{
  "id": 127257,
  "title": "rank68_the_ simplest idea to get the first medal",
  "url": "/competitions/tensorflow2-question-answering/discussion/127257",
  "author_name": "van73",
  "post_date": "2020-01-23T04:35:28.708000",
  "votes": 6,
  "comment_count": 3,
  "views": 0,
  "content": "<p><strong>The whole process of solving this problem in my way：</strong></p>\n\n<p><em><strong>source</strong></em>:fork from bert joint baseline notebook</p>\n\n<p>**<em>problems</em> **(before starting this competition): </p>\n\n<p>1.the first time to deal with QA </p>\n\n<p>2.just have a month to go</p>\n\n<p>3.need to understand source codes in tf2.0 ver(just familiar with pytorch)</p>\n\n<p>4.GPU (just have a 1080ti)</p>\n\n<p><strong>Based on current problems, I came up with some solutions:</strong></p>\n\n<p>1.read tf2.0 source code（necessary） #done</p>\n\n<p>2.fine-tune bert-joint-model (<strong>the simplest solution to get started and indeed understand the whole process</strong>)#done</p>\n\n<p>3.try to convert tf2.0 code to pytorch ver (but have not a better result untill now )#done</p>\n\n<p>4.change the pipeline(Some effort)</p>\n\n<p>In this competition, I think the first and second points are the most important for the first time runners，and the main changes of mine are as follows：</p>\n\n<p>1.FLAGS=DummyObject(skip_nested_contexts=True,\n                 max_position=50,\n                 <strong>max_contexts=130</strong>,\n                 max_query_length=64,\n                 max_seq_length=512,\n                 doc_stride=128,\n                 include_unknowns=0.02,\n                 n_best_size=50,\n                 max_answer_length=60)</p>\n\n<p>2.entry[\"short_answer_score\"] &lt; 7.5:</p>\n\n<p>3.entry[\"long_answer_score\"] &lt; 1.5:</p>\n\n<p><strong>I hope to be of some help to those who have just started a new task with limited time and resources</strong></p>\n\n<p>kernel:<a href=\"https://www.kaggle.com/vanle73/rank68-the-simplest-idea-to-get-medal?scriptVersionId=26631305\">rank68_the_ simplest idea to get medal</a></p>",
  "messages": [
    {
      "id": 726553,
      "postDate": "2020-01-23T04:35:28.710Z",
      "content": "<p><strong>The whole process of solving this problem in my way：</strong></p>\n\n<p><em><strong>source</strong></em>:fork from bert joint baseline notebook</p>\n\n<p>**<em>problems</em> **(before starting this competition): </p>\n\n<p>1.the first time to deal with QA </p>\n\n<p>2.just have a month to go</p>\n\n<p>3.need to understand source codes in tf2.0 ver(just familiar with pytorch)</p>\n\n<p>4.GPU (just have a 1080ti)</p>\n\n<p><strong>Based on current problems, I came up with some solutions:</strong></p>\n\n<p>1.read tf2.0 source code（necessary） #done</p>\n\n<p>2.fine-tune bert-joint-model (<strong>the simplest solution to get started and indeed understand the whole process</strong>)#done</p>\n\n<p>3.try to convert tf2.0 code to pytorch ver (but have not a better result untill now )#done</p>\n\n<p>4.change the pipeline(Some effort)</p>\n\n<p>In this competition, I think the first and second points are the most important for the first time runners，and the main changes of mine are as follows：</p>\n\n<p>1.FLAGS=DummyObject(skip_nested_contexts=True,\n                 max_position=50,\n                 <strong>max_contexts=130</strong>,\n                 max_query_length=64,\n                 max_seq_length=512,\n                 doc_stride=128,\n                 include_unknowns=0.02,\n                 n_best_size=50,\n                 max_answer_length=60)</p>\n\n<p>2.entry[\"short_answer_score\"] &lt; 7.5:</p>\n\n<p>3.entry[\"long_answer_score\"] &lt; 1.5:</p>\n\n<p><strong>I hope to be of some help to those who have just started a new task with limited time and resources</strong></p>\n\n<p>kernel:<a href=\"https://www.kaggle.com/vanle73/rank68-the-simplest-idea-to-get-medal?scriptVersionId=26631305\">rank68_the_ simplest idea to get medal</a></p>",
      "rawMarkdown": "**The whole process of solving this problem in my way：**\n\n***source***:fork from bert joint baseline notebook\n\n***problems* **(before starting this competition): \n\n1.the first time to deal with QA \n\n2.just have a month to go\n\n3.need to understand source codes in tf2.0 ver(just familiar with pytorch)\n\n4.GPU (just have a 1080ti)\n\n**Based on current problems, I came up with some solutions:**\n\n1.read tf2.0 source code（necessary） #done\n\n2.fine-tune bert-joint-model (**the simplest solution to get started and indeed understand the whole process**)#done\n\n3.try to convert tf2.0 code to pytorch ver (but have not a better result untill now )#done\n\n4.change the pipeline(Some effort)\n\nIn this competition, I think the first and second points are the most important for the first time runners，and the main changes of mine are as follows：\n\n1.FLAGS=DummyObject(skip_nested_contexts=True,\n                 max_position=50,\n                 **max_contexts=130**,\n                 max_query_length=64,\n                 max_seq_length=512,\n                 doc_stride=128,\n                 include_unknowns=0.02,\n                 n_best_size=50,\n                 max_answer_length=60)\n                 \n2.entry[\"short_answer_score\"] &lt; 7.5:\n\n3.entry[\"long_answer_score\"] &lt; 1.5:\n\n\n**I hope to be of some help to those who have just started a new task with limited time and resources**\n\nkernel:[rank68_the_ simplest idea to get medal](https://www.kaggle.com/vanle73/rank68-the-simplest-idea-to-get-medal?scriptVersionId=26631305)",
      "votes": 6
    },
    {
      "id": 728497,
      "postDate": "2020-01-24T19:57:01.527Z",
      "content": "<p>Congrats &amp; Thanks for sharing your solution🎉 😄 👍 </p>",
      "rawMarkdown": "Congrats &amp; Thanks for sharing your solution🎉 😄 👍 "
    },
    {
      "id": 726566,
      "postDate": "2020-01-23T04:50:11.507Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 730151,
      "postDate": "2020-01-27T06:44:32.130Z",
      "content": "<p>nice share,thanks</p>",
      "rawMarkdown": "nice share,thanks"
    }
  ],
  "comments": [
    {
      "id": 728497,
      "author_name": "Miyabon",
      "author_url": "",
      "post_date": "2020-01-24T19:57:01.527000",
      "content": "<p>Congrats &amp; Thanks for sharing your solution🎉 😄 👍 </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 726566,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-01-23T04:50:11.507000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 730151,
      "author_name": "YunliangLi",
      "author_url": "",
      "post_date": "2020-01-27T06:44:32.130000",
      "content": "<p>nice share,thanks</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "726553": "**The whole process of solving this problem in my way：**\n\n***source***:fork from bert joint baseline notebook\n\n***problems* **(before starting this competition): \n\n1.the first time to deal with QA \n\n2.just have a month to go\n\n3.need to understand source codes in tf2.0 ver(just familiar with pytorch)\n\n4.GPU (just have a 1080ti)\n\n**Based on current problems, I came up with some solutions:**\n\n1.read tf2.0 source code（necessary） #done\n\n2.fine-tune bert-joint-model (**the simplest solution to get started and indeed understand the whole process**)#done\n\n3.try to convert tf2.0 code to pytorch ver (but have not a better result untill now )#done\n\n4.change the pipeline(Some effort)\n\nIn this competition, I think the first and second points are the most important for the first time runners，and the main changes of mine are as follows：\n\n1.FLAGS=DummyObject(skip_nested_contexts=True,\n                 max_position=50,\n                 **max_contexts=130**,\n                 max_query_length=64,\n                 max_seq_length=512,\n                 doc_stride=128,\n                 include_unknowns=0.02,\n                 n_best_size=50,\n                 max_answer_length=60)\n                 \n2.entry[\"short_answer_score\"] &lt; 7.5:\n\n3.entry[\"long_answer_score\"] &lt; 1.5:\n\n\n**I hope to be of some help to those who have just started a new task with limited time and resources**\n\nkernel:[rank68_the_ simplest idea to get medal](https://www.kaggle.com/vanle73/rank68-the-simplest-idea-to-get-medal?scriptVersionId=26631305)",
    "728497": "Congrats &amp; Thanks for sharing your solution🎉 😄 👍 ",
    "726566": "",
    "730151": "nice share,thanks"
  }
}