{
  "id": 202287,
  "title": "ROC-AUC dependence from number of previous steps",
  "url": "/competitions/riiid-test-answer-prediction/discussion/202287",
  "author_name": "Elena K.",
  "post_date": "2020-12-09T08:51:54.730000",
  "votes": 8,
  "comment_count": 7,
  "views": 0,
  "content": "<p>All time I have been working with that challenge I tried to use more data.<br>\nBut yesterday I decided to build roc-auc dependence curve from number of previous steps.<br>\nThat's what I got. Probably it may be helpful for someone.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F572545%2Faf2702fdce4e4bb640b0e6acc55adf82%2Fdependency.png?generation=1607503958423796&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 1106953,
      "postDate": "2020-12-09T08:51:54.730Z",
      "content": "<p>All time I have been working with that challenge I tried to use more data.<br>\nBut yesterday I decided to build roc-auc dependence curve from number of previous steps.<br>\nThat's what I got. Probably it may be helpful for someone.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F572545%2Faf2702fdce4e4bb640b0e6acc55adf82%2Fdependency.png?generation=1607503958423796&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "All time I have been working with that challenge I tried to use more data.\nBut yesterday I decided to build roc-auc dependence curve from number of previous steps.\nThat's what I got. Probably it may be helpful for someone.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F572545%2Faf2702fdce4e4bb640b0e6acc55adf82%2Fdependency.png?generation=1607503958423796&alt=media)\n\n",
      "votes": 7
    },
    {
      "id": 1110440,
      "postDate": "2020-12-12T18:28:32.930Z",
      "content": "<p>great work man ! keep it up</p>",
      "rawMarkdown": "great work man ! keep it up",
      "votes": 1
    },
    {
      "id": 1107109,
      "postDate": "2020-12-09T11:56:39.867Z",
      "content": "<p>Hello thanks for sharing. Can you elaborate this analysis here? With just two last step you achieve 79 auc?</p>",
      "rawMarkdown": "Hello thanks for sharing. Can you elaborate this analysis here? With just two last step you achieve 79 auc?",
      "replies": [
        {
          "id": 1107112,
          "postDate": "2020-12-09T12:01:56.003Z",
          "content": "<p>It's not finished now therefore It's quite raw. But I got this on my valid set includes 10 last step of the dataset. And the train dataset is the previous steps before the valid</p>",
          "rawMarkdown": "It's not finished now therefore It's quite raw. But I got this on my valid set includes 10 last step of the dataset. And the train dataset is the previous steps before the valid"
        },
        {
          "id": 1107120,
          "postDate": "2020-12-09T12:08:36.523Z",
          "content": "<p>I understad. thank you for explanation. In a more realistic cross validation environment, and with a bigger data, it may fail you.  I tried to featurize last 30 content information and last 30 target information and it failed me.</p>",
          "rawMarkdown": "I understad. thank you for explanation. In a more realistic cross validation environment, and with a bigger data, it may fail you.  I tried to featurize last 30 content information and last 30 target information and it failed me."
        },
        {
          "id": 1107182,
          "postDate": "2020-12-09T13:15:58.700Z",
          "content": "<p>Yes, you're totally right. I try to adjust more appropriate cv</p>",
          "rawMarkdown": "Yes, you're totally right. I try to adjust more appropriate cv"
        }
      ]
    },
    {
      "id": 1108022,
      "postDate": "2020-12-10T07:33:33.513Z",
      "rawMarkdown": "",
      "isDeleted": true,
      "replies": [
        {
          "id": 1108031,
          "postDate": "2020-12-10T07:47:20.390Z",
          "content": "<p>user interaction (question answering)</p>",
          "rawMarkdown": "user interaction (question answering)"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1110440,
      "author_name": "Nitesh Ranjan Singh",
      "author_url": "",
      "post_date": "2020-12-12T18:28:32.930000",
      "content": "<p>great work man ! keep it up</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1107109,
      "author_name": "bturan19",
      "author_url": "",
      "post_date": "2020-12-09T11:56:39.867000",
      "content": "<p>Hello thanks for sharing. Can you elaborate this analysis here? With just two last step you achieve 79 auc?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1107112,
          "author_name": "Elena K.",
          "author_url": "",
          "post_date": "2020-12-09T12:01:56.003000",
          "content": "<p>It's not finished now therefore It's quite raw. But I got this on my valid set includes 10 last step of the dataset. And the train dataset is the previous steps before the valid</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1107120,
          "author_name": "bturan19",
          "author_url": "",
          "post_date": "2020-12-09T12:08:36.523000",
          "content": "<p>I understad. thank you for explanation. In a more realistic cross validation environment, and with a bigger data, it may fail you.  I tried to featurize last 30 content information and last 30 target information and it failed me.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1107182,
          "author_name": "Elena K.",
          "author_url": "",
          "post_date": "2020-12-09T13:15:58.700000",
          "content": "<p>Yes, you're totally right. I try to adjust more appropriate cv</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1108022,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-12-10T07:33:33.513000",
      "content": "",
      "votes": 0,
      "replies": [
        {
          "id": 1108031,
          "author_name": "Elena K.",
          "author_url": "",
          "post_date": "2020-12-10T07:47:20.390000",
          "content": "<p>user interaction (question answering)</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1106953": "All time I have been working with that challenge I tried to use more data.\nBut yesterday I decided to build roc-auc dependence curve from number of previous steps.\nThat's what I got. Probably it may be helpful for someone.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F572545%2Faf2702fdce4e4bb640b0e6acc55adf82%2Fdependency.png?generation=1607503958423796&alt=media)\n\n",
    "1110440": "great work man ! keep it up",
    "1107109": "Hello thanks for sharing. Can you elaborate this analysis here? With just two last step you achieve 79 auc?",
    "1108022": ""
  }
}