Seems like all of the submissions so far are variants of the default submission (all zeros) or random. Has anyone (other than @ryches) succeeded in submitting anything non-trivial so far?
\n\nIn my case, out of 6 submissions (all GPUs), none were successful. I made sure it was committed correctly, and I even made sure to only run my GPU model on a single test example (everything else was default submission), I still got an error telling me \"Notebook Exceeded Allowed Compute\" although it clearly shouldn't take more than minimal RAM usage (and run in less than 100s).
\n\n@kaggleteam let me know if you have any idea.
\n\n@everyone I'd be happy to share my kernel if you'd like to take a look, if you have an idea how to solve it; i was planning to publicize it once I got LB results but seems like it will take a while.
", "messages": [ { "id": "696066", "postDate": "12/16/2019 04:40:53", "content": "Seems like all of the submissions so far are variants of the default submission (all zeros) or random. Has anyone (other than @ryches) succeeded in submitting anything non-trivial so far?
\n\nIn my case, out of 6 submissions (all GPUs), none were successful. I made sure it was committed correctly, and I even made sure to only run my GPU model on a single test example (everything else was default submission), I still got an error telling me \"Notebook Exceeded Allowed Compute\" although it clearly shouldn't take more than minimal RAM usage (and run in less than 100s).
\n\n@kaggleteam let me know if you have any idea.
\n\n@everyone I'd be happy to share my kernel if you'd like to take a look, if you have an idea how to solve it; i was planning to publicize it once I got LB results but seems like it will take a while.
", "rawMarkdown": "Seems like all of the submissions so far are variants of the default submission (all zeros) or random. Has anyone (other than @ryches) succeeded in submitting anything non-trivial so far?\n\nIn my case, out of 6 submissions (all GPUs), none were successful. I made sure it was committed correctly, and I even made sure to only run my GPU model on a single test example (everything else was default submission), I still got an error telling me \"Notebook Exceeded Allowed Compute\" although it clearly shouldn't take more than minimal RAM usage (and run in less than 100s).\n\n@kaggleteam let me know if you have any idea. \n\n@everyone I'd be happy to share my kernel if you'd like to take a look, if you have an idea how to solve it; i was planning to publicize it once I got LB results but seems like it will take a while.", "votes": null }, { "id": "696155", "postDate": "12/16/2019 07:31:32", "content": "Lol . Go Ahead @xhlulu . I am sure , you are way better than me in this thing , but sometimes a different perspective works . However , does Everyone means Everyone ? or is it the User Everyone ?
", "rawMarkdown": "Lol . Go Ahead @xhlulu . I am sure , you are way better than me in this thing , but sometimes a different perspective works . However , does Everyone means Everyone ? or is it the User Everyone ?", "votes": null }, { "id": "696238", "postDate": "12/16/2019 10:26:56", "content": "I made a submission training only on the train sample (400 videos) on a simple CNN just to test and it worked, although the results were worse than a random submission :)
\n\nEDIT: Just to put it out there that I didn't even do face detection, I trained on the whole image resized to 128x128
", "rawMarkdown": "I made a submission training only on the train sample (400 videos) on a simple CNN just to test and it worked, although the results were worse than a random submission :)\n\nEDIT: Just to put it out there that I didn't even do face detection, I trained on the whole image resized to 128x128", "votes": null }, { "id": "696242", "postDate": "12/16/2019 10:34:45", "content": "I don't think it should matter but my inference has been cpu based so far. Only because I'm bad at pytorch. Might be worth trying cpu inference and see if that works and gpu is the issue.
", "rawMarkdown": "I don't think it should matter but my inference has been cpu based so far. Only because I'm bad at pytorch. Might be worth trying cpu inference and see if that works and gpu is the issue.", "votes": null }, { "id": "696311", "postDate": "12/16/2019 12:59:08", "content": "Checkout the lb ;)
", "rawMarkdown": "Checkout the lb ;)", "votes": null }, { "id": "696612", "postDate": "12/16/2019 20:52:53", "content": "Did you use a GPU or a CPU kernel? Also would love to take a look at your kernel if you were planning to make it public :)
", "rawMarkdown": "Did you use a GPU or a CPU kernel? Also would love to take a look at your kernel if you were planning to make it public :)", "votes": null }, { "id": "696613", "postDate": "12/16/2019 20:53:24", "content": "I really hope it's not because of GPU issues... But i'll try CPU 👍
", "rawMarkdown": "I really hope it's not because of GPU issues... But i'll try CPU 👍", "votes": null }, { "id": "696665", "postDate": "12/16/2019 23:03:05", "content": "I made a kernel here. Just a simple CNN detecting pre-cropped face. But I can't submit it, it always says \"Submission File not Found\". So recently, I am trying to submit it in other kernel.\nEdit: I managed to successfully submit it but the LB was pretty low. Maybe I should increase the image size.
", "rawMarkdown": "I made a kernel [here](https://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model). Just a simple CNN detecting pre-cropped face. But I can't submit it, it always says \"Submission File not Found\". So recently, I am trying to submit it in other kernel.\nEdit: I managed to successfully submit it but the LB was pretty low. Maybe I should increase the image size.", "votes": null }, { "id": "696950", "postDate": "12/17/2019 09:28:30", "content": "I used a GPU kernel. I'm not planning to make it public at this moment as I have trained offline and just loaded the weights for the network and did the predictions on Kaggle, I might turn it public in the near future and include the training part as is quite simple. But right now I'm focusing on preparing a robust face detection for my pipeline, maybe when I'm done with that.
", "rawMarkdown": "I used a GPU kernel. I'm not planning to make it public at this moment as I have trained offline and just loaded the weights for the network and did the predictions on Kaggle, I might turn it public in the near future and include the training part as is quite simple. But right now I'm focusing on preparing a robust face detection for my pipeline, maybe when I'm done with that.", "votes": null }, { "id": "697265", "postDate": "12/17/2019 17:03:32", "content": "I'm really confused. I changed my kernel from GPU to cpu and it suddenly works, so I thought GPU kernels were broken. But I guess it isn't?
", "rawMarkdown": "I'm really confused. I changed my kernel from GPU to cpu and it suddenly works, so I thought GPU kernels were broken. But I guess it isn't?", "votes": null }, { "id": "697423", "postDate": "12/17/2019 22:17:11", "content": "Now I am having issues with \"Submission CSV Not Found\" as well...
", "rawMarkdown": "Now I am having issues with \"Submission CSV Not Found\" as well...", "votes": null }, { "id": "697435", "postDate": "12/17/2019 22:56:38", "content": "Me too. My first two submissions with a GPU kernel worked without problems, now I'm also getting \"Submission CSV Not Found\"
", "rawMarkdown": "Me too. My first two submissions with a GPU kernel worked without problems, now I'm also getting \"Submission CSV Not Found\"", "votes": null }, { "id": "697704", "postDate": "12/18/2019 09:42:04", "content": "looks like 0.5 in submission is doing the trick till now.
", "rawMarkdown": "looks like 0.5 in submission is doing the trick till now.", "votes": null }, { "id": "698413", "postDate": "12/19/2019 07:37:03", "content": "Should the variable \"x\" in your definition of function \"read_image\" be replaced with variable \"name\"?🤔
", "rawMarkdown": "Should the variable \"x\" in your definition of function \"read_image\" be replaced with variable \"name\"?🤔", "votes": null } ], "comments": [ { "id": 696155, "author_name": "phoenix9032", "author_url": "", "post_date": "12/16/2019 07:31:32", "content": "Lol . Go Ahead @xhlulu . I am sure , you are way better than me in this thing , but sometimes a different perspective works . However , does Everyone means Everyone ? or is it the User Everyone ?
", "votes": null, "replies": [ { "id": 696311, "author_name": "xhlulu", "author_url": "", "post_date": "12/16/2019 12:59:08", "content": "Checkout the lb ;)
", "votes": null, "replies": [] } ] }, { "id": 696238, "author_name": "pedromb", "author_url": "", "post_date": "12/16/2019 10:26:56", "content": "I made a submission training only on the train sample (400 videos) on a simple CNN just to test and it worked, although the results were worse than a random submission :)
\n\nEDIT: Just to put it out there that I didn't even do face detection, I trained on the whole image resized to 128x128
", "votes": null, "replies": [ { "id": 696612, "author_name": "xhlulu", "author_url": "", "post_date": "12/16/2019 20:52:53", "content": "Did you use a GPU or a CPU kernel? Also would love to take a look at your kernel if you were planning to make it public :)
", "votes": null, "replies": [] }, { "id": 696950, "author_name": "pedromb", "author_url": "", "post_date": "12/17/2019 09:28:30", "content": "I used a GPU kernel. I'm not planning to make it public at this moment as I have trained offline and just loaded the weights for the network and did the predictions on Kaggle, I might turn it public in the near future and include the training part as is quite simple. But right now I'm focusing on preparing a robust face detection for my pipeline, maybe when I'm done with that.
", "votes": null, "replies": [] }, { "id": 697265, "author_name": "xhlulu", "author_url": "", "post_date": "12/17/2019 17:03:32", "content": "I'm really confused. I changed my kernel from GPU to cpu and it suddenly works, so I thought GPU kernels were broken. But I guess it isn't?
", "votes": null, "replies": [] } ] }, { "id": 696242, "author_name": "ryches", "author_url": "", "post_date": "12/16/2019 10:34:45", "content": "I don't think it should matter but my inference has been cpu based so far. Only because I'm bad at pytorch. Might be worth trying cpu inference and see if that works and gpu is the issue.
", "votes": null, "replies": [ { "id": 696613, "author_name": "xhlulu", "author_url": "", "post_date": "12/16/2019 20:53:24", "content": "I really hope it's not because of GPU issues... But i'll try CPU 👍
", "votes": null, "replies": [] }, { "id": 697423, "author_name": "ryches", "author_url": "", "post_date": "12/17/2019 22:17:11", "content": "Now I am having issues with \"Submission CSV Not Found\" as well...
", "votes": null, "replies": [] }, { "id": 697435, "author_name": "pedromb", "author_url": "", "post_date": "12/17/2019 22:56:38", "content": "Me too. My first two submissions with a GPU kernel worked without problems, now I'm also getting \"Submission CSV Not Found\"
", "votes": null, "replies": [] } ] }, { "id": 696665, "author_name": "unkownhihi", "author_url": "", "post_date": "12/16/2019 23:03:05", "content": "I made a kernel here. Just a simple CNN detecting pre-cropped face. But I can't submit it, it always says \"Submission File not Found\". So recently, I am trying to submit it in other kernel.\nEdit: I managed to successfully submit it but the LB was pretty low. Maybe I should increase the image size.
", "votes": null, "replies": [] }, { "id": 697704, "author_name": "hemanth007", "author_url": "", "post_date": "12/18/2019 09:42:04", "content": "looks like 0.5 in submission is doing the trick till now.
", "votes": null, "replies": [] }, { "id": 698413, "author_name": "haozshno1", "author_url": "", "post_date": "12/19/2019 07:37:03", "content": "Should the variable \"x\" in your definition of function \"read_image\" be replaced with variable \"name\"?🤔
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "696066": "Seems like all of the submissions so far are variants of the default submission (all zeros) or random. Has anyone (other than @ryches) succeeded in submitting anything non-trivial so far?\n\nIn my case, out of 6 submissions (all GPUs), none were successful. I made sure it was committed correctly, and I even made sure to only run my GPU model on a single test example (everything else was default submission), I still got an error telling me \"Notebook Exceeded Allowed Compute\" although it clearly shouldn't take more than minimal RAM usage (and run in less than 100s).\n\n@kaggleteam let me know if you have any idea. \n\n@everyone I'd be happy to share my kernel if you'd like to take a look, if you have an idea how to solve it; i was planning to publicize it once I got LB results but seems like it will take a while.", "696155": "Lol . Go Ahead @xhlulu . I am sure , you are way better than me in this thing , but sometimes a different perspective works . However , does Everyone means Everyone ? or is it the User Everyone ?", "696238": "I made a submission training only on the train sample (400 videos) on a simple CNN just to test and it worked, although the results were worse than a random submission :)\n\nEDIT: Just to put it out there that I didn't even do face detection, I trained on the whole image resized to 128x128", "696242": "I don't think it should matter but my inference has been cpu based so far. Only because I'm bad at pytorch. Might be worth trying cpu inference and see if that works and gpu is the issue.", "696311": "Checkout the lb ;)", "696612": "Did you use a GPU or a CPU kernel? Also would love to take a look at your kernel if you were planning to make it public :)", "696613": "I really hope it's not because of GPU issues... But i'll try CPU 👍", "696665": "I made a kernel [here](https://www.kaggle.com/unkownhihi/starter-kernel-with-cnn-model). Just a simple CNN detecting pre-cropped face. But I can't submit it, it always says \"Submission File not Found\". So recently, I am trying to submit it in other kernel.\nEdit: I managed to successfully submit it but the LB was pretty low. Maybe I should increase the image size.", "696950": "I used a GPU kernel. I'm not planning to make it public at this moment as I have trained offline and just loaded the weights for the network and did the predictions on Kaggle, I might turn it public in the near future and include the training part as is quite simple. But right now I'm focusing on preparing a robust face detection for my pipeline, maybe when I'm done with that.", "697265": "I'm really confused. I changed my kernel from GPU to cpu and it suddenly works, so I thought GPU kernels were broken. But I guess it isn't?", "697423": "Now I am having issues with \"Submission CSV Not Found\" as well...", "697435": "Me too. My first two submissions with a GPU kernel worked without problems, now I'm also getting \"Submission CSV Not Found\"", "697704": "looks like 0.5 in submission is doing the trick till now.", "698413": "Should the variable \"x\" in your definition of function \"read_image\" be replaced with variable \"name\"?🤔" }, "source": "meta" }