{
  "id": 90145,
  "title": "What Deep Learning Framework to use ?",
  "url": "/competitions/imet-2019-fgvc6/discussion/90145",
  "author_name": "",
  "post_date": "2019-04-20T22:19:44.295828500Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I love using <strong>keras</strong> with tensorflow as backend because it's easy and fast.\nIf I need some advanced things I use tensorflow (I don't like it so much)\nSomeone told me that I have to master Tensorflow (because of its production deployment simplicity ), Keras, and Pytorch (because of its dynamic graph) to have a good position between the Deep learning developer's.</p>\n\n<p>What about you?</p>",
  "messages": [
    {
      "id": "520394",
      "postDate": "04/20/2019 22:19:44",
      "content": "<p>I love using <strong>keras</strong> with tensorflow as backend because it's easy and fast.\nIf I need some advanced things I use tensorflow (I don't like it so much)\nSomeone told me that I have to master Tensorflow (because of its production deployment simplicity ), Keras, and Pytorch (because of its dynamic graph) to have a good position between the Deep learning developer's.</p>\n\n<p>What about you?</p>",
      "rawMarkdown": "I love using **keras** with tensorflow as backend because it's easy and fast.\nIf I need some advanced things I use tensorflow (I don't like it so much)\nSomeone told me that I have to master Tensorflow (because of its production deployment simplicity ), Keras, and Pytorch (because of its dynamic graph) to have a good position between the Deep learning developer's.\n\nWhat about you?",
      "votes": null
    },
    {
      "id": "520455",
      "postDate": "04/21/2019 02:46:09",
      "content": "<p>I love pytorch, because it has brief  APIs and it is easy to use. But in some Kaggle Competition, many person use fastai which is a higher-API of pytorch. Fastai is born for competition and we can use several minutes to build a program to solve a problem using it.</p>",
      "rawMarkdown": "I love pytorch, because it has brief  APIs and it is easy to use. But in some Kaggle Competition, many person use fastai which is a higher-API of pytorch. Fastai is born for competition and we can use several minutes to build a program to solve a problem using it.",
      "votes": null
    },
    {
      "id": "520799",
      "postDate": "04/21/2019 19:04:37",
      "content": "<p>Pytorch is cool too, but fast.ai I find it more obscure</p>",
      "rawMarkdown": "Pytorch is cool too, but fast.ai I find it more obscure",
      "votes": null
    },
    {
      "id": "521764",
      "postDate": "04/23/2019 11:47:22",
      "content": "<p>Though Tensorflow may be simple to deploy to production, it is not simple to try different models using it. So for competition purposes Keras and Pytorch are more preferable, I suppose. The same is Fast.AI, this is even higher level API which uses Pytorch inside.</p>",
      "rawMarkdown": "Though Tensorflow may be simple to deploy to production, it is not simple to try different models using it. So for competition purposes Keras and Pytorch are more preferable, I suppose. The same is Fast.AI, this is even higher level API which uses Pytorch inside.",
      "votes": null
    },
    {
      "id": "524341",
      "postDate": "04/28/2019 14:47:16",
      "content": "<p>I also study mainly keras and tensorflow, but when I check the pubic kernel I think I have the opportunity to learn pytorch naturally. Especially in Japan, chainer is one of the famous frameworks.\nThis↓ kernel is interesting and helpful as a result of research on the deep learning framework.\n<a href=\"https://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018\">https://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018</a>\nI also want a 2019 version:)</p>",
      "rawMarkdown": "I also study mainly keras and tensorflow, but when I check the pubic kernel I think I have the opportunity to learn pytorch naturally. Especially in Japan, chainer is one of the famous frameworks.\nThis↓ kernel is interesting and helpful as a result of research on the deep learning framework.\nhttps://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018\nI also want a 2019 version:)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 520455,
      "author_name": "seefun",
      "author_url": "",
      "post_date": "04/21/2019 02:46:09",
      "content": "<p>I love pytorch, because it has brief  APIs and it is easy to use. But in some Kaggle Competition, many person use fastai which is a higher-API of pytorch. Fastai is born for competition and we can use several minutes to build a program to solve a problem using it.</p>",
      "votes": null,
      "replies": [
        {
          "id": 520799,
          "author_name": "jmourad100",
          "author_url": "",
          "post_date": "04/21/2019 19:04:37",
          "content": "<p>Pytorch is cool too, but fast.ai I find it more obscure</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 521764,
      "author_name": "demonplus",
      "author_url": "",
      "post_date": "04/23/2019 11:47:22",
      "content": "<p>Though Tensorflow may be simple to deploy to production, it is not simple to try different models using it. So for competition purposes Keras and Pytorch are more preferable, I suppose. The same is Fast.AI, this is even higher level API which uses Pytorch inside.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 524341,
      "author_name": "dhaqui",
      "author_url": "",
      "post_date": "04/28/2019 14:47:16",
      "content": "<p>I also study mainly keras and tensorflow, but when I check the pubic kernel I think I have the opportunity to learn pytorch naturally. Especially in Japan, chainer is one of the famous frameworks.\nThis↓ kernel is interesting and helpful as a result of research on the deep learning framework.\n<a href=\"https://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018\">https://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018</a>\nI also want a 2019 version:)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "520394": "I love using **keras** with tensorflow as backend because it's easy and fast.\nIf I need some advanced things I use tensorflow (I don't like it so much)\nSomeone told me that I have to master Tensorflow (because of its production deployment simplicity ), Keras, and Pytorch (because of its dynamic graph) to have a good position between the Deep learning developer's.\n\nWhat about you?",
    "520455": "I love pytorch, because it has brief  APIs and it is easy to use. But in some Kaggle Competition, many person use fastai which is a higher-API of pytorch. Fastai is born for competition and we can use several minutes to build a program to solve a problem using it.",
    "520799": "Pytorch is cool too, but fast.ai I find it more obscure",
    "521764": "Though Tensorflow may be simple to deploy to production, it is not simple to try different models using it. So for competition purposes Keras and Pytorch are more preferable, I suppose. The same is Fast.AI, this is even higher level API which uses Pytorch inside.",
    "524341": "I also study mainly keras and tensorflow, but when I check the pubic kernel I think I have the opportunity to learn pytorch naturally. Especially in Japan, chainer is one of the famous frameworks.\nThis↓ kernel is interesting and helpful as a result of research on the deep learning framework.\nhttps://www.kaggle.com/discdiver/deep-learning-framework-power-scores-2018\nI also want a 2019 version:)"
  },
  "source": "meta"
}