{
  "id": 232334,
  "title": "Evaluation metrics understanding help.",
  "url": "/competitions/hpa-single-cell-image-classification/discussion/232334",
  "author_name": "olga puntous",
  "post_date": "2021-04-13T09:56:20.856000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello, </p>\n<p>I am quite new to kaggle competitions with segmentation task. </p>\n<p>I did my submission with my baseline approach and I got 0 ( expected value was 0.01), des anybody has an idea why ? ( I descibe my submission below)</p>\n<p>Here is my baseline approach:</p>\n<p>I trained 18 models (one model for each label ). I used balanced single labels datasets to train my  models. So to make a final prediction I made a loop with 18 models  to predict if the label is here or not.However the accuracy for each label varies from 0.6 to 0.8, recall is between 0.4  and  0.6. ( I did evaluation on separate test set)</p>\n<p>Also, I ve decided to cheat a bit on segmentation task and instead of mask of unique cell  for label I gave for each label mask of all cells presented on the image.</p>\n<p>Then I made my submission. I made submission with assumption that public test set is the part off pivate test set (it is true!!!) , so I filled submission template with resuts already calculated on test set.(like this I was hoping to get at least 0.001 score  with intersection of my solution and real submission numbers)!!! But I got 0.000 score .</p>\n<p>Could somebody explain me why?</p>",
  "messages": [
    {
      "id": 1272198,
      "postDate": "2021-04-13T09:56:20.857Z",
      "content": "<p>Hello, </p>\n<p>I am quite new to kaggle competitions with segmentation task. </p>\n<p>I did my submission with my baseline approach and I got 0 ( expected value was 0.01), des anybody has an idea why ? ( I descibe my submission below)</p>\n<p>Here is my baseline approach:</p>\n<p>I trained 18 models (one model for each label ). I used balanced single labels datasets to train my  models. So to make a final prediction I made a loop with 18 models  to predict if the label is here or not.However the accuracy for each label varies from 0.6 to 0.8, recall is between 0.4  and  0.6. ( I did evaluation on separate test set)</p>\n<p>Also, I ve decided to cheat a bit on segmentation task and instead of mask of unique cell  for label I gave for each label mask of all cells presented on the image.</p>\n<p>Then I made my submission. I made submission with assumption that public test set is the part off pivate test set (it is true!!!) , so I filled submission template with resuts already calculated on test set.(like this I was hoping to get at least 0.001 score  with intersection of my solution and real submission numbers)!!! But I got 0.000 score .</p>\n<p>Could somebody explain me why?</p>",
      "rawMarkdown": "Hello, \n\nI am quite new to kaggle competitions with segmentation task. \n\nI did my submission with my baseline approach and I got 0 ( expected value was 0.01), des anybody has an idea why ? ( I descibe my submission below)\n\nHere is my baseline approach:\n\nI trained 18 models (one model for each label ). I used balanced single labels datasets to train my  models. So to make a final prediction I made a loop with 18 models  to predict if the label is here or not.However the accuracy for each label varies from 0.6 to 0.8, recall is between 0.4  and  0.6. ( I did evaluation on separate test set)\n\nAlso, I ve decided to cheat a bit on segmentation task and instead of mask of unique cell  for label I gave for each label mask of all cells presented on the image.\n\nThen I made my submission. I made submission with assumption that public test set is the part off pivate test set (it is true!!!) , so I filled submission template with resuts already calculated on test set.(like this I was hoping to get at least 0.001 score  with intersection of my solution and real submission numbers)!!! But I got 0.000 score .\n\nCould somebody explain me why?\n\n",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1272198": "Hello, \n\nI am quite new to kaggle competitions with segmentation task. \n\nI did my submission with my baseline approach and I got 0 ( expected value was 0.01), des anybody has an idea why ? ( I descibe my submission below)\n\nHere is my baseline approach:\n\nI trained 18 models (one model for each label ). I used balanced single labels datasets to train my  models. So to make a final prediction I made a loop with 18 models  to predict if the label is here or not.However the accuracy for each label varies from 0.6 to 0.8, recall is between 0.4  and  0.6. ( I did evaluation on separate test set)\n\nAlso, I ve decided to cheat a bit on segmentation task and instead of mask of unique cell  for label I gave for each label mask of all cells presented on the image.\n\nThen I made my submission. I made submission with assumption that public test set is the part off pivate test set (it is true!!!) , so I filled submission template with resuts already calculated on test set.(like this I was hoping to get at least 0.001 score  with intersection of my solution and real submission numbers)!!! But I got 0.000 score .\n\nCould somebody explain me why?\n\n"
  }
}