![\"Screenshot\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2261609%2F2d8e8e7cac73cd0f50b0fc6b3e3add21%2FScreenshot%20from%202020-01-04%2019-52-45.png?generation=1578147818564893&alt=media\")
I'm unable to submit my predictions, getting submission error multiple times. Please check my notebook https://www.kaggle.com/suryaparsa/bengaliai-resnet50
\n\nIs the test set of 12 images only or its expected more than that?
\n\n
I'm unable to submit my predictions, getting submission error multiple times. Please check my notebook https://www.kaggle.com/suryaparsa/bengaliai-resnet50
\n\nIs the test set of 12 images only or its expected more than that?
\n\n
Hi there,
\n\nAs you may have guessed, no. The actual test set used for scoring is of the same size that the training set. The organizer only provided 4 small test parquets file to help us create kernels that could work with any 4 kinds of parquets file.\nThe problem of submission scoring error has been discussed in other post like this one but it is linked to a Python error raised in your kernel. In my own experience, it was a MemoryError when manipulating the 4 test parquet files. While the oens provided by the organizer were not raising any error (containing only 12 images), the actual parquet files contains hundreds of thousands of images, making MemoryError or other performance related errors more likely to be raised.
\n\nTo make sure your code will not raised such error and to debug it, I would advise you to try to run inference on the train parquet files as they are supposed to be very similar in term of size to the test parquet files. If the execution goes well, then put back your code as it was before, using the test parquet files and all should be ok !
\n\nJust note that the inference execution needs to take less than 9 hours to compute
", "rawMarkdown": "Hi there,\n\nAs you may have guessed, no. The actual test set used for scoring is of the same size that the training set. The organizer only provided 4 small test parquets file to help us create kernels that could work with any 4 kinds of parquets file.\nThe problem of submission scoring error has been discussed in other post like [this one](https://www.kaggle.com/c/bengaliai-cv19/discussion/123212) but it is linked to a Python error raised in your kernel. In my own experience, it was a MemoryError when manipulating the 4 test parquet files. While the oens provided by the organizer were not raising any error (containing only 12 images), the actual parquet files contains hundreds of thousands of images, making MemoryError or other performance related errors more likely to be raised.\n\nTo make sure your code will not raised such error and to debug it, I would advise you to try to run inference on the **train parquet files** as they are supposed to be very similar in term of size to the test parquet files. If the execution goes well, then put back your code as it was before, using the **test parquet files** and all should be ok !\n\nJust note that the inference execution needs to take less than 9 hours to compute", "votes": null }, { "id": "710550", "postDate": "01/04/2020 21:53:23", "content": "Thanks Thomas, Its the memory error and number of test images are expected to be much greater not 12.
", "rawMarkdown": "Thanks Thomas, Its the memory error and number of test images are expected to be much greater not 12.", "votes": null } ], "comments": [ { "id": 710324, "author_name": "dimartinot", "author_url": "", "post_date": "01/04/2020 15:39:41", "content": "Hi there,
\n\nAs you may have guessed, no. The actual test set used for scoring is of the same size that the training set. The organizer only provided 4 small test parquets file to help us create kernels that could work with any 4 kinds of parquets file.\nThe problem of submission scoring error has been discussed in other post like this one but it is linked to a Python error raised in your kernel. In my own experience, it was a MemoryError when manipulating the 4 test parquet files. While the oens provided by the organizer were not raising any error (containing only 12 images), the actual parquet files contains hundreds of thousands of images, making MemoryError or other performance related errors more likely to be raised.
\n\nTo make sure your code will not raised such error and to debug it, I would advise you to try to run inference on the train parquet files as they are supposed to be very similar in term of size to the test parquet files. If the execution goes well, then put back your code as it was before, using the test parquet files and all should be ok !
\n\nJust note that the inference execution needs to take less than 9 hours to compute
", "votes": null, "replies": [ { "id": 710550, "author_name": "suryaparsa", "author_url": "", "post_date": "01/04/2020 21:53:23", "content": "Thanks Thomas, Its the memory error and number of test images are expected to be much greater not 12.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "710272": "I'm unable to submit my predictions, getting submission error multiple times. Please check my notebook https://www.kaggle.com/suryaparsa/bengaliai-resnet50\n\nIs the test set of 12 images only or its expected more than that?\n\n", "710324": "Hi there,\n\nAs you may have guessed, no. The actual test set used for scoring is of the same size that the training set. The organizer only provided 4 small test parquets file to help us create kernels that could work with any 4 kinds of parquets file.\nThe problem of submission scoring error has been discussed in other post like [this one](https://www.kaggle.com/c/bengaliai-cv19/discussion/123212) but it is linked to a Python error raised in your kernel. In my own experience, it was a MemoryError when manipulating the 4 test parquet files. While the oens provided by the organizer were not raising any error (containing only 12 images), the actual parquet files contains hundreds of thousands of images, making MemoryError or other performance related errors more likely to be raised.\n\nTo make sure your code will not raised such error and to debug it, I would advise you to try to run inference on the **train parquet files** as they are supposed to be very similar in term of size to the test parquet files. If the execution goes well, then put back your code as it was before, using the **test parquet files** and all should be ok !\n\nJust note that the inference execution needs to take less than 9 hours to compute", "710550": "Thanks Thomas, Its the memory error and number of test images are expected to be much greater not 12." }, "source": "meta" }