 looks pretty promising, although I am worried about the randomness of keras CuDNNLSTM and CuDNNGRU.</p>",
      "rawMarkdown": "I am getting 0.699 LB with the same features as [this kernel](https://www.kaggle.com/braquino/5-fold-lstm-attention-fully-commented-0-694), but with a completely different architecture. I have also engineered few extra signal features. My CV at around 0.754 looks pretty promising, although I am worried about the randomness of keras CuDNNLSTM and CuDNNGRU.",
      "votes": null
    },
    {
      "id": "478643",
      "postDate": "02/26/2019 12:41:52",
      "content": "<p>Have you been able to get the same/similar score twice? :)</p>",
      "rawMarkdown": "Have you been able to get the same/similar score twice? :)",
      "votes": null
    },
    {
      "id": "482276",
      "postDate": "03/02/2019 16:12:39",
      "content": "<p>It seems your model already achieves a good score</p>",
      "rawMarkdown": "It seems your model already achieves a good score",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 469987,
      "author_name": "simonnada",
      "author_url": "",
      "post_date": "02/12/2019 07:14:47",
      "content": "<p>I have tried some DTFT with LSTM, got score 0.519,  not very well</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 478079,
      "author_name": "tarunpaparaju",
      "author_url": "",
      "post_date": "02/25/2019 17:43:23",
      "content": "<p>I am getting 0.699 LB with the same features as <a href=\"https://www.kaggle.com/braquino/5-fold-lstm-attention-fully-commented-0-694\">this kernel</a>, but with a completely different architecture. I have also engineered few extra signal features. My CV at around 0.754 looks pretty promising, although I am worried about the randomness of keras CuDNNLSTM and CuDNNGRU.</p>",
      "votes": null,
      "replies": [
        {
          "id": 478643,
          "author_name": "mpekalski",
          "author_url": "",
          "post_date": "02/26/2019 12:41:52",
          "content": "<p>Have you been able to get the same/similar score twice? :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 482276,
      "author_name": "strideradu",
      "author_url": "",
      "post_date": "03/02/2019 16:12:39",
      "content": "<p>It seems your model already achieves a good score</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "465825": "My model is basic LSTM with 0.714.. I think I overfitted the LB. What's yours?",
    "469987": "I have tried some DTFT with LSTM, got score 0.519,  not very well",
    "478079": "I am getting 0.699 LB with the same features as [this kernel](https://www.kaggle.com/braquino/5-fold-lstm-attention-fully-commented-0-694), but with a completely different architecture. I have also engineered few extra signal features. My CV at around 0.754 looks pretty promising, although I am worried about the randomness of keras CuDNNLSTM and CuDNNGRU.",
    "478643": "Have you been able to get the same/similar score twice? :)",
    "482276": "It seems your model already achieves a good score"
  },
  "source": "meta"
}