{
  "id": 78529,
  "title": "This is really weired",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/78529",
  "author_name": "",
  "post_date": "2019-01-25T03:47:06.380159100Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Every time i am running my model i am getting very varied results on LB\nIs there anyone who is able to get a stable model ?</p>",
  "messages": [
    {
      "id": "461035",
      "postDate": "01/25/2019 03:47:06",
      "content": "<p>Every time i am running my model i am getting very varied results on LB\nIs there anyone who is able to get a stable model ?</p>",
      "rawMarkdown": "Every time i am running my model i am getting very varied results on LB\nIs there anyone who is able to get a stable model ?",
      "votes": null
    },
    {
      "id": "461043",
      "postDate": "01/25/2019 04:26:38",
      "content": "<p>Did you use CuDNNGRU / CuDNNLSTM with Keras? According <a href=\"https://machinelearningmastery.com/reproducible-results-neural-networks-keras/\">this post</a>, it seems that using Nvidia cuDNN in your stack may introduce additional sources of randomness and prevent the exact reproducibility of your results.  </p>\n\n<p>Here are some references, but I don't know too much how to avoid the randomness...  </p>\n\n<ul>\n<li><a href=\"https://github.com/keras-team/keras/issues/2280\">https://github.com/keras-team/keras/issues/2280</a>  </li>\n<li><a href=\"https://github.com/keras-team/keras/issues/2479#issuecomment-213987747\">https://github.com/keras-team/keras/issues/2479#issuecomment-213987747</a>  </li>\n</ul>",
      "rawMarkdown": "Did you use CuDNNGRU / CuDNNLSTM with Keras? According [this post](https://machinelearningmastery.com/reproducible-results-neural-networks-keras/), it seems that using Nvidia cuDNN in your stack may introduce additional sources of randomness and prevent the exact reproducibility of your results.  \n\nHere are some references, but I don't know too much how to avoid the randomness...  \n\n- https://github.com/keras-team/keras/issues/2280  \n- https://github.com/keras-team/keras/issues/2479#issuecomment-213987747",
      "votes": null
    },
    {
      "id": "461059",
      "postDate": "01/25/2019 05:57:12",
      "content": "<p>Make sure you fix the seeds. Even with different seeds you can see a lot of variation. you can use the following to make it reproducible.\n<a href=\"https://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development\">https://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development</a>.</p>",
      "rawMarkdown": "Make sure you fix the seeds. Even with different seeds you can see a lot of variation. you can use the following to make it reproducible.\nhttps://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 461043,
      "author_name": "takaishikawa",
      "author_url": "",
      "post_date": "01/25/2019 04:26:38",
      "content": "<p>Did you use CuDNNGRU / CuDNNLSTM with Keras? According <a href=\"https://machinelearningmastery.com/reproducible-results-neural-networks-keras/\">this post</a>, it seems that using Nvidia cuDNN in your stack may introduce additional sources of randomness and prevent the exact reproducibility of your results.  </p>\n\n<p>Here are some references, but I don't know too much how to avoid the randomness...  </p>\n\n<ul>\n<li><a href=\"https://github.com/keras-team/keras/issues/2280\">https://github.com/keras-team/keras/issues/2280</a>  </li>\n<li><a href=\"https://github.com/keras-team/keras/issues/2479#issuecomment-213987747\">https://github.com/keras-team/keras/issues/2479#issuecomment-213987747</a>  </li>\n</ul>",
      "votes": null,
      "replies": []
    },
    {
      "id": 461059,
      "author_name": "harshit92",
      "author_url": "",
      "post_date": "01/25/2019 05:57:12",
      "content": "<p>Make sure you fix the seeds. Even with different seeds you can see a lot of variation. you can use the following to make it reproducible.\n<a href=\"https://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development\">https://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development</a>.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "461035": "Every time i am running my model i am getting very varied results on LB\nIs there anyone who is able to get a stable model ?",
    "461043": "Did you use CuDNNGRU / CuDNNLSTM with Keras? According [this post](https://machinelearningmastery.com/reproducible-results-neural-networks-keras/), it seems that using Nvidia cuDNN in your stack may introduce additional sources of randomness and prevent the exact reproducibility of your results.  \n\nHere are some references, but I don't know too much how to avoid the randomness...  \n\n- https://github.com/keras-team/keras/issues/2280  \n- https://github.com/keras-team/keras/issues/2479#issuecomment-213987747",
    "461059": "Make sure you fix the seeds. Even with different seeds you can see a lot of variation. you can use the following to make it reproducible.\nhttps://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development."
  },
  "source": "meta"
}