{
  "id": 236703,
  "title": "What kind of cross-validation should be done for this competition? ",
  "url": "/competitions/birdclef-2021/discussion/236703",
  "author_name": "udon",
  "post_date": "2021-05-05T13:39:58.114000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I will ask the question in two languages, English and Japanese.</p>\n<p>What kind of cross-validation should be done for this competition? I don't think the method of dividing the data into 10 pieces, training with 9 pieces, and testing with the remaining one, and then taking the average of the 10 pieces, as shown in the table, will give an appropriate evaluation value this time. This is because we think that if the test data contains birds that have not been learned from the training data, we will not be able to make a correct judgment.</p>\n<p>今回のコンペでは、どのような交差検証を行えばいいのでしょうか。表形式のようにデータを10個に分けて、9個で学習し残りの一つでテストするというのを、10回行い、その平均をとるといった手法は今回適切な評価値をだせないと思います。なぜなら、訓練データから学習していない鳥がテストデータに含まれているとき、正しく判断できないと考えるからです。</p>",
  "messages": [
    {
      "id": 1294221,
      "postDate": "2021-05-05T13:39:58.113Z",
      "content": "<p>I will ask the question in two languages, English and Japanese.</p>\n<p>What kind of cross-validation should be done for this competition? I don't think the method of dividing the data into 10 pieces, training with 9 pieces, and testing with the remaining one, and then taking the average of the 10 pieces, as shown in the table, will give an appropriate evaluation value this time. This is because we think that if the test data contains birds that have not been learned from the training data, we will not be able to make a correct judgment.</p>\n<p>今回のコンペでは、どのような交差検証を行えばいいのでしょうか。表形式のようにデータを10個に分けて、9個で学習し残りの一つでテストするというのを、10回行い、その平均をとるといった手法は今回適切な評価値をだせないと思います。なぜなら、訓練データから学習していない鳥がテストデータに含まれているとき、正しく判断できないと考えるからです。</p>",
      "rawMarkdown": "I will ask the question in two languages, English and Japanese.\n\nWhat kind of cross-validation should be done for this competition? I don't think the method of dividing the data into 10 pieces, training with 9 pieces, and testing with the remaining one, and then taking the average of the 10 pieces, as shown in the table, will give an appropriate evaluation value this time. This is because we think that if the test data contains birds that have not been learned from the training data, we will not be able to make a correct judgment.\n\n今回のコンペでは、どのような交差検証を行えばいいのでしょうか。表形式のようにデータを10個に分けて、9個で学習し残りの一つでテストするというのを、10回行い、その平均をとるといった手法は今回適切な評価値をだせないと思います。なぜなら、訓練データから学習していない鳥がテストデータに含まれているとき、正しく判断できないと考えるからです。",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1294221": "I will ask the question in two languages, English and Japanese.\n\nWhat kind of cross-validation should be done for this competition? I don't think the method of dividing the data into 10 pieces, training with 9 pieces, and testing with the remaining one, and then taking the average of the 10 pieces, as shown in the table, will give an appropriate evaluation value this time. This is because we think that if the test data contains birds that have not been learned from the training data, we will not be able to make a correct judgment.\n\n今回のコンペでは、どのような交差検証を行えばいいのでしょうか。表形式のようにデータを10個に分けて、9個で学習し残りの一つでテストするというのを、10回行い、その平均をとるといった手法は今回適切な評価値をだせないと思います。なぜなら、訓練データから学習していない鳥がテストデータに含まれているとき、正しく判断できないと考えるからです。"
  }
}