I did refer to notebooks. I've created 2 datasets, one contains pretrained weights which I've kept public and other contains my externally trained model and it's private. But I'm getting \"Submission CSV Not Found\" error on submission. When I make the externally trained model public, the submission becomes successful. As in the guidelines it's written external data should be publicly available, so does that mean externally trained model has to be publicly available? Pl shelp me solve the error!
\n\nAs you can see, on committing my kernel does produce a \"submission.csv\" file.\n.png?generation=1582433114881907&alt=media)
I did refer to notebooks. I've created 2 datasets, one contains pretrained weights which I've kept public and other contains my externally trained model and it's private. But I'm getting \"Submission CSV Not Found\" error on submission. When I make the externally trained model public, the submission becomes successful. As in the guidelines it's written external data should be publicly available, so does that mean externally trained model has to be publicly available? Pl shelp me solve the error!
\n\nAs you can see, on committing my kernel does produce a \"submission.csv\" file.\n.png?generation=1582433114881907&alt=media)
so does that mean externally trained model has to be publicly available
No, you can definitely train your model, upload it in your private Kaggle dataset, and use it for inference.
", "rawMarkdown": "``` so does that mean externally trained model has to be publicly available ```\n\nNo, you can definitely train your model, upload it in your private Kaggle dataset, and use it for inference.", "votes": null }, { "id": "754123", "postDate": "02/23/2020 04:42:17", "content": "Then why am I getting this error? And on making it public, it works fine!
", "rawMarkdown": "Then why am I getting this error? And on making it public, it works fine!", "votes": null }, { "id": "754145", "postDate": "02/23/2020 05:36:55", "content": "These are tricky errors, and if you can't solve I would suggest start with a working public inference kernel to test. I remember, I had this error some time ago, I literally started with a public kernel and modified for my purpose. It's not scientific, but saved me from wasting tons of debugging time.
", "rawMarkdown": "These are tricky errors, and if you can't solve I would suggest start with a working public inference kernel to test. I remember, I had this error some time ago, I literally started with a public kernel and modified for my purpose. It's not scientific, but saved me from wasting tons of debugging time.", "votes": null }, { "id": "754261", "postDate": "02/23/2020 09:58:58", "content": "This error comes when your code is unable to process some videos in test set. You can get rid of it by doing exception handling.
", "rawMarkdown": "This error comes when your code is unable to process some videos in test set. You can get rid of it by doing exception handling.", "votes": null }, { "id": "754504", "postDate": "02/23/2020 16:55:03", "content": "As of now, my model doesn't detect faces. So there shouldn't be any exceptions I think. On every video, I am capturing 5 frames only. That should also not throw errors. After that just passing the images to model. What I've noticed is making external data public doesn't throw this error in my case. But one shouldn't make their model public right?
", "rawMarkdown": "As of now, my model doesn't detect faces. So there shouldn't be any exceptions I think. On every video, I am capturing 5 frames only. That should also not throw errors. After that just passing the images to model. What I've noticed is making external data public doesn't throw this error in my case. But one shouldn't make their model public right?", "votes": null }, { "id": "754513", "postDate": "02/23/2020 17:08:21", "content": "Worth trying what @nishagh said, here is why:
\n\nThere are files in the private set that OpenCV fails to read. The sad part, OpenCV does not throw error unless you are performing operations such as BRG2RGB. Therefore, use exceptions even if you are just reading video frames.
\n\nP.S. Again, I am inferencing with models private to me, so that's not an issue for me.
", "rawMarkdown": "Worth trying what @nishagh said, here is why:\n\nThere are files in the private set that OpenCV fails to read. The sad part, OpenCV does not throw error unless you are performing operations such as BRG2RGB. Therefore, use exceptions even if you are just reading video frames.\n\nP.S. Again, I am inferencing with models private to me, so that's not an issue for me.", "votes": null }, { "id": "754527", "postDate": "02/23/2020 17:35:24", "content": "Thank you so much! Will definitely try that out.
\n\nIn case there is an exception, do you guys just give a random probability to that video? And I just don't understand how come changing it from private to public makes the submission possible, probably a kaggle bug!
", "rawMarkdown": "Thank you so much! Will definitely try that out.\n\nIn case there is an exception, do you guys just give a random probability to that video? And I just don't understand how come changing it from private to public makes the submission possible, probably a kaggle bug!", "votes": null }, { "id": "754538", "postDate": "02/23/2020 17:59:51", "content": "Yes, that's pretty weird! That's why I gave my first suggestion to start from scratch.
\n\nRegarding random probability, I use 0.5 to be safe.
\n\nP.S. If you are using Keras, this Kernel used to work with a private model: https://www.kaggle.com/debanga/submission-with-externally-trained-keras-model. Now, I moved to PyTorch, so I cannot help much with this kernel anymore.
", "rawMarkdown": "Yes, that's pretty weird! That's why I gave my first suggestion to start from scratch.\n\nRegarding random probability, I use 0.5 to be safe.\n\nP.S. If you are using Keras, this Kernel used to work with a private model: https://www.kaggle.com/debanga/submission-with-externally-trained-keras-model. Now, I moved to PyTorch, so I cannot help much with this kernel anymore.", "votes": null }, { "id": "758014", "postDate": "02/27/2020 11:08:28", "content": "down load your model to your PC , then create a new dataset by uploading your model from PC .
", "rawMarkdown": "down load your model to your PC , then create a new dataset by uploading your model from PC .", "votes": null } ], "comments": [ { "id": 754104, "author_name": "debanga", "author_url": "", "post_date": "02/23/2020 04:10:24", "content": "so does that mean externally trained model has to be publicly available
No, you can definitely train your model, upload it in your private Kaggle dataset, and use it for inference.
", "votes": null, "replies": [ { "id": 754123, "author_name": "saanikagupta", "author_url": "", "post_date": "02/23/2020 04:42:17", "content": "Then why am I getting this error? And on making it public, it works fine!
", "votes": null, "replies": [] }, { "id": 754145, "author_name": "debanga", "author_url": "", "post_date": "02/23/2020 05:36:55", "content": "These are tricky errors, and if you can't solve I would suggest start with a working public inference kernel to test. I remember, I had this error some time ago, I literally started with a public kernel and modified for my purpose. It's not scientific, but saved me from wasting tons of debugging time.
", "votes": null, "replies": [] }, { "id": 754261, "author_name": "nishagh", "author_url": "", "post_date": "02/23/2020 09:58:58", "content": "This error comes when your code is unable to process some videos in test set. You can get rid of it by doing exception handling.
", "votes": null, "replies": [] }, { "id": 754504, "author_name": "saanikagupta", "author_url": "", "post_date": "02/23/2020 16:55:03", "content": "As of now, my model doesn't detect faces. So there shouldn't be any exceptions I think. On every video, I am capturing 5 frames only. That should also not throw errors. After that just passing the images to model. What I've noticed is making external data public doesn't throw this error in my case. But one shouldn't make their model public right?
", "votes": null, "replies": [] }, { "id": 754513, "author_name": "debanga", "author_url": "", "post_date": "02/23/2020 17:08:21", "content": "Worth trying what @nishagh said, here is why:
\n\nThere are files in the private set that OpenCV fails to read. The sad part, OpenCV does not throw error unless you are performing operations such as BRG2RGB. Therefore, use exceptions even if you are just reading video frames.
\n\nP.S. Again, I am inferencing with models private to me, so that's not an issue for me.
", "votes": null, "replies": [] }, { "id": 754527, "author_name": "saanikagupta", "author_url": "", "post_date": "02/23/2020 17:35:24", "content": "Thank you so much! Will definitely try that out.
\n\nIn case there is an exception, do you guys just give a random probability to that video? And I just don't understand how come changing it from private to public makes the submission possible, probably a kaggle bug!
", "votes": null, "replies": [] }, { "id": 754538, "author_name": "debanga", "author_url": "", "post_date": "02/23/2020 17:59:51", "content": "Yes, that's pretty weird! That's why I gave my first suggestion to start from scratch.
\n\nRegarding random probability, I use 0.5 to be safe.
\n\nP.S. If you are using Keras, this Kernel used to work with a private model: https://www.kaggle.com/debanga/submission-with-externally-trained-keras-model. Now, I moved to PyTorch, so I cannot help much with this kernel anymore.
", "votes": null, "replies": [] } ] }, { "id": 758014, "author_name": "bosbos", "author_url": "", "post_date": "02/27/2020 11:08:28", "content": "down load your model to your PC , then create a new dataset by uploading your model from PC .
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "754098": "I did refer to notebooks. I've created 2 datasets, one contains pretrained weights which I've kept public and other contains my externally trained model and it's private. But I'm getting \"Submission CSV Not Found\" error on submission. When I make the externally trained model public, the submission becomes successful. As in the guidelines it's written external data should be publicly available, so does that mean externally trained model has to be publicly available? Pl shelp me solve the error!\n\nAs you can see, on committing my kernel does produce a \"submission.csv\" file.\n.png?generation=1582433114881907&alt=media)", "754104": "``` so does that mean externally trained model has to be publicly available ```\n\nNo, you can definitely train your model, upload it in your private Kaggle dataset, and use it for inference.", "754123": "Then why am I getting this error? And on making it public, it works fine!", "754145": "These are tricky errors, and if you can't solve I would suggest start with a working public inference kernel to test. I remember, I had this error some time ago, I literally started with a public kernel and modified for my purpose. It's not scientific, but saved me from wasting tons of debugging time.", "754261": "This error comes when your code is unable to process some videos in test set. You can get rid of it by doing exception handling.", "754504": "As of now, my model doesn't detect faces. So there shouldn't be any exceptions I think. On every video, I am capturing 5 frames only. That should also not throw errors. After that just passing the images to model. What I've noticed is making external data public doesn't throw this error in my case. But one shouldn't make their model public right?", "754513": "Worth trying what @nishagh said, here is why:\n\nThere are files in the private set that OpenCV fails to read. The sad part, OpenCV does not throw error unless you are performing operations such as BRG2RGB. Therefore, use exceptions even if you are just reading video frames.\n\nP.S. Again, I am inferencing with models private to me, so that's not an issue for me.", "754527": "Thank you so much! Will definitely try that out.\n\nIn case there is an exception, do you guys just give a random probability to that video? And I just don't understand how come changing it from private to public makes the submission possible, probably a kaggle bug!", "754538": "Yes, that's pretty weird! That's why I gave my first suggestion to start from scratch.\n\nRegarding random probability, I use 0.5 to be safe.\n\nP.S. If you are using Keras, this Kernel used to work with a private model: https://www.kaggle.com/debanga/submission-with-externally-trained-keras-model. Now, I moved to PyTorch, so I cannot help much with this kernel anymore.", "758014": "down load your model to your PC , then create a new dataset by uploading your model from PC ." }, "source": "meta" }