{
  "id": 20231,
  "title": "Study Material on Keras and Theano",
  "url": "/competitions/expedia-hotel-recommendations/discussion/20231",
  "author_name": "",
  "post_date": "2016-04-18T17:24:22.170Z",
  "votes": 1,
  "comment_count": 6,
  "views": 1813,
  "content": "<p>Hi..</p>\n\n<p>It will be helpful ,if I can get suggestions regarding good basic tutorials (books, video, MOOC etc) on Keras and Theano. I am trying to build up my knowledge on this , so that I can apply it on the problem set.</p>\n\n<p>Thanks a lot.. </p>",
  "messages": [
    {
      "id": "115431",
      "postDate": "04/18/2016 17:24:22",
      "content": "<p>Hi..</p>\n\n<p>It will be helpful ,if I can get suggestions regarding good basic tutorials (books, video, MOOC etc) on Keras and Theano. I am trying to build up my knowledge on this , so that I can apply it on the problem set.</p>\n\n<p>Thanks a lot.. </p>",
      "rawMarkdown": "Hi..\r\n\r\nIt will be helpful ,if I can get suggestions regarding good basic tutorials (books, video, MOOC etc) on Keras and Theano. I am trying to build up my knowledge on this , so that I can apply it on the problem set.\r\n\r\nThanks a lot..",
      "votes": null
    },
    {
      "id": "118787",
      "postDate": "05/05/2016 10:13:24",
      "content": "<p>Shan, \nGo to Keras github location and see the examples directory. They have solutions to a number of Kaggle competitions using Keras leveraging Theano backend. </p>\n\n<p>Hope this helps. </p>\n\n<p>Pradeep</p>",
      "rawMarkdown": "Shan, \r\nGo to Keras github location and see the examples directory. They have solutions to a number of Kaggle competitions using Keras leveraging Theano backend. \r\n\r\nHope this helps. \r\n\r\nPradeep",
      "votes": null
    },
    {
      "id": "118819",
      "postDate": "05/05/2016 14:45:12",
      "content": "<p>Shan, I have found Keras and Theano relatively straightforward to use <em>as long as</em> one understands the theory behind it correctly. I can point you to good docs on the underlying methods if you'd like. We can also team up in applying keras methods to this problem, I am interested in it too. Let me know. </p>",
      "rawMarkdown": "Shan, I have found Keras and Theano relatively straightforward to use *as long as* one understands the theory behind it correctly. I can point you to good docs on the underlying methods if you'd like. We can also team up in applying keras methods to this problem, I am interested in it too. Let me know.",
      "votes": null
    },
    {
      "id": "118852",
      "postDate": "05/05/2016 17:47:35",
      "content": "<p>Thank you @PradeepHegde . The docs are informative. Would also like to go through some video tutorials if you can suggest.</p>\n\n<p>Thank you @BenjaminTannenbaum. Sure. Would like to go through docs and videos. You may mail me:  shan2016exp@gmail.com</p>",
      "rawMarkdown": "Thank you @PradeepHegde . The docs are informative. Would also like to go through some video tutorials if you can suggest.\r\n\r\nThank you @BenjaminTannenbaum. Sure. Would like to go through docs and videos. You may mail me:  shan2016exp@gmail.com",
      "votes": null
    },
    {
      "id": "118861",
      "postDate": "05/05/2016 19:30:32",
      "content": "<p>Hi Shan, \nI found <a href=\"http://benjaminbolte.com/blog/2016/keras-language-modeling.html\" title=\"this one\">this one</a> very useful for starting.</p>",
      "rawMarkdown": "Hi Shan, \r\nI found [this one][1] very useful for starting.\r\n\r\n\r\n  [1]: http://benjaminbolte.com/blog/2016/keras-language-modeling.html \"this one\"",
      "votes": null
    },
    {
      "id": "118957",
      "postDate": "05/06/2016 11:01:21",
      "content": "<p>Hi, maybe you'll find this useful.</p>\n\n<p>There is a <a href=\"https://www.kaggle.com/c/rossmann-store-sales/forums/t/17974/code-sharing-3rd-place-category-embedding-with-deep-neural-network/103140#post103140\">forum post</a> where @entron shared his code for 3rd place in Rossmann competition using neural networks with entity embeddings. There's also a link to their article on the subject. I thought this might be a good fit for this competition since there's lots of high cardinality categorical features. </p>\n\n<p>So far I've been able to produce 0.30 local / 0.277 lb score with a neural net model based on @entron's code, but I'm not using all the features yet and definitely don't have the best parameters. Maybe there's yet room for growth =).</p>",
      "rawMarkdown": "Hi, maybe you'll find this useful.\r\n\r\nThere is a [forum post](https://www.kaggle.com/c/rossmann-store-sales/forums/t/17974/code-sharing-3rd-place-category-embedding-with-deep-neural-network/103140#post103140) where @entron shared his code for 3rd place in Rossmann competition using neural networks with entity embeddings. There's also a link to their article on the subject. I thought this might be a good fit for this competition since there's lots of high cardinality categorical features. \r\n\r\nSo far I've been able to produce 0.30 local / 0.277 lb score with a neural net model based on @entron's code, but I'm not using all the features yet and definitely don't have the best parameters. Maybe there's yet room for growth =).",
      "votes": null
    },
    {
      "id": "121589",
      "postDate": "05/27/2016 16:10:44",
      "content": "<p>@dune_dweller Thank you for sharing. Did you do this by downsampling or use the whole train set? And what's your method to handle such big dataset? Thank you.</p>",
      "rawMarkdown": "dune_dweller Thank you for sharing. Did you do this by downsampling or use the whole train set? And what's your method to handle such big dataset? Thank you.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 118787,
      "author_name": "phegde",
      "author_url": "",
      "post_date": "05/05/2016 10:13:24",
      "content": "<p>Shan, \nGo to Keras github location and see the examples directory. They have solutions to a number of Kaggle competitions using Keras leveraging Theano backend. </p>\n\n<p>Hope this helps. </p>\n\n<p>Pradeep</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 118819,
      "author_name": "bentannenbaum",
      "author_url": "",
      "post_date": "05/05/2016 14:45:12",
      "content": "<p>Shan, I have found Keras and Theano relatively straightforward to use <em>as long as</em> one understands the theory behind it correctly. I can point you to good docs on the underlying methods if you'd like. We can also team up in applying keras methods to this problem, I am interested in it too. Let me know. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 118852,
      "author_name": "shan4224",
      "author_url": "",
      "post_date": "05/05/2016 17:47:35",
      "content": "<p>Thank you @PradeepHegde . The docs are informative. Would also like to go through some video tutorials if you can suggest.</p>\n\n<p>Thank you @BenjaminTannenbaum. Sure. Would like to go through docs and videos. You may mail me:  shan2016exp@gmail.com</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 118861,
      "author_name": "zarnold",
      "author_url": "",
      "post_date": "05/05/2016 19:30:32",
      "content": "<p>Hi Shan, \nI found <a href=\"http://benjaminbolte.com/blog/2016/keras-language-modeling.html\" title=\"this one\">this one</a> very useful for starting.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 118957,
      "author_name": "dvasyukova",
      "author_url": "",
      "post_date": "05/06/2016 11:01:21",
      "content": "<p>Hi, maybe you'll find this useful.</p>\n\n<p>There is a <a href=\"https://www.kaggle.com/c/rossmann-store-sales/forums/t/17974/code-sharing-3rd-place-category-embedding-with-deep-neural-network/103140#post103140\">forum post</a> where @entron shared his code for 3rd place in Rossmann competition using neural networks with entity embeddings. There's also a link to their article on the subject. I thought this might be a good fit for this competition since there's lots of high cardinality categorical features. </p>\n\n<p>So far I've been able to produce 0.30 local / 0.277 lb score with a neural net model based on @entron's code, but I'm not using all the features yet and definitely don't have the best parameters. Maybe there's yet room for growth =).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 121589,
      "author_name": "beedata",
      "author_url": "",
      "post_date": "05/27/2016 16:10:44",
      "content": "<p>@dune_dweller Thank you for sharing. Did you do this by downsampling or use the whole train set? And what's your method to handle such big dataset? Thank you.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "115431": "Hi..\r\n\r\nIt will be helpful ,if I can get suggestions regarding good basic tutorials (books, video, MOOC etc) on Keras and Theano. I am trying to build up my knowledge on this , so that I can apply it on the problem set.\r\n\r\nThanks a lot..",
    "118787": "Shan, \r\nGo to Keras github location and see the examples directory. They have solutions to a number of Kaggle competitions using Keras leveraging Theano backend. \r\n\r\nHope this helps. \r\n\r\nPradeep",
    "118819": "Shan, I have found Keras and Theano relatively straightforward to use *as long as* one understands the theory behind it correctly. I can point you to good docs on the underlying methods if you'd like. We can also team up in applying keras methods to this problem, I am interested in it too. Let me know.",
    "118852": "Thank you @PradeepHegde . The docs are informative. Would also like to go through some video tutorials if you can suggest.\r\n\r\nThank you @BenjaminTannenbaum. Sure. Would like to go through docs and videos. You may mail me:  shan2016exp@gmail.com",
    "118861": "Hi Shan, \r\nI found [this one][1] very useful for starting.\r\n\r\n\r\n  [1]: http://benjaminbolte.com/blog/2016/keras-language-modeling.html \"this one\"",
    "118957": "Hi, maybe you'll find this useful.\r\n\r\nThere is a [forum post](https://www.kaggle.com/c/rossmann-store-sales/forums/t/17974/code-sharing-3rd-place-category-embedding-with-deep-neural-network/103140#post103140) where @entron shared his code for 3rd place in Rossmann competition using neural networks with entity embeddings. There's also a link to their article on the subject. I thought this might be a good fit for this competition since there's lots of high cardinality categorical features. \r\n\r\nSo far I've been able to produce 0.30 local / 0.277 lb score with a neural net model based on @entron's code, but I'm not using all the features yet and definitely don't have the best parameters. Maybe there's yet room for growth =).",
    "121589": "dune_dweller Thank you for sharing. Did you do this by downsampling or use the whole train set? And what's your method to handle such big dataset? Thank you."
  },
  "source": "meta"
}