{
  "id": 155963,
  "title": "TPU HTTP Connection Error",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/155963",
  "author_name": "",
  "post_date": "2020-06-03T19:30:41.819794800Z",
  "votes": 9,
  "comment_count": 17,
  "views": 0,
  "content": "<p>Is anyone else getting a TPU HTTP Connection error today?  </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F844a208edaa2c5b2263b69f5debec45d%2FScreen%20Shot%202020-06-03%20at%2012.28.32%20PM.png?generation=1591212524318603&amp;alt=media\" alt=\"\"></p>\n\n<p>I can load my Kaggle dataset TFRecords fine, but I get the above error message when trying to load the competition's <code>siim-isic-melanoma-classification</code> folder. I don't think i'm making a code mistake because to verify, I have forked 3 public notebooks but all 3 public notebooks fail today.</p>\n\n<p>This error occurs in TPU interactive notebook. (I haven't tried committing).</p>\n\n<p>===================</p>\n\n<p>UPDATE: Kaggle Team has responded below and hopefully fixed this issue. Thank you Timo.</p>",
  "messages": [
    {
      "id": "873093",
      "postDate": "06/03/2020 19:30:41",
      "content": "<p>Is anyone else getting a TPU HTTP Connection error today?  </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F844a208edaa2c5b2263b69f5debec45d%2FScreen%20Shot%202020-06-03%20at%2012.28.32%20PM.png?generation=1591212524318603&amp;alt=media\" alt=\"\"></p>\n\n<p>I can load my Kaggle dataset TFRecords fine, but I get the above error message when trying to load the competition's <code>siim-isic-melanoma-classification</code> folder. I don't think i'm making a code mistake because to verify, I have forked 3 public notebooks but all 3 public notebooks fail today.</p>\n\n<p>This error occurs in TPU interactive notebook. (I haven't tried committing).</p>\n\n<p>===================</p>\n\n<p>UPDATE: Kaggle Team has responded below and hopefully fixed this issue. Thank you Timo.</p>",
      "rawMarkdown": "Is anyone else getting a TPU HTTP Connection error today?  \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F844a208edaa2c5b2263b69f5debec45d%2FScreen%20Shot%202020-06-03%20at%2012.28.32%20PM.png?generation=1591212524318603&amp;alt=media)\n  \nI can load my Kaggle dataset TFRecords fine, but I get the above error message when trying to load the competition's `siim-isic-melanoma-classification` folder. I don't think i'm making a code mistake because to verify, I have forked 3 public notebooks but all 3 public notebooks fail today.\n\nThis error occurs in TPU interactive notebook. (I haven't tried committing).\n\n===================\n\nUPDATE: Kaggle Team has responded below and hopefully fixed this issue. Thank you Timo.",
      "votes": null
    },
    {
      "id": "873109",
      "postDate": "06/03/2020 19:52:00",
      "content": "<p>same - not only tpu - notonly interactive</p>",
      "rawMarkdown": "same - not only tpu - notonly interactive",
      "votes": null
    },
    {
      "id": "873131",
      "postDate": "06/03/2020 20:27:18",
      "content": "<p>get_gcs_path does not work. I can run on GPUs by directly reading from local dir</p>",
      "rawMarkdown": "get_gcs_path does not work. I can run on GPUs by directly reading from local dir",
      "votes": null
    },
    {
      "id": "873142",
      "postDate": "06/03/2020 20:47:04",
      "content": "<p><code>get_gcs_path</code> works for me if i load my own TFRecords from my Kaggle dataset <a href=\"https://www.kaggle.com/cdeotte/melanoma-512x512\">here</a>, but it won't load the competition TFRecords for me.</p>",
      "rawMarkdown": "`get_gcs_path` works for me if i load my own TFRecords from my Kaggle dataset [here][1], but it won't load the competition TFRecords for me.\n\n[1]: https://www.kaggle.com/cdeotte/melanoma-512x512",
      "votes": null
    },
    {
      "id": "873147",
      "postDate": "06/03/2020 21:01:55",
      "content": "<p>The same here. Trying to execute <code>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</code> results in a very long delay after which I get <code>HTTPError: HTTP Error 502: Bad Gateway</code>.</p>",
      "rawMarkdown": "The same here. Trying to execute `GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')` results in a very long delay after which I get `HTTPError: HTTP Error 502: Bad Gateway`.",
      "votes": null
    },
    {
      "id": "873172",
      "postDate": "06/03/2020 21:49:19",
      "content": "<p>get_gcs_path('melanoma-256x256') works on cpu and gpu for me.\nget_gcs_path('siim-isic-melanoma-classification') does not...</p>",
      "rawMarkdown": "get_gcs_path('melanoma-256x256') works on cpu and gpu for me.\nget_gcs_path('siim-isic-melanoma-classification') does not...",
      "votes": null
    },
    {
      "id": "873198",
      "postDate": "06/03/2020 22:22:30",
      "content": "<p>I got same error.\nSo I commented out the code and put GCS_PATH directly. Of course it worked.\n```</p>\n\n<h1>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</h1>\n\n<p>GCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```</p>",
      "rawMarkdown": "I got same error.\nSo I commented out the code and put GCS_PATH directly. Of course it worked.\n```\n#GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')\nGCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```",
      "votes": null
    },
    {
      "id": "873200",
      "postDate": "06/03/2020 22:26:28",
      "content": "<p>That's awesome. That solved the problem. How did you find the direct path?</p>",
      "rawMarkdown": "That's awesome. That solved the problem. How did you find the direct path?",
      "votes": null
    },
    {
      "id": "873205",
      "postDate": "06/03/2020 22:40:03",
      "content": "<p>My public kernel (when <code>get_gcs_path</code> worked) had printed out the path!\n<a href=\"https://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training\">https://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training</a></p>",
      "rawMarkdown": "My public kernel (when ```get_gcs_path``` worked) had printed out the path!\nhttps://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training",
      "votes": null
    },
    {
      "id": "880134",
      "postDate": "06/10/2020 03:09:42",
      "content": "<p>Hi, I am experiencing the same error now,  I followed this\n```</p>\n\n<h1>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</h1>\n\n<p>GCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```</p>\n\n<p>But still the files are not loading,\n```\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/tfrecords/train*.tfrec')</p>\n\n<p>```This list is empty still. What can I do now?</p>",
      "rawMarkdown": "Hi, I am experiencing the same error now,  I followed this\n```\n#GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')\nGCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```\n\nBut still the files are not loading,\n```\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/tfrecords/train*.tfrec')\n\n```This list is empty still. What can I do now?",
      "votes": null
    },
    {
      "id": "880138",
      "postDate": "06/10/2020 03:17:32",
      "content": "<p>This file path</p>\n\n<pre><code>gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4\n</code></pre>\n\n<p>was some sort of temporary path because it hasn't worked in a week. There is probably a new recent weird file path like this but i don't know what it is.</p>",
      "rawMarkdown": "This file path\n\n    gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4\n\nwas some sort of temporary path because it hasn't worked in a week. There is probably a new recent weird file path like this but i don't know what it is.",
      "votes": null
    },
    {
      "id": "880142",
      "postDate": "06/10/2020 03:28:46",
      "content": "<p>Yes I found it. I also tried the path on the <a href=\"/yasufuminakama\">@yasufuminakama</a> 's notebook \n<code>\"gs://kds-f941eda182a90288c86bd91ec807ec5f6c58e5c4e0fbf9ad2114f762\"</code> even this too is not working. So what can I do for now <a href=\"/cdeotte\">@cdeotte</a> ? Do this one\n<code>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</code>  works for you?\nIf yes then can you please post the latest path as Nakama did.</p>",
      "rawMarkdown": "Yes I found it. I also tried the path on the @yasufuminakama 's notebook \n`\"gs://kds-f941eda182a90288c86bd91ec807ec5f6c58e5c4e0fbf9ad2114f762\"` even this too is not working. So what can I do for now @cdeotte ? Do this one\n`GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')`  works for you?\nIf yes then can you please post the latest path as Nakama did.",
      "votes": null
    },
    {
      "id": "880145",
      "postDate": "06/10/2020 03:40:34",
      "content": "<p>I'm using this address to access the competition TFRecords data on Colab right now.</p>\n\n<p><code>gs://kds-c89313da1d85616eec461ab327fed61e1335defb486fb7729cf897b1</code></p>",
      "rawMarkdown": "I'm using this address to access the competition TFRecords data on Colab right now.\n\n`gs://kds-c89313da1d85616eec461ab327fed61e1335defb486fb7729cf897b1`",
      "votes": null
    },
    {
      "id": "881273",
      "postDate": "06/10/2020 21:53:49",
      "content": "<p>Hi <a href=\"/cdeotte\">@cdeotte</a> and others,</p>\n\n<p>just jumping in to clarify what's happening when you do a <code>KaggleDatasets().getgcs_path()</code> in your notebook:</p>\n\n<ul>\n<li>TPUs need the data to be in regional buckets that are accessible to them (same region as your TPU is located). To make this transparent for all users, we created the a mechanism to copy the data from the comp to such a bucket and return the path via <code>getgcs_path</code></li>\n<li>Obviously we don't want to do this for every single notebook, so we're reusing these copies. For a whole bunch of reasons, these copies can't live forever, so expire every couple of days. Whoever needs access to the data first after that will trigger a new copy. That's when it gets slow and might get over a client timeout (the data here is ~110GB).</li>\n<li>I spent some time optimizing that process on our end, when we implemented it first it didn't work well with a lot of data (both the large file count as well as the size matters here).</li>\n</ul>\n\n<p>With the improvements I made, you should be fine now to always use <code>KaggleDatasets().getgcs_path()</code>. I don't recommend using a static path, since this will only work for a couple of days and then the data will expire.</p>\n\n<p>Sorry for the hassle. Hope that helps understanding the issue and the behavior on our end.</p>\n\n<p>Cheers,\nTimo</p>",
      "rawMarkdown": "Hi @cdeotte and others,\n\njust jumping in to clarify what's happening when you do a `KaggleDatasets().getgcs_path()` in your notebook:\n\n- TPUs need the data to be in regional buckets that are accessible to them (same region as your TPU is located). To make this transparent for all users, we created the a mechanism to copy the data from the comp to such a bucket and return the path via `getgcs_path`\n- Obviously we don't want to do this for every single notebook, so we're reusing these copies. For a whole bunch of reasons, these copies can't live forever, so expire every couple of days. Whoever needs access to the data first after that will trigger a new copy. That's when it gets slow and might get over a client timeout (the data here is ~110GB).\n- I spent some time optimizing that process on our end, when we implemented it first it didn't work well with a lot of data (both the large file count as well as the size matters here).\n\nWith the improvements I made, you should be fine now to always use `KaggleDatasets().getgcs_path()`. I don't recommend using a static path, since this will only work for a couple of days and then the data will expire.\n\nSorry for the hassle. Hope that helps understanding the issue and the behavior on our end.\n\nCheers,\nTimo",
      "votes": null
    },
    {
      "id": "881290",
      "postDate": "06/10/2020 22:11:21",
      "content": "<p>Thanks Timo. I updated my post to not include suggesting using the temporary link.</p>",
      "rawMarkdown": "Thanks Timo. I updated my post to not include suggesting using the temporary link.",
      "votes": null
    },
    {
      "id": "881424",
      "postDate": "06/11/2020 03:07:50",
      "content": "<p>Thanks Timo for clarifying this.</p>",
      "rawMarkdown": "Thanks Timo for clarifying this.",
      "votes": null
    },
    {
      "id": "881483",
      "postDate": "06/11/2020 04:38:37",
      "content": "<p><a href=\"/timoboz\">@timoboz</a> The syntax should be <code>get_gcs_path()</code> right?\nBecause when i use <code>getgcs_path()</code> im getting no module named error</p>",
      "rawMarkdown": "timoboz The syntax should be `get_gcs_path()` right?\nBecause when i use `getgcs_path()` im getting no module named error",
      "votes": null
    },
    {
      "id": "881498",
      "postDate": "06/11/2020 05:20:36",
      "content": "<p>Yes, It is get_gcs_path(). </p>",
      "rawMarkdown": "Yes, It is get_gcs_path().",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 873109,
      "author_name": "romanweilguny",
      "author_url": "",
      "post_date": "06/03/2020 19:52:00",
      "content": "<p>same - not only tpu - notonly interactive</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 873131,
      "author_name": "chikenfeet",
      "author_url": "",
      "post_date": "06/03/2020 20:27:18",
      "content": "<p>get_gcs_path does not work. I can run on GPUs by directly reading from local dir</p>",
      "votes": null,
      "replies": [
        {
          "id": 873142,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "06/03/2020 20:47:04",
          "content": "<p><code>get_gcs_path</code> works for me if i load my own TFRecords from my Kaggle dataset <a href=\"https://www.kaggle.com/cdeotte/melanoma-512x512\">here</a>, but it won't load the competition TFRecords for me.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 873172,
          "author_name": "romanweilguny",
          "author_url": "",
          "post_date": "06/03/2020 21:49:19",
          "content": "<p>get_gcs_path('melanoma-256x256') works on cpu and gpu for me.\nget_gcs_path('siim-isic-melanoma-classification') does not...</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 873147,
      "author_name": "graf10a",
      "author_url": "",
      "post_date": "06/03/2020 21:01:55",
      "content": "<p>The same here. Trying to execute <code>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</code> results in a very long delay after which I get <code>HTTPError: HTTP Error 502: Bad Gateway</code>.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 873198,
      "author_name": "yasufuminakama",
      "author_url": "",
      "post_date": "06/03/2020 22:22:30",
      "content": "<p>I got same error.\nSo I commented out the code and put GCS_PATH directly. Of course it worked.\n```</p>\n\n<h1>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</h1>\n\n<p>GCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```</p>",
      "votes": null,
      "replies": [
        {
          "id": 873200,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "06/03/2020 22:26:28",
          "content": "<p>That's awesome. That solved the problem. How did you find the direct path?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 873205,
          "author_name": "yasufuminakama",
          "author_url": "",
          "post_date": "06/03/2020 22:40:03",
          "content": "<p>My public kernel (when <code>get_gcs_path</code> worked) had printed out the path!\n<a href=\"https://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training\">https://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 880134,
          "author_name": "aakashveera",
          "author_url": "",
          "post_date": "06/10/2020 03:09:42",
          "content": "<p>Hi, I am experiencing the same error now,  I followed this\n```</p>\n\n<h1>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</h1>\n\n<p>GCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```</p>\n\n<p>But still the files are not loading,\n```\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/tfrecords/train*.tfrec')</p>\n\n<p>```This list is empty still. What can I do now?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 880138,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "06/10/2020 03:17:32",
          "content": "<p>This file path</p>\n\n<pre><code>gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4\n</code></pre>\n\n<p>was some sort of temporary path because it hasn't worked in a week. There is probably a new recent weird file path like this but i don't know what it is.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 880142,
          "author_name": "aakashveera",
          "author_url": "",
          "post_date": "06/10/2020 03:28:46",
          "content": "<p>Yes I found it. I also tried the path on the <a href=\"/yasufuminakama\">@yasufuminakama</a> 's notebook \n<code>\"gs://kds-f941eda182a90288c86bd91ec807ec5f6c58e5c4e0fbf9ad2114f762\"</code> even this too is not working. So what can I do for now <a href=\"/cdeotte\">@cdeotte</a> ? Do this one\n<code>GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')</code>  works for you?\nIf yes then can you please post the latest path as Nakama did.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 880145,
          "author_name": "mutantspore",
          "author_url": "",
          "post_date": "06/10/2020 03:40:34",
          "content": "<p>I'm using this address to access the competition TFRecords data on Colab right now.</p>\n\n<p><code>gs://kds-c89313da1d85616eec461ab327fed61e1335defb486fb7729cf897b1</code></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 881273,
      "author_name": "timoboz",
      "author_url": "",
      "post_date": "06/10/2020 21:53:49",
      "content": "<p>Hi <a href=\"/cdeotte\">@cdeotte</a> and others,</p>\n\n<p>just jumping in to clarify what's happening when you do a <code>KaggleDatasets().getgcs_path()</code> in your notebook:</p>\n\n<ul>\n<li>TPUs need the data to be in regional buckets that are accessible to them (same region as your TPU is located). To make this transparent for all users, we created the a mechanism to copy the data from the comp to such a bucket and return the path via <code>getgcs_path</code></li>\n<li>Obviously we don't want to do this for every single notebook, so we're reusing these copies. For a whole bunch of reasons, these copies can't live forever, so expire every couple of days. Whoever needs access to the data first after that will trigger a new copy. That's when it gets slow and might get over a client timeout (the data here is ~110GB).</li>\n<li>I spent some time optimizing that process on our end, when we implemented it first it didn't work well with a lot of data (both the large file count as well as the size matters here).</li>\n</ul>\n\n<p>With the improvements I made, you should be fine now to always use <code>KaggleDatasets().getgcs_path()</code>. I don't recommend using a static path, since this will only work for a couple of days and then the data will expire.</p>\n\n<p>Sorry for the hassle. Hope that helps understanding the issue and the behavior on our end.</p>\n\n<p>Cheers,\nTimo</p>",
      "votes": null,
      "replies": [
        {
          "id": 881290,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "06/10/2020 22:11:21",
          "content": "<p>Thanks Timo. I updated my post to not include suggesting using the temporary link.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 881424,
          "author_name": "aakashveera",
          "author_url": "",
          "post_date": "06/11/2020 03:07:50",
          "content": "<p>Thanks Timo for clarifying this.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 881483,
          "author_name": "msharuk589",
          "author_url": "",
          "post_date": "06/11/2020 04:38:37",
          "content": "<p><a href=\"/timoboz\">@timoboz</a> The syntax should be <code>get_gcs_path()</code> right?\nBecause when i use <code>getgcs_path()</code> im getting no module named error</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 881498,
          "author_name": "aakashveera",
          "author_url": "",
          "post_date": "06/11/2020 05:20:36",
          "content": "<p>Yes, It is get_gcs_path(). </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "873093": "Is anyone else getting a TPU HTTP Connection error today?  \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F844a208edaa2c5b2263b69f5debec45d%2FScreen%20Shot%202020-06-03%20at%2012.28.32%20PM.png?generation=1591212524318603&amp;alt=media)\n  \nI can load my Kaggle dataset TFRecords fine, but I get the above error message when trying to load the competition's `siim-isic-melanoma-classification` folder. I don't think i'm making a code mistake because to verify, I have forked 3 public notebooks but all 3 public notebooks fail today.\n\nThis error occurs in TPU interactive notebook. (I haven't tried committing).\n\n===================\n\nUPDATE: Kaggle Team has responded below and hopefully fixed this issue. Thank you Timo.",
    "873109": "same - not only tpu - notonly interactive",
    "873131": "get_gcs_path does not work. I can run on GPUs by directly reading from local dir",
    "873142": "`get_gcs_path` works for me if i load my own TFRecords from my Kaggle dataset [here][1], but it won't load the competition TFRecords for me.\n\n[1]: https://www.kaggle.com/cdeotte/melanoma-512x512",
    "873147": "The same here. Trying to execute `GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')` results in a very long delay after which I get `HTTPError: HTTP Error 502: Bad Gateway`.",
    "873172": "get_gcs_path('melanoma-256x256') works on cpu and gpu for me.\nget_gcs_path('siim-isic-melanoma-classification') does not...",
    "873198": "I got same error.\nSo I commented out the code and put GCS_PATH directly. Of course it worked.\n```\n#GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')\nGCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```",
    "873200": "That's awesome. That solved the problem. How did you find the direct path?",
    "873205": "My public kernel (when ```get_gcs_path``` worked) had printed out the path!\nhttps://www.kaggle.com/yasufuminakama/tpu-siim-isic-efficientnetb3-training",
    "880134": "Hi, I am experiencing the same error now,  I followed this\n```\n#GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')\nGCS_PATH = 'gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4'\n```\n\nBut still the files are not loading,\n```\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/tfrecords/train*.tfrec')\n\n```This list is empty still. What can I do now?",
    "880138": "This file path\n\n    gs://kds-0b2c68d2b2fa4692fcffc1029c606b32dd6a88de8d6da08fcd30d0c4\n\nwas some sort of temporary path because it hasn't worked in a week. There is probably a new recent weird file path like this but i don't know what it is.",
    "880142": "Yes I found it. I also tried the path on the @yasufuminakama 's notebook \n`\"gs://kds-f941eda182a90288c86bd91ec807ec5f6c58e5c4e0fbf9ad2114f762\"` even this too is not working. So what can I do for now @cdeotte ? Do this one\n`GCS_PATH = KaggleDatasets().get_gcs_path('siim-isic-melanoma-classification')`  works for you?\nIf yes then can you please post the latest path as Nakama did.",
    "880145": "I'm using this address to access the competition TFRecords data on Colab right now.\n\n`gs://kds-c89313da1d85616eec461ab327fed61e1335defb486fb7729cf897b1`",
    "881273": "Hi @cdeotte and others,\n\njust jumping in to clarify what's happening when you do a `KaggleDatasets().getgcs_path()` in your notebook:\n\n- TPUs need the data to be in regional buckets that are accessible to them (same region as your TPU is located). To make this transparent for all users, we created the a mechanism to copy the data from the comp to such a bucket and return the path via `getgcs_path`\n- Obviously we don't want to do this for every single notebook, so we're reusing these copies. For a whole bunch of reasons, these copies can't live forever, so expire every couple of days. Whoever needs access to the data first after that will trigger a new copy. That's when it gets slow and might get over a client timeout (the data here is ~110GB).\n- I spent some time optimizing that process on our end, when we implemented it first it didn't work well with a lot of data (both the large file count as well as the size matters here).\n\nWith the improvements I made, you should be fine now to always use `KaggleDatasets().getgcs_path()`. I don't recommend using a static path, since this will only work for a couple of days and then the data will expire.\n\nSorry for the hassle. Hope that helps understanding the issue and the behavior on our end.\n\nCheers,\nTimo",
    "881290": "Thanks Timo. I updated my post to not include suggesting using the temporary link.",
    "881424": "Thanks Timo for clarifying this.",
    "881483": "timoboz The syntax should be `get_gcs_path()` right?\nBecause when i use `getgcs_path()` im getting no module named error",
    "881498": "Yes, It is get_gcs_path()."
  },
  "source": "meta"
}