{
  "id": 71756,
  "title": "Overfitting?? Any ideas.",
  "url": "/competitions/quora-insincere-questions-classification/discussion/71756",
  "author_name": "",
  "post_date": "2018-11-16T08:16:54.670778400Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>The LSTM Model starts to overfit after 3-4 epochs in keras, any any solution to prevent overfitting and improve performance</p>",
  "messages": [
    {
      "id": "422440",
      "postDate": "11/16/2018 08:16:54",
      "content": "<p>The LSTM Model starts to overfit after 3-4 epochs in keras, any any solution to prevent overfitting and improve performance</p>",
      "rawMarkdown": "The LSTM Model starts to overfit after 3-4 epochs in keras, any any solution to prevent overfitting and improve performance",
      "votes": null
    },
    {
      "id": "422459",
      "postDate": "11/16/2018 08:51:06",
      "content": "<p>I use SpatialDropout1D after emb,\nDropout after concatenate,\nSmaller vocabulary,\nbut it starts to overfit after 4-5 epochs</p>",
      "rawMarkdown": "I use SpatialDropout1D after emb,\nDropout after concatenate,\nSmaller vocabulary,\nbut it starts to overfit after 4-5 epochs",
      "votes": null
    },
    {
      "id": "422475",
      "postDate": "11/16/2018 09:13:54",
      "content": "<p>Which is better SpatialDropout1D or Dropout </p>",
      "rawMarkdown": "Which is better SpatialDropout1D or Dropout",
      "votes": null
    },
    {
      "id": "422477",
      "postDate": "11/16/2018 09:19:56",
      "content": "<p>SpatialDropout1D  i use 0.1-0.2,  dropout 0.4-0.5,\nThis may not be reasonable.</p>",
      "rawMarkdown": "SpatialDropout1D  i use 0.1-0.2,  dropout 0.4-0.5,\nThis may not be reasonable.",
      "votes": null
    },
    {
      "id": "422586",
      "postDate": "11/16/2018 12:55:50",
      "content": "<p>I use some callbacks to make sure I don't overfit too much, take a look to <a href=\"https://keras.io/callbacks/\">https://keras.io/callbacks/</a>\nI use ModelCheckpoint mostly, but you can use EarlyStopping as well. </p>\n\n<p>Otherwise, if you want to modify your model, you can use Dropout as mentionned by @ggsddu. Maxpooling is fine as well I guess.</p>",
      "rawMarkdown": "I use some callbacks to make sure I don't overfit too much, take a look to https://keras.io/callbacks/\nI use ModelCheckpoint mostly, but you can use EarlyStopping as well. \n\nOtherwise, if you want to modify your model, you can use Dropout as mentionned by @ggsddu. Maxpooling is fine as well I guess.",
      "votes": null
    },
    {
      "id": "422940",
      "postDate": "11/17/2018 05:39:27",
      "content": "<p>Increasing Batchsize helps</p>",
      "rawMarkdown": "Increasing Batchsize helps",
      "votes": null
    },
    {
      "id": "423068",
      "postDate": "11/17/2018 12:03:27",
      "content": "<p>Maybe you can try earlyStoping.</p>",
      "rawMarkdown": "Maybe you can try earlyStoping.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 422459,
      "author_name": "tianshuaifei",
      "author_url": "",
      "post_date": "11/16/2018 08:51:06",
      "content": "<p>I use SpatialDropout1D after emb,\nDropout after concatenate,\nSmaller vocabulary,\nbut it starts to overfit after 4-5 epochs</p>",
      "votes": null,
      "replies": [
        {
          "id": 422475,
          "author_name": "nikhilroxtomar",
          "author_url": "",
          "post_date": "11/16/2018 09:13:54",
          "content": "<p>Which is better SpatialDropout1D or Dropout </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 422477,
          "author_name": "tianshuaifei",
          "author_url": "",
          "post_date": "11/16/2018 09:19:56",
          "content": "<p>SpatialDropout1D  i use 0.1-0.2,  dropout 0.4-0.5,\nThis may not be reasonable.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 422586,
      "author_name": "theoviel",
      "author_url": "",
      "post_date": "11/16/2018 12:55:50",
      "content": "<p>I use some callbacks to make sure I don't overfit too much, take a look to <a href=\"https://keras.io/callbacks/\">https://keras.io/callbacks/</a>\nI use ModelCheckpoint mostly, but you can use EarlyStopping as well. </p>\n\n<p>Otherwise, if you want to modify your model, you can use Dropout as mentionned by @ggsddu. Maxpooling is fine as well I guess.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 422940,
      "author_name": "christofhenkel",
      "author_url": "",
      "post_date": "11/17/2018 05:39:27",
      "content": "<p>Increasing Batchsize helps</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 423068,
      "author_name": "xiaobai1123q",
      "author_url": "",
      "post_date": "11/17/2018 12:03:27",
      "content": "<p>Maybe you can try earlyStoping.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "422440": "The LSTM Model starts to overfit after 3-4 epochs in keras, any any solution to prevent overfitting and improve performance",
    "422459": "I use SpatialDropout1D after emb,\nDropout after concatenate,\nSmaller vocabulary,\nbut it starts to overfit after 4-5 epochs",
    "422475": "Which is better SpatialDropout1D or Dropout",
    "422477": "SpatialDropout1D  i use 0.1-0.2,  dropout 0.4-0.5,\nThis may not be reasonable.",
    "422586": "I use some callbacks to make sure I don't overfit too much, take a look to https://keras.io/callbacks/\nI use ModelCheckpoint mostly, but you can use EarlyStopping as well. \n\nOtherwise, if you want to modify your model, you can use Dropout as mentionned by @ggsddu. Maxpooling is fine as well I guess.",
    "422940": "Increasing Batchsize helps",
    "423068": "Maybe you can try earlyStoping."
  },
  "source": "meta"
}