{
  "id": 207557,
  "title": "Be careful for history with only lectures!",
  "url": "/competitions/riiid-test-answer-prediction/discussion/207557",
  "author_name": "nadare",
  "post_date": "2020-12-30T08:08:23.045000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I used a feature that was the sum of the average accuracy rates of the questions divided by the number of questions. </p>\n<p>However, since the numerator that filled the defect due to the lecture with the average correct answer rate was divided by the total number of problems excluding the lecture + eps, the ratio became very large and damaged the model. </p>\n<pre><code>my_wrong_feature = mean_accuracy_history.fillna(accuracy_mean).sum() / sum_of_question\n</code></pre>\n<p>I took a few days to fix this issue. I will share this failure with you to have fun fighting the last few days of this competition.</p>",
  "messages": [
    {
      "id": 1132163,
      "postDate": "2020-12-30T08:08:23.047Z",
      "content": "<p>I used a feature that was the sum of the average accuracy rates of the questions divided by the number of questions. </p>\n<p>However, since the numerator that filled the defect due to the lecture with the average correct answer rate was divided by the total number of problems excluding the lecture + eps, the ratio became very large and damaged the model. </p>\n<pre><code>my_wrong_feature = mean_accuracy_history.fillna(accuracy_mean).sum() / sum_of_question\n</code></pre>\n<p>I took a few days to fix this issue. I will share this failure with you to have fun fighting the last few days of this competition.</p>",
      "rawMarkdown": "I used a feature that was the sum of the average accuracy rates of the questions divided by the number of questions. \n\nHowever, since the numerator that filled the defect due to the lecture with the average correct answer rate was divided by the total number of problems excluding the lecture + eps, the ratio became very large and damaged the model. \n\n```\nmy_wrong_feature = mean_accuracy_history.fillna(accuracy_mean).sum() / sum_of_question\n```\n\nI took a few days to fix this issue. I will share this failure with you to have fun fighting the last few days of this competition.",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1132163": "I used a feature that was the sum of the average accuracy rates of the questions divided by the number of questions. \n\nHowever, since the numerator that filled the defect due to the lecture with the average correct answer rate was divided by the total number of problems excluding the lecture + eps, the ratio became very large and damaged the model. \n\n```\nmy_wrong_feature = mean_accuracy_history.fillna(accuracy_mean).sum() / sum_of_question\n```\n\nI took a few days to fix this issue. I will share this failure with you to have fun fighting the last few days of this competition."
  }
}