{
  "id": 315368,
  "title": "Someone knows if there is any way to check the compatibility of libraries with the tensorflow version",
  "url": "/competitions/happy-whale-and-dolphin/discussion/315368",
  "author_name": "",
  "post_date": "2022-03-27T18:46:32.315615800Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everybody! <br>\nI often encounter the problem that some libraries are not compatible with this or that version of tensorflow, and if you use TPU, the problem becomes terrible because some libraries do not support TPU. </p>\n<p>Now the question is, what is a universal way to check libraries with which version of TPU they are compatible?</p>\n<p>The latest bug with Nan in this race is just a shock, in fact it means that the prediction results of your models may differ due to the tf version with the same model weights.</p>\n<p>I don't work as a DS, it's just a hobby for me, it would be interesting to hear the opinions of people who come across this every day.</p>",
  "messages": [
    {
      "id": "1736808",
      "postDate": "03/27/2022 18:46:32",
      "content": "<p>Hello everybody! <br>\nI often encounter the problem that some libraries are not compatible with this or that version of tensorflow, and if you use TPU, the problem becomes terrible because some libraries do not support TPU. </p>\n<p>Now the question is, what is a universal way to check libraries with which version of TPU they are compatible?</p>\n<p>The latest bug with Nan in this race is just a shock, in fact it means that the prediction results of your models may differ due to the tf version with the same model weights.</p>\n<p>I don't work as a DS, it's just a hobby for me, it would be interesting to hear the opinions of people who come across this every day.</p>",
      "rawMarkdown": "Hello everybody! \nI often encounter the problem that some libraries are not compatible with this or that version of tensorflow, and if you use TPU, the problem becomes terrible because some libraries do not support TPU. \n\nNow the question is, what is a universal way to check libraries with which version of TPU they are compatible?\n\nThe latest bug with Nan in this race is just a shock, in fact it means that the prediction results of your models may differ due to the tf version with the same model weights.\n\nI don't work as a DS, it's just a hobby for me, it would be interesting to hear the opinions of people who come across this every day.",
      "votes": null
    },
    {
      "id": "1737089",
      "postDate": "03/28/2022 05:26:58",
      "content": "<p>first I found NaN on Colab .  now  NaN on kaggle.  </p>",
      "rawMarkdown": "first I found NaN on Colab .  now  NaN on kaggle.",
      "votes": null
    },
    {
      "id": "1746549",
      "postDate": "04/05/2022 23:00:11",
      "content": "<p>I took a loooooong time but finally switching to docker solved most of my issues. I think you could use the Kaggle docker image everywhere and not have to worry about the setup versions. </p>",
      "rawMarkdown": "I took a loooooong time but finally switching to docker solved most of my issues. I think you could use the Kaggle docker image everywhere and not have to worry about the setup versions.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1737089,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "03/28/2022 05:26:58",
      "content": "<p>first I found NaN on Colab .  now  NaN on kaggle.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1746549,
      "author_name": "init27",
      "author_url": "",
      "post_date": "04/05/2022 23:00:11",
      "content": "<p>I took a loooooong time but finally switching to docker solved most of my issues. I think you could use the Kaggle docker image everywhere and not have to worry about the setup versions. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1736808": "Hello everybody! \nI often encounter the problem that some libraries are not compatible with this or that version of tensorflow, and if you use TPU, the problem becomes terrible because some libraries do not support TPU. \n\nNow the question is, what is a universal way to check libraries with which version of TPU they are compatible?\n\nThe latest bug with Nan in this race is just a shock, in fact it means that the prediction results of your models may differ due to the tf version with the same model weights.\n\nI don't work as a DS, it's just a hobby for me, it would be interesting to hear the opinions of people who come across this every day.",
    "1737089": "first I found NaN on Colab .  now  NaN on kaggle.",
    "1746549": "I took a loooooong time but finally switching to docker solved most of my issues. I think you could use the Kaggle docker image everywhere and not have to worry about the setup versions."
  },
  "source": "meta"
}