{
  "id": 80664,
  "title": "125th solution. 12folds average",
  "url": "/competitions/quora-insincere-questions-classification/discussion/80664",
  "author_name": "Tamaki",
  "post_date": "2019-02-15T08:24:58.560000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Here is our solution based on Benjamin's great kernel.</p>\n\n<p><a href=\"https://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed\">https://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed</a></p>\n\n<p><a href=\"https://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245\">https://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245</a></p>\n\n<p>These are the main differences.</p>\n\n<ul>\n<li>Use maxlen = 35, n_folds = 12</li>\n<li>Make 2 datasets by truncating like this.</li>\n</ul>\n\n<p><code>\nx_train_pre = pad_sequences(x_train, maxlen=maxlen, truncating='pre')\n</code>\n<code>\nx_train_post = pad_sequences(x_train, maxlen=maxlen, truncating='post')\n</code></p>\n\n<ul>\n<li>At fitting, we use 1 of 2 alternately. </li>\n<li>At predicting, we use both of them and average the predictions.</li>\n</ul>\n\n<p>CV: 0.694 -&gt; 0.701</p>\n\n<p>Public Leaderboard: 0.696 -&gt; 0.703</p>",
  "messages": [
    {
      "id": 472024,
      "postDate": "2019-02-15T08:24:58.560Z",
      "content": "<p>Here is our solution based on Benjamin's great kernel.</p>\n\n<p><a href=\"https://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed\">https://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed</a></p>\n\n<p><a href=\"https://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245\">https://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245</a></p>\n\n<p>These are the main differences.</p>\n\n<ul>\n<li>Use maxlen = 35, n_folds = 12</li>\n<li>Make 2 datasets by truncating like this.</li>\n</ul>\n\n<p><code>\nx_train_pre = pad_sequences(x_train, maxlen=maxlen, truncating='pre')\n</code>\n<code>\nx_train_post = pad_sequences(x_train, maxlen=maxlen, truncating='post')\n</code></p>\n\n<ul>\n<li>At fitting, we use 1 of 2 alternately. </li>\n<li>At predicting, we use both of them and average the predictions.</li>\n</ul>\n\n<p>CV: 0.694 -&gt; 0.701</p>\n\n<p>Public Leaderboard: 0.696 -&gt; 0.703</p>",
      "rawMarkdown": "Here is our solution based on Benjamin's great kernel.\n\nhttps://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed\n\nhttps://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245\n\nThese are the main differences.\n\n* Use maxlen = 35, n_folds = 12\n* Make 2 datasets by truncating like this.\n\n```\nx_train_pre = pad_sequences(x_train, maxlen=maxlen, truncating='pre')\n```\n```\nx_train_post = pad_sequences(x_train, maxlen=maxlen, truncating='post')\n```\n\n* At fitting, we use 1 of 2 alternately. \n* At predicting, we use both of them and average the predictions.\n\nCV: 0.694 -&gt; 0.701\n\nPublic Leaderboard: 0.696 -&gt; 0.703",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "472024": "Here is our solution based on Benjamin's great kernel.\n\nhttps://www.kaggle.com/bminixhofer/a-validation-framework-impact-of-the-random-seed\n\nhttps://www.kaggle.com/decoflight/tranc-submit?scriptVersionId=10104245\n\nThese are the main differences.\n\n* Use maxlen = 35, n_folds = 12\n* Make 2 datasets by truncating like this.\n\n```\nx_train_pre = pad_sequences(x_train, maxlen=maxlen, truncating='pre')\n```\n```\nx_train_post = pad_sequences(x_train, maxlen=maxlen, truncating='post')\n```\n\n* At fitting, we use 1 of 2 alternately. \n* At predicting, we use both of them and average the predictions.\n\nCV: 0.694 -&gt; 0.701\n\nPublic Leaderboard: 0.696 -&gt; 0.703"
  }
}