{
  "id": 159015,
  "title": "Using colab vs kaggle notebooks",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/159015",
  "author_name": "",
  "post_date": "2020-06-16T05:47:16.672436700Z",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I'm new to Kaggle and a frequent user of colab with a pro account. Based on your experience, what could be the better platform to run models for this problem (colab or kaggle notebooks)?</p>",
  "messages": [
    {
      "id": "888089",
      "postDate": "06/16/2020 05:47:16",
      "content": "<p>I'm new to Kaggle and a frequent user of colab with a pro account. Based on your experience, what could be the better platform to run models for this problem (colab or kaggle notebooks)?</p>",
      "rawMarkdown": "I'm new to Kaggle and a frequent user of colab with a pro account. Based on your experience, what could be the better platform to run models for this problem (colab or kaggle notebooks)?",
      "votes": null
    },
    {
      "id": "888118",
      "postDate": "06/16/2020 06:26:48",
      "content": "<p>Well, colab is little bit slower but we can get more runtime there. Where in kaggle with TPU we can get only 3hr of Runtime</p>",
      "rawMarkdown": "Well, colab is little bit slower but we can get more runtime there. Where in kaggle with TPU we can get only 3hr of Runtime",
      "votes": null
    },
    {
      "id": "888161",
      "postDate": "06/16/2020 06:57:37",
      "content": "<p>Thanks a lot. Do you have a rough estimation on how much time will go to train full data set with a standard CNN on Kagge TPUs? I know that depends on the architecture. I'm just trying to have a very rough idea on the expectations. Is that few minutes or couple of hours?</p>",
      "rawMarkdown": "Thanks a lot. Do you have a rough estimation on how much time will go to train full data set with a standard CNN on Kagge TPUs? I know that depends on the architecture. I'm just trying to have a very rough idea on the expectations. Is that few minutes or couple of hours?",
      "votes": null
    },
    {
      "id": "888238",
      "postDate": "06/16/2020 08:06:39",
      "content": "<p>It totally depends on the model and the dataset. Just for example, with EfficientNetB3 it takes 3min/epoch for B7 it takes around 7.5 min/epoch with 512x512 in kaggle and for colab it added 2-3 mins more per epoch</p>",
      "rawMarkdown": "It totally depends on the model and the dataset. Just for example, with EfficientNetB3 it takes 3min/epoch for B7 it takes around 7.5 min/epoch with 512x512 in kaggle and for colab it added 2-3 mins more per epoch",
      "votes": null
    },
    {
      "id": "888600",
      "postDate": "06/16/2020 13:05:03",
      "content": "<p>Thanks. That gives an idea.</p>",
      "rawMarkdown": "Thanks. That gives an idea.",
      "votes": null
    },
    {
      "id": "911519",
      "postDate": "07/01/2020 19:46:31",
      "content": "<p>Being a pro user on colab, do you think there's anything missing in it or something you don't like? I believe the GPU on colab suffices most of your tasks. But would love to know from your experience if there's any pain point in it?</p>",
      "rawMarkdown": "Being a pro user on colab, do you think there's anything missing in it or something you don't like? I believe the GPU on colab suffices most of your tasks. But would love to know from your experience if there's any pain point in it?",
      "votes": null
    },
    {
      "id": "911713",
      "postDate": "07/02/2020 01:00:39",
      "content": "<p>Both are quite same ! But as you mentioned you are the pro user of google colab then you must have the access to the powerful gpu's and other hardware.\n Hence, i'll recommend you to stay on the google colab </p>",
      "rawMarkdown": "Both are quite same ! But as you mentioned you are the pro user of google colab then you must have the access to the powerful gpu's and other hardware.\n Hence, i'll recommend you to stay on the google colab",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 888118,
      "author_name": "awsaf49",
      "author_url": "",
      "post_date": "06/16/2020 06:26:48",
      "content": "<p>Well, colab is little bit slower but we can get more runtime there. Where in kaggle with TPU we can get only 3hr of Runtime</p>",
      "votes": null,
      "replies": [
        {
          "id": 888161,
          "author_name": "dimuthuupe1990",
          "author_url": "",
          "post_date": "06/16/2020 06:57:37",
          "content": "<p>Thanks a lot. Do you have a rough estimation on how much time will go to train full data set with a standard CNN on Kagge TPUs? I know that depends on the architecture. I'm just trying to have a very rough idea on the expectations. Is that few minutes or couple of hours?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 888238,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "06/16/2020 08:06:39",
          "content": "<p>It totally depends on the model and the dataset. Just for example, with EfficientNetB3 it takes 3min/epoch for B7 it takes around 7.5 min/epoch with 512x512 in kaggle and for colab it added 2-3 mins more per epoch</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 888600,
          "author_name": "dimuthuupe1990",
          "author_url": "",
          "post_date": "06/16/2020 13:05:03",
          "content": "<p>Thanks. That gives an idea.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 911519,
      "author_name": "genesis96839",
      "author_url": "",
      "post_date": "07/01/2020 19:46:31",
      "content": "<p>Being a pro user on colab, do you think there's anything missing in it or something you don't like? I believe the GPU on colab suffices most of your tasks. But would love to know from your experience if there's any pain point in it?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 911713,
      "author_name": "sudarshanpatil",
      "author_url": "",
      "post_date": "07/02/2020 01:00:39",
      "content": "<p>Both are quite same ! But as you mentioned you are the pro user of google colab then you must have the access to the powerful gpu's and other hardware.\n Hence, i'll recommend you to stay on the google colab </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "888089": "I'm new to Kaggle and a frequent user of colab with a pro account. Based on your experience, what could be the better platform to run models for this problem (colab or kaggle notebooks)?",
    "888118": "Well, colab is little bit slower but we can get more runtime there. Where in kaggle with TPU we can get only 3hr of Runtime",
    "888161": "Thanks a lot. Do you have a rough estimation on how much time will go to train full data set with a standard CNN on Kagge TPUs? I know that depends on the architecture. I'm just trying to have a very rough idea on the expectations. Is that few minutes or couple of hours?",
    "888238": "It totally depends on the model and the dataset. Just for example, with EfficientNetB3 it takes 3min/epoch for B7 it takes around 7.5 min/epoch with 512x512 in kaggle and for colab it added 2-3 mins more per epoch",
    "888600": "Thanks. That gives an idea.",
    "911519": "Being a pro user on colab, do you think there's anything missing in it or something you don't like? I believe the GPU on colab suffices most of your tasks. But would love to know from your experience if there's any pain point in it?",
    "911713": "Both are quite same ! But as you mentioned you are the pro user of google colab then you must have the access to the powerful gpu's and other hardware.\n Hence, i'll recommend you to stay on the google colab"
  },
  "source": "meta"
}