{
  "id": 166820,
  "title": "Multi-label dataset",
  "url": "/competitions/birdsong-recognition/discussion/166820",
  "author_name": "hawkey",
  "post_date": "2020-07-14T06:25:32.506000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I made a notebook with multi-label dataset in pytorch: \n<a href=\"https://www.kaggle.com/hawkey/birdsong-multi-label-dataset\">https://www.kaggle.com/hawkey/birdsong-multi-label-dataset</a></p>\n\n<p>Extra labels are extracted from <code>secondary_labels</code> or <code>background</code> meta-data in <code>train.csv</code>. </p>\n\n<p>Since the test dataset is multi-labeled, maybe training with multi-labels can help. But the main problem for this competition is noise and the exact time range corresponding to each label, so this may not be that helpful.  </p>",
  "messages": [
    {
      "id": 928670,
      "postDate": "2020-07-14T06:25:32.507Z",
      "content": "<p>I made a notebook with multi-label dataset in pytorch: \n<a href=\"https://www.kaggle.com/hawkey/birdsong-multi-label-dataset\">https://www.kaggle.com/hawkey/birdsong-multi-label-dataset</a></p>\n\n<p>Extra labels are extracted from <code>secondary_labels</code> or <code>background</code> meta-data in <code>train.csv</code>. </p>\n\n<p>Since the test dataset is multi-labeled, maybe training with multi-labels can help. But the main problem for this competition is noise and the exact time range corresponding to each label, so this may not be that helpful.  </p>",
      "rawMarkdown": "I made a notebook with multi-label dataset in pytorch: \nhttps://www.kaggle.com/hawkey/birdsong-multi-label-dataset\n\nExtra labels are extracted from `secondary_labels` or `background` meta-data in `train.csv`. \n\nSince the test dataset is multi-labeled, maybe training with multi-labels can help. But the main problem for this competition is noise and the exact time range corresponding to each label, so this may not be that helpful.  ",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "928670": "I made a notebook with multi-label dataset in pytorch: \nhttps://www.kaggle.com/hawkey/birdsong-multi-label-dataset\n\nExtra labels are extracted from `secondary_labels` or `background` meta-data in `train.csv`. \n\nSince the test dataset is multi-labeled, maybe training with multi-labels can help. But the main problem for this competition is noise and the exact time range corresponding to each label, so this may not be that helpful.  "
  }
}