{
  "id": 73365,
  "title": "any improvement with using the external data?",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/73365",
  "author_name": "",
  "post_date": "2018-12-02T11:31:04.228699600Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>anyone who use the external data to imporve the classification result? I use the external data in the website: <a href=\"https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv\">https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv</a> .  But did not to improve, who can give some tips?</p>",
  "messages": [
    {
      "id": "431487",
      "postDate": "12/02/2018 11:31:04",
      "content": "<p>anyone who use the external data to imporve the classification result? I use the external data in the website: <a href=\"https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv\">https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv</a> .  But did not to improve, who can give some tips?</p>",
      "rawMarkdown": "anyone who use the external data to imporve the classification result? I use the external data in the website: https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv .  But did not to improve, who can give some tips?",
      "votes": null
    },
    {
      "id": "432920",
      "postDate": "12/04/2018 13:21:13",
      "content": "<p>I also add external data into the train set and I found that my CV score improve but the LB score become worse. It didn't make sane at all cuz external data have 10000 more images and the original data only have about 30000 images. I guess that maybe the labels in external data's are different from original data set so that's why the LB score become even worse. The next step I would try is to use external data to pretrain the model and then use original data to train.</p>",
      "rawMarkdown": "I also add external data into the train set and I found that my CV score improve but the LB score become worse. It didn't make sane at all cuz external data have 10000 more images and the original data only have about 30000 images. I guess that maybe the labels in external data's are different from original data set so that's why the LB score become even worse. The next step I would try is to use external data to pretrain the model and then use original data to train.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 432920,
      "author_name": "johnnyjana730",
      "author_url": "",
      "post_date": "12/04/2018 13:21:13",
      "content": "<p>I also add external data into the train set and I found that my CV score improve but the LB score become worse. It didn't make sane at all cuz external data have 10000 more images and the original data only have about 30000 images. I guess that maybe the labels in external data's are different from original data set so that's why the LB score become even worse. The next step I would try is to use external data to pretrain the model and then use original data to train.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "431487": "anyone who use the external data to imporve the classification result? I use the external data in the website: https://www.kaggle.com/artemtprv/external-data-for-protein-atlas#subcellular_location.tsv .  But did not to improve, who can give some tips?",
    "432920": "I also add external data into the train set and I found that my CV score improve but the LB score become worse. It didn't make sane at all cuz external data have 10000 more images and the original data only have about 30000 images. I guess that maybe the labels in external data's are different from original data set so that's why the LB score become even worse. The next step I would try is to use external data to pretrain the model and then use original data to train."
  },
  "source": "meta"
}