{
  "id": 182549,
  "title": "Comparison of the Networks",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/182549",
  "author_name": "Praveen Kumar",
  "post_date": "2020-09-13T10:24:22.250000",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<ul>\n<li>From my experience, GRUs train faster and perform better than LSTMs on less training data if you are doing language modeling (not sure about other tasks).</li>\n<li>GRUs are simpler and thus easier to modify, for example adding new gates in case of additional input to the network. It’s just less code in general.</li>\n<li>LSTMs should, in theory, remember longer sequences than GRUs and outperform them in tasks requiring modeling long-distance relations.</li>\n<li>The GRUs also have less parameter complexity than LSTM which can be seen from the model summaries above.</li>\n<li>The simple RNNs only have simple recurrent operations without any gates to control the flow of information among the cells.</li>\n</ul>",
  "messages": [
    {
      "id": 1008701,
      "postDate": "2020-09-13T10:24:22.250Z",
      "content": "<ul>\n<li>From my experience, GRUs train faster and perform better than LSTMs on less training data if you are doing language modeling (not sure about other tasks).</li>\n<li>GRUs are simpler and thus easier to modify, for example adding new gates in case of additional input to the network. It’s just less code in general.</li>\n<li>LSTMs should, in theory, remember longer sequences than GRUs and outperform them in tasks requiring modeling long-distance relations.</li>\n<li>The GRUs also have less parameter complexity than LSTM which can be seen from the model summaries above.</li>\n<li>The simple RNNs only have simple recurrent operations without any gates to control the flow of information among the cells.</li>\n</ul>",
      "rawMarkdown": "- From my experience, GRUs train faster and perform better than LSTMs on less training data if you are doing language modeling (not sure about other tasks).\n- GRUs are simpler and thus easier to modify, for example adding new gates in case of additional input to the network. It’s just less code in general.\n- LSTMs should, in theory, remember longer sequences than GRUs and outperform them in tasks requiring modeling long-distance relations.\n- The GRUs also have less parameter complexity than LSTM which can be seen from the model summaries above.\n- The simple RNNs only have simple recurrent operations without any gates to control the flow of information among the cells.",
      "votes": 5
    },
    {
      "id": 1058693,
      "postDate": "2020-10-24T05:47:19.147Z",
      "content": "<p>Has anyone tried LSTM in this competition?<br>\nLast year's winners have used LSTM… checking in if it was similarly helpful in the current competition…</p>",
      "rawMarkdown": "Has anyone tried LSTM in this competition?\nLast year's winners have used LSTM... checking in if it was similarly helpful in the current competition..."
    },
    {
      "id": 1008751,
      "postDate": "2020-09-13T11:04:01.090Z",
      "content": "<p>Nice explanation 👍<br>\nUpvoted.</p>",
      "rawMarkdown": "Nice explanation 👍\nUpvoted."
    },
    {
      "id": 1008795,
      "postDate": "2020-09-13T12:00:42.827Z",
      "content": "<p><a href=\"https://www.kaggle.com/oneplustricks\" target=\"_blank\">@oneplustricks</a>  Thank you</p>",
      "rawMarkdown": "@oneplustricks  Thank you",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1058693,
      "author_name": "Kamal Das",
      "author_url": "",
      "post_date": "2020-10-24T05:47:19.147000",
      "content": "<p>Has anyone tried LSTM in this competition?<br>\nLast year's winners have used LSTM… checking in if it was similarly helpful in the current competition…</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1008751,
      "author_name": "Parth Chhabra",
      "author_url": "",
      "post_date": "2020-09-13T11:04:01.090000",
      "content": "<p>Nice explanation 👍<br>\nUpvoted.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1008795,
      "author_name": "Praveen Kumar",
      "author_url": "",
      "post_date": "2020-09-13T12:00:42.827000",
      "content": "<p><a href=\"https://www.kaggle.com/oneplustricks\" target=\"_blank\">@oneplustricks</a>  Thank you</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1008701": "- From my experience, GRUs train faster and perform better than LSTMs on less training data if you are doing language modeling (not sure about other tasks).\n- GRUs are simpler and thus easier to modify, for example adding new gates in case of additional input to the network. It’s just less code in general.\n- LSTMs should, in theory, remember longer sequences than GRUs and outperform them in tasks requiring modeling long-distance relations.\n- The GRUs also have less parameter complexity than LSTM which can be seen from the model summaries above.\n- The simple RNNs only have simple recurrent operations without any gates to control the flow of information among the cells.",
    "1058693": "Has anyone tried LSTM in this competition?\nLast year's winners have used LSTM... checking in if it was similarly helpful in the current competition...",
    "1008751": "Nice explanation 👍\nUpvoted.",
    "1008795": "@oneplustricks  Thank you"
  }
}