{
  "id": 160380,
  "title": "How to make the curve smoother?",
  "url": "/competitions/digit-recognizer/discussion/160380",
  "author_name": "Fatin Ishraq ",
  "post_date": "2020-06-21T04:14:27.790000",
  "votes": 0,
  "comment_count": 1,
  "views": null,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2F88bfec40e1eefa2ba1ebefb67b2b4f85%2FCapture.PNG?generation=1592712639799820&amp;alt=media\" alt=\"\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2Ff74ff305268b2a19a3330ac9b34c8d74%2F2.PNG?generation=1592712641453053&amp;alt=media\" alt=\"\">\nTo make the curve smooth, the things we should try out:\n1. <strong>regularization</strong>, so that the overfitting does not occur\n2. <strong>EarlyStopping callback</strong>, we can use this callback using tensorflow.keras and set the monitor to val_loss or val_accuracy and set patience = 3/4/5 so that the code stops executing whenever there is a failure in the improvement of results.</p>",
  "messages": [
    {
      "id": 932216,
      "postDate": "2020-07-16T20:25:50.997Z",
      "content": "<p>I think it's worth to mention a <strong>dropout</strong> technique and <strong>data augmentation</strong>, which also really help to keep the curve smooth.</p>",
      "rawMarkdown": "I think it's worth to mention a **dropout** technique and **data augmentation**, which also really help to keep the curve smooth."
    },
    {
      "id": 895045,
      "postDate": "2020-06-21T04:14:27.790Z",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2F88bfec40e1eefa2ba1ebefb67b2b4f85%2FCapture.PNG?generation=1592712639799820&amp;alt=media\" alt=\"\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2Ff74ff305268b2a19a3330ac9b34c8d74%2F2.PNG?generation=1592712641453053&amp;alt=media\" alt=\"\">\nTo make the curve smooth, the things we should try out:\n1. <strong>regularization</strong>, so that the overfitting does not occur\n2. <strong>EarlyStopping callback</strong>, we can use this callback using tensorflow.keras and set the monitor to val_loss or val_accuracy and set patience = 3/4/5 so that the code stops executing whenever there is a failure in the improvement of results.</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2F88bfec40e1eefa2ba1ebefb67b2b4f85%2FCapture.PNG?generation=1592712639799820&amp;alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2Ff74ff305268b2a19a3330ac9b34c8d74%2F2.PNG?generation=1592712641453053&amp;alt=media)\nTo make the curve smooth, the things we should try out:\n1. **regularization**, so that the overfitting does not occur\n2. **EarlyStopping callback**, we can use this callback using tensorflow.keras and set the monitor to val_loss or val_accuracy and set patience = 3/4/5 so that the code stops executing whenever there is a failure in the improvement of results."
    }
  ],
  "comments": [
    {
      "id": 932216,
      "author_name": "Szymon Bednorz",
      "author_url": "",
      "post_date": "2020-07-16T20:25:50.997000",
      "content": "<p>I think it's worth to mention a <strong>dropout</strong> technique and <strong>data augmentation</strong>, which also really help to keep the curve smooth.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "932216": "I think it's worth to mention a **dropout** technique and **data augmentation**, which also really help to keep the curve smooth.",
    "895045": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2F88bfec40e1eefa2ba1ebefb67b2b4f85%2FCapture.PNG?generation=1592712639799820&amp;alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F5048836%2Ff74ff305268b2a19a3330ac9b34c8d74%2F2.PNG?generation=1592712641453053&amp;alt=media)\nTo make the curve smooth, the things we should try out:\n1. **regularization**, so that the overfitting does not occur\n2. **EarlyStopping callback**, we can use this callback using tensorflow.keras and set the monitor to val_loss or val_accuracy and set patience = 3/4/5 so that the code stops executing whenever there is a failure in the improvement of results."
  }
}