{
  "id": 131886,
  "title": "is 168 classes for grapheme root enough?",
  "url": "/competitions/bengaliai-cv19/discussion/131886",
  "author_name": "",
  "post_date": "2020-02-22T11:32:26.268411800Z",
  "votes": 8,
  "comment_count": 2,
  "views": 0,
  "content": "<p>there are a few cases like this shown below</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F70a13a71e651f58f1eda486eba322370%2FSelection_094.png?generation=1582371142145658&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "753566",
      "postDate": "02/22/2020 11:32:26",
      "content": "<p>there are a few cases like this shown below</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F70a13a71e651f58f1eda486eba322370%2FSelection_094.png?generation=1582371142145658&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "there are a few cases like this shown below\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F70a13a71e651f58f1eda486eba322370%2FSelection_094.png?generation=1582371142145658&amp;alt=media)",
      "votes": null
    },
    {
      "id": "753616",
      "postDate": "02/22/2020 13:23:24",
      "content": "<p>Nice, that should improve the classification, but did you check if your network is actually struggling with 13 in particular? If it is not performing specifically bad for this class, we could infer that it has already learned that there are 2 styles for it?</p>",
      "rawMarkdown": "Nice, that should improve the classification, but did you check if your network is actually struggling with 13 in particular? If it is not performing specifically bad for this class, we could infer that it has already learned that there are 2 styles for it?",
      "votes": null
    },
    {
      "id": "760714",
      "postDate": "03/01/2020 16:39:29",
      "content": "<p><a href=\"/hengck23\">@hengck23</a> hello, would you please suggest what's the best approaches we can do to handle grapheme_root class? The best validation score I get of it is about 97% but I'm not sure if we need to balance it anyway before training. The other two classes are easily going to 98 or 99%.  I need some advice from you. Thank you. </p>",
      "rawMarkdown": "hengck23 hello, would you please suggest what's the best approaches we can do to handle grapheme_root class? The best validation score I get of it is about 97% but I'm not sure if we need to balance it anyway before training. The other two classes are easily going to 98 or 99%.  I need some advice from you. Thank you.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 753616,
      "author_name": "bsridatta",
      "author_url": "",
      "post_date": "02/22/2020 13:23:24",
      "content": "<p>Nice, that should improve the classification, but did you check if your network is actually struggling with 13 in particular? If it is not performing specifically bad for this class, we could infer that it has already learned that there are 2 styles for it?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 760714,
      "author_name": "ipythonx",
      "author_url": "",
      "post_date": "03/01/2020 16:39:29",
      "content": "<p><a href=\"/hengck23\">@hengck23</a> hello, would you please suggest what's the best approaches we can do to handle grapheme_root class? The best validation score I get of it is about 97% but I'm not sure if we need to balance it anyway before training. The other two classes are easily going to 98 or 99%.  I need some advice from you. Thank you. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "753566": "there are a few cases like this shown below\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F70a13a71e651f58f1eda486eba322370%2FSelection_094.png?generation=1582371142145658&amp;alt=media)",
    "753616": "Nice, that should improve the classification, but did you check if your network is actually struggling with 13 in particular? If it is not performing specifically bad for this class, we could infer that it has already learned that there are 2 styles for it?",
    "760714": "hengck23 hello, would you please suggest what's the best approaches we can do to handle grapheme_root class? The best validation score I get of it is about 97% but I'm not sure if we need to balance it anyway before training. The other two classes are easily going to 98 or 99%.  I need some advice from you. Thank you."
  },
  "source": "meta"
}