{
  "id": 240950,
  "title": "I can't find out what is missing for my submission file.",
  "url": "/competitions/coleridgeinitiative-show-us-the-data/discussion/240950",
  "author_name": "",
  "post_date": "2021-05-22T09:30:46.693254Z",
  "votes": 1,
  "comment_count": 14,
  "views": 0,
  "content": "<p>Dear Kaggle Team-members,</p>\n<p>I made a basic LSTM model for initial testing and get reasonable predictions, code is showing in the following link.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/gavinjang/my-rnn-model</a></p>\n<p>Yet, the public score is 0.000, and I am sure my prediction is not that bad.</p>\n<p>For instance, the following code from Mr. <a href=\"https://www.kaggle.com/mlconsult\" target=\"_blank\">@mlconsult</a> has a relatively closed result with mine, and his score is 0.573.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/mlconsult/score-57ish-with-additional-govt-datasets?select=submission.csv</a></p>\n<p>Therefore, I am wondering which part is missing for my submission file and I look into the formation, yet, mine looks as fine as Ken's. I got my text cleaned with \"def text_clean\", the separation is ',' and have the pipe symbols for multi-labeled output. </p>\n<p>Could someone help me to find out what happened to my submission?</p>\n<p>I appreciate and looking forward to your kind assistance. </p>\n<p>Sincerely,</p>\n<p>Gavin</p>",
  "messages": [
    {
      "id": "1318383",
      "postDate": "05/22/2021 09:30:46",
      "content": "<p>Dear Kaggle Team-members,</p>\n<p>I made a basic LSTM model for initial testing and get reasonable predictions, code is showing in the following link.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/gavinjang/my-rnn-model</a></p>\n<p>Yet, the public score is 0.000, and I am sure my prediction is not that bad.</p>\n<p>For instance, the following code from Mr. <a href=\"https://www.kaggle.com/mlconsult\" target=\"_blank\">@mlconsult</a> has a relatively closed result with mine, and his score is 0.573.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/mlconsult/score-57ish-with-additional-govt-datasets?select=submission.csv</a></p>\n<p>Therefore, I am wondering which part is missing for my submission file and I look into the formation, yet, mine looks as fine as Ken's. I got my text cleaned with \"def text_clean\", the separation is ',' and have the pipe symbols for multi-labeled output. </p>\n<p>Could someone help me to find out what happened to my submission?</p>\n<p>I appreciate and looking forward to your kind assistance. </p>\n<p>Sincerely,</p>\n<p>Gavin</p>",
      "rawMarkdown": "Dear Kaggle Team-members,\n\nI made a basic LSTM model for initial testing and get reasonable predictions, code is showing in the following link.\n\n[https://www.kaggle.com/gavinjang/my-rnn-model](url)\n\nYet, the public score is 0.000, and I am sure my prediction is not that bad.\n\nFor instance, the following code from Mr. @mlconsult has a relatively closed result with mine, and his score is 0.573.\n\n[https://www.kaggle.com/mlconsult/score-57ish-with-additional-govt-datasets?select=submission.csv](url)\n\nTherefore, I am wondering which part is missing for my submission file and I look into the formation, yet, mine looks as fine as Ken's. I got my text cleaned with \"def text_clean\", the separation is ',' and have the pipe symbols for multi-labeled output. \n\nCould someone help me to find out what happened to my submission?\n\nI appreciate and looking forward to your kind assistance. \n\nSincerely,\n\nGavin",
      "votes": null
    },
    {
      "id": "1318424",
      "postDate": "05/22/2021 10:19:58",
      "content": "<p>If I understand your code correctly, you only predict datasets for the first 44 publications.</p>",
      "rawMarkdown": "If I understand your code correctly, you only predict datasets for the first 44 publications.",
      "votes": null
    },
    {
      "id": "1318462",
      "postDate": "05/22/2021 11:03:11",
      "content": "<p>Oh, I sort all publications into 44 categories according to the [\"dataset_title\"]. So, the output dimension is 44.</p>",
      "rawMarkdown": "Oh, I sort all publications into 44 categories according to the [\"dataset_title\"]. So, the output dimension is 44.",
      "votes": null
    },
    {
      "id": "1318592",
      "postDate": "05/22/2021 12:24:02",
      "content": "<p>Ok, so I think I misunderstood your code.</p>",
      "rawMarkdown": "Ok, so I think I misunderstood your code.",
      "votes": null
    },
    {
      "id": "1318705",
      "postDate": "05/22/2021 14:05:46",
      "content": "<p>same as yours, mine 0.004, I can not figure it out till now, look forward to having your kind assistance too</p>",
      "rawMarkdown": "same as yours, mine 0.004, I can not figure it out till now, look forward to having your kind assistance too",
      "votes": null
    },
    {
      "id": "1318712",
      "postDate": "05/22/2021 14:11:56",
      "content": "<p>I had a look again. </p>\n<blockquote>\n  <p>text_sub = pd.read_csv(\"../input/preprocessed-for-coleridgeinitiative/test_text.csv\")</p>\n</blockquote>\n<p>You get the texts from a presaved file. That doesn't work because when submitting the hidden test set needs to be loaded (by refering to the test ids in sumbission_df and the json-files in your notebook).</p>",
      "rawMarkdown": "I had a look again. \n> \ntext_sub = pd.read_csv(\"../input/preprocessed-for-coleridgeinitiative/test_text.csv\")\n\nYou get the texts from a presaved file. That doesn't work because when submitting the hidden test set needs to be loaded (by refering to the test ids in sumbission_df and the json-files in your notebook).",
      "votes": null
    },
    {
      "id": "1318741",
      "postDate": "05/22/2021 14:29:33",
      "content": "<p>I understand that they will have a private test set and my code won't work in that scenario. However, there still shall be the public score for my initial test.</p>",
      "rawMarkdown": "I understand that they will have a private test set and my code won't work in that scenario. However, there still shall be the public score for my initial test.",
      "votes": null
    },
    {
      "id": "1318751",
      "postDate": "05/22/2021 14:42:25",
      "content": "<p>No, the public test set is also hidden. The four examples which can be seen are not used to calculate your public and private test score.</p>",
      "rawMarkdown": "No, the public test set is also hidden. The four examples which can be seen are not used to calculate your public and private test score.",
      "votes": null
    },
    {
      "id": "1318756",
      "postDate": "05/22/2021 14:51:16",
      "content": "<p>Oh, I see, thank you very much for your time.</p>",
      "rawMarkdown": "Oh, I see, thank you very much for your time.",
      "votes": null
    },
    {
      "id": "1342316",
      "postDate": "06/09/2021 10:50:06",
      "content": "<p>Hi Linh,</p>\n<p>Sorry for my late response.</p>\n<p>In my case, I pre-processed and tokenized all text and turn into CSV files on local machine. So, my code won't get real public and private text, and obviously, I would have 0.000.</p>\n<p>But you did get some point, it means that the problem may result from your model. You may need some more tuning for your model.</p>\n<p>Sincerely,</p>\n<p>Gavin</p>",
      "rawMarkdown": "Hi Linh,\n\nSorry for my late response.\n\n In my case, I pre-processed and tokenized all text and turn into CSV files on local machine. So, my code won't get real public and private text, and obviously, I would have 0.000.\n\nBut you did get some point, it means that the problem may result from your model. You may need some more tuning for your model.\n\nSincerely,\n\nGavin",
      "votes": null
    },
    {
      "id": "1342479",
      "postDate": "06/09/2021 13:46:39",
      "content": "<p>Thanks for your reply, <br>\nthis is my first competition, and working around Kaggle, yesterday I can use GPU, get score of 0.505 and finished running within 2 hours, consider it is my big progress :), and now despite busy working but trying my best to make score little higher before the competition closes</p>",
      "rawMarkdown": "Thanks for your reply, \nthis is my first competition, and working around Kaggle, yesterday I can use GPU, get score of 0.505 and finished running within 2 hours, consider it is my big progress :), and now despite busy working but trying my best to make score little higher before the competition closes",
      "votes": null
    },
    {
      "id": "1343267",
      "postDate": "06/10/2021 06:20:06",
      "content": "<p>Wow! That's a pretty high score. I am still struggling with insufficient RAM for my submission.</p>\n<p>Sincerely,<br>\nGavin</p>",
      "rawMarkdown": "Wow! That's a pretty high score. I am still struggling with insufficient RAM for my submission.\n\nSincerely,\nGavin",
      "votes": null
    },
    {
      "id": "1343594",
      "postDate": "06/10/2021 09:45:36",
      "content": "<p>yes, like mine, i think you can split into sentences (using nltk for instance) and filter sentences contains words that you might think to have dataset name it must have words such as: \"data\", \"dataset\", \"database\", \"survey\", etc, and try if possible filtering by POS like verb that you find it high frequency in train set, \"used\", \"using\", \"collected\", etc (bag of verbs) for example, so will reduce huge un-qualified sentences </p>",
      "rawMarkdown": "yes, like mine, i think you can split into sentences (using nltk for instance) and filter sentences contains words that you might think to have dataset name it must have words such as: \"data\", \"dataset\", \"database\", \"survey\", etc, and try if possible filtering by POS like verb that you find it high frequency in train set, \"used\", \"using\", \"collected\", etc (bag of verbs) for example, so will reduce huge un-qualified sentences",
      "votes": null
    },
    {
      "id": "1343864",
      "postDate": "06/10/2021 13:41:02",
      "content": "<p>I have found the solution this afternoon. It is too much memory consumption if I cram all test text into TensorFlow object. My code works fine now. Anyway, thank you very much for your sharing.</p>",
      "rawMarkdown": "I have found the solution this afternoon. It is too much memory consumption if I cram all test text into TensorFlow object. My code works fine now. Anyway, thank you very much for your sharing.",
      "votes": null
    },
    {
      "id": "1343879",
      "postDate": "06/10/2021 13:47:29",
      "content": "<p>👍 great to hear that, good luck to your next!</p>",
      "rawMarkdown": "👍 great to hear that, good luck to your next!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1318424,
      "author_name": "hannes82",
      "author_url": "",
      "post_date": "05/22/2021 10:19:58",
      "content": "<p>If I understand your code correctly, you only predict datasets for the first 44 publications.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1318462,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "05/22/2021 11:03:11",
          "content": "<p>Oh, I sort all publications into 44 categories according to the [\"dataset_title\"]. So, the output dimension is 44.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1318592,
          "author_name": "hannes82",
          "author_url": "",
          "post_date": "05/22/2021 12:24:02",
          "content": "<p>Ok, so I think I misunderstood your code.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1318712,
          "author_name": "hannes82",
          "author_url": "",
          "post_date": "05/22/2021 14:11:56",
          "content": "<p>I had a look again. </p>\n<blockquote>\n  <p>text_sub = pd.read_csv(\"../input/preprocessed-for-coleridgeinitiative/test_text.csv\")</p>\n</blockquote>\n<p>You get the texts from a presaved file. That doesn't work because when submitting the hidden test set needs to be loaded (by refering to the test ids in sumbission_df and the json-files in your notebook).</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1318741,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "05/22/2021 14:29:33",
          "content": "<p>I understand that they will have a private test set and my code won't work in that scenario. However, there still shall be the public score for my initial test.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1318751,
          "author_name": "hannes82",
          "author_url": "",
          "post_date": "05/22/2021 14:42:25",
          "content": "<p>No, the public test set is also hidden. The four examples which can be seen are not used to calculate your public and private test score.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1318756,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "05/22/2021 14:51:16",
          "content": "<p>Oh, I see, thank you very much for your time.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1318705,
      "author_name": "nguyenvanlinh",
      "author_url": "",
      "post_date": "05/22/2021 14:05:46",
      "content": "<p>same as yours, mine 0.004, I can not figure it out till now, look forward to having your kind assistance too</p>",
      "votes": null,
      "replies": [
        {
          "id": 1342316,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "06/09/2021 10:50:06",
          "content": "<p>Hi Linh,</p>\n<p>Sorry for my late response.</p>\n<p>In my case, I pre-processed and tokenized all text and turn into CSV files on local machine. So, my code won't get real public and private text, and obviously, I would have 0.000.</p>\n<p>But you did get some point, it means that the problem may result from your model. You may need some more tuning for your model.</p>\n<p>Sincerely,</p>\n<p>Gavin</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1342479,
          "author_name": "nguyenvanlinh",
          "author_url": "",
          "post_date": "06/09/2021 13:46:39",
          "content": "<p>Thanks for your reply, <br>\nthis is my first competition, and working around Kaggle, yesterday I can use GPU, get score of 0.505 and finished running within 2 hours, consider it is my big progress :), and now despite busy working but trying my best to make score little higher before the competition closes</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1343267,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "06/10/2021 06:20:06",
          "content": "<p>Wow! That's a pretty high score. I am still struggling with insufficient RAM for my submission.</p>\n<p>Sincerely,<br>\nGavin</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1343594,
          "author_name": "nguyenvanlinh",
          "author_url": "",
          "post_date": "06/10/2021 09:45:36",
          "content": "<p>yes, like mine, i think you can split into sentences (using nltk for instance) and filter sentences contains words that you might think to have dataset name it must have words such as: \"data\", \"dataset\", \"database\", \"survey\", etc, and try if possible filtering by POS like verb that you find it high frequency in train set, \"used\", \"using\", \"collected\", etc (bag of verbs) for example, so will reduce huge un-qualified sentences </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1343864,
          "author_name": "gavinjang",
          "author_url": "",
          "post_date": "06/10/2021 13:41:02",
          "content": "<p>I have found the solution this afternoon. It is too much memory consumption if I cram all test text into TensorFlow object. My code works fine now. Anyway, thank you very much for your sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1343879,
          "author_name": "nguyenvanlinh",
          "author_url": "",
          "post_date": "06/10/2021 13:47:29",
          "content": "<p>👍 great to hear that, good luck to your next!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1318383": "Dear Kaggle Team-members,\n\nI made a basic LSTM model for initial testing and get reasonable predictions, code is showing in the following link.\n\n[https://www.kaggle.com/gavinjang/my-rnn-model](url)\n\nYet, the public score is 0.000, and I am sure my prediction is not that bad.\n\nFor instance, the following code from Mr. @mlconsult has a relatively closed result with mine, and his score is 0.573.\n\n[https://www.kaggle.com/mlconsult/score-57ish-with-additional-govt-datasets?select=submission.csv](url)\n\nTherefore, I am wondering which part is missing for my submission file and I look into the formation, yet, mine looks as fine as Ken's. I got my text cleaned with \"def text_clean\", the separation is ',' and have the pipe symbols for multi-labeled output. \n\nCould someone help me to find out what happened to my submission?\n\nI appreciate and looking forward to your kind assistance. \n\nSincerely,\n\nGavin",
    "1318424": "If I understand your code correctly, you only predict datasets for the first 44 publications.",
    "1318462": "Oh, I sort all publications into 44 categories according to the [\"dataset_title\"]. So, the output dimension is 44.",
    "1318592": "Ok, so I think I misunderstood your code.",
    "1318705": "same as yours, mine 0.004, I can not figure it out till now, look forward to having your kind assistance too",
    "1318712": "I had a look again. \n> \ntext_sub = pd.read_csv(\"../input/preprocessed-for-coleridgeinitiative/test_text.csv\")\n\nYou get the texts from a presaved file. That doesn't work because when submitting the hidden test set needs to be loaded (by refering to the test ids in sumbission_df and the json-files in your notebook).",
    "1318741": "I understand that they will have a private test set and my code won't work in that scenario. However, there still shall be the public score for my initial test.",
    "1318751": "No, the public test set is also hidden. The four examples which can be seen are not used to calculate your public and private test score.",
    "1318756": "Oh, I see, thank you very much for your time.",
    "1342316": "Hi Linh,\n\nSorry for my late response.\n\n In my case, I pre-processed and tokenized all text and turn into CSV files on local machine. So, my code won't get real public and private text, and obviously, I would have 0.000.\n\nBut you did get some point, it means that the problem may result from your model. You may need some more tuning for your model.\n\nSincerely,\n\nGavin",
    "1342479": "Thanks for your reply, \nthis is my first competition, and working around Kaggle, yesterday I can use GPU, get score of 0.505 and finished running within 2 hours, consider it is my big progress :), and now despite busy working but trying my best to make score little higher before the competition closes",
    "1343267": "Wow! That's a pretty high score. I am still struggling with insufficient RAM for my submission.\n\nSincerely,\nGavin",
    "1343594": "yes, like mine, i think you can split into sentences (using nltk for instance) and filter sentences contains words that you might think to have dataset name it must have words such as: \"data\", \"dataset\", \"database\", \"survey\", etc, and try if possible filtering by POS like verb that you find it high frequency in train set, \"used\", \"using\", \"collected\", etc (bag of verbs) for example, so will reduce huge un-qualified sentences",
    "1343864": "I have found the solution this afternoon. It is too much memory consumption if I cram all test text into TensorFlow object. My code works fine now. Anyway, thank you very much for your sharing.",
    "1343879": "👍 great to hear that, good luck to your next!"
  },
  "source": "meta"
}