{
  "id": 124639,
  "title": "Problem with accuracy when uploading a trained model",
  "url": "/competitions/bengaliai-cv19/discussion/124639",
  "author_name": "Anoop Kumar yadav",
  "post_date": "2020-01-05T14:58:02.164000",
  "votes": 0,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I downloaded data on google colab and trained model there during training I was getting accuracy of approx 90% on valid set. I used fastai library and also converted input files to images and trained and saved model. When I uploaded model here I got score of 0.0614. Can any one explain what happened? </p>",
  "messages": [
    {
      "id": 711787,
      "postDate": "2020-01-06T14:39:14.387Z",
      "content": "<p>That is the score of the sample_submission.csv file, I think you doing something wrong while saving the submission file.</p>",
      "rawMarkdown": "That is the score of the sample_submission.csv file, I think you doing something wrong while saving the submission file.\n"
    },
    {
      "id": 711079,
      "postDate": "2020-01-05T16:58:42.073Z",
      "content": "<p>While submitting, the test image data will be replaced by huge one. So predict using parquet file. Not using the sample_esubmission.csv because sample_submission has only 36 rows</p>",
      "rawMarkdown": "While submitting, the test image data will be replaced by huge one. So predict using parquet file. Not using the sample_esubmission.csv because sample_submission has only 36 rows"
    },
    {
      "id": 711070,
      "postDate": "2020-01-05T16:41:56.663Z",
      "content": "<p>The one explanation I can think of is overfitting. Try splitting your data set into a 3 way split between the train, cross-validation and test sets if you haven't already. </p>",
      "rawMarkdown": "The one explanation I can think of is overfitting. Try splitting your data set into a 3 way split between the train, cross-validation and test sets if you haven't already. "
    },
    {
      "id": 711013,
      "postDate": "2020-01-05T14:58:02.163Z",
      "content": "<p>I downloaded data on google colab and trained model there during training I was getting accuracy of approx 90% on valid set. I used fastai library and also converted input files to images and trained and saved model. When I uploaded model here I got score of 0.0614. Can any one explain what happened? </p>",
      "rawMarkdown": "I downloaded data on google colab and trained model there during training I was getting accuracy of approx 90% on valid set. I used fastai library and also converted input files to images and trained and saved model. When I uploaded model here I got score of 0.0614. Can any one explain what happened? "
    },
    {
      "id": 712811,
      "postDate": "2020-01-07T15:52:56.953Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 711787,
      "author_name": "ShubhankarMohan",
      "author_url": "",
      "post_date": "2020-01-06T14:39:14.387000",
      "content": "<p>That is the score of the sample_submission.csv file, I think you doing something wrong while saving the submission file.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 711079,
      "author_name": "Bibhas Mondal",
      "author_url": "",
      "post_date": "2020-01-05T16:58:42.073000",
      "content": "<p>While submitting, the test image data will be replaced by huge one. So predict using parquet file. Not using the sample_esubmission.csv because sample_submission has only 36 rows</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 711070,
      "author_name": "Brett Odle",
      "author_url": "",
      "post_date": "2020-01-05T16:41:56.663000",
      "content": "<p>The one explanation I can think of is overfitting. Try splitting your data set into a 3 way split between the train, cross-validation and test sets if you haven't already. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 712811,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-01-07T15:52:56.953000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "711787": "That is the score of the sample_submission.csv file, I think you doing something wrong while saving the submission file.\n",
    "711079": "While submitting, the test image data will be replaced by huge one. So predict using parquet file. Not using the sample_esubmission.csv because sample_submission has only 36 rows",
    "711070": "The one explanation I can think of is overfitting. Try splitting your data set into a 3 way split between the train, cross-validation and test sets if you haven't already. ",
    "711013": "I downloaded data on google colab and trained model there during training I was getting accuracy of approx 90% on valid set. I used fastai library and also converted input files to images and trained and saved model. When I uploaded model here I got score of 0.0614. Can any one explain what happened? ",
    "712811": ""
  }
}