{
  "id": 68663,
  "title": "Good metric for training for competition",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/68663",
  "author_name": "",
  "post_date": "2018-10-15T18:14:21.832820400Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I'm having some trouble picking training metric for the competition. I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?</p>",
  "messages": [
    {
      "id": "404455",
      "postDate": "10/15/2018 18:14:21",
      "content": "<p>I'm having some trouble picking training metric for the competition. I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?</p>",
      "rawMarkdown": "I'm having some trouble picking training metric for the competition. I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?",
      "votes": null
    },
    {
      "id": "404488",
      "postDate": "10/15/2018 19:54:37",
      "content": "<p>Are you looking for a good loss function? Categorical crossentropy works fine. You can adjust the loss based on class imbalances in your training set. </p>",
      "rawMarkdown": "Are you looking for a good loss function? Categorical crossentropy works fine. You can adjust the loss based on class imbalances in your training set.",
      "votes": null
    },
    {
      "id": "404496",
      "postDate": "10/15/2018 20:19:35",
      "content": "<blockquote>\n  <p>I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?</p>\n</blockquote>\n\n<p>Use categorical cross-entropy (log-loss) for training, f1-macro score for early stopping.</p>",
      "rawMarkdown": "&gt; I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?\n\nUse categorical cross-entropy (log-loss) for training, f1-macro score for early stopping.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 404488,
      "author_name": "nastikavidwan",
      "author_url": "",
      "post_date": "10/15/2018 19:54:37",
      "content": "<p>Are you looking for a good loss function? Categorical crossentropy works fine. You can adjust the loss based on class imbalances in your training set. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 404496,
      "author_name": "tilii7",
      "author_url": "",
      "post_date": "10/15/2018 20:19:35",
      "content": "<blockquote>\n  <p>I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?</p>\n</blockquote>\n\n<p>Use categorical cross-entropy (log-loss) for training, f1-macro score for early stopping.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "404455": "I'm having some trouble picking training metric for the competition. I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?",
    "404488": "Are you looking for a good loss function? Categorical crossentropy works fine. You can adjust the loss based on class imbalances in your training set.",
    "404496": "&gt; I'm using keras but obviosly you can't train on f1 macro, but are there any similar alternatives that you can train on?\n\nUse categorical cross-entropy (log-loss) for training, f1-macro score for early stopping."
  },
  "source": "meta"
}