I've read the other posts on resolving submission scoring error and nothing seems to be working. I've made my code to be flexible and work with any amount of files in the directory, I'm checking for frames when I open videos with opencv, I'm ensuring MTCNN is returning faces(and returning .5 confidence otherwise) and everything works correctly on the public validation(400 video) set. I'm confused as to what else could cause the run to fail. I'll share my code if necessary. Does anyone have any ideas?
\n\nThanks
", "messages": [ { "id": "728764", "postDate": "01/25/2020 08:18:43", "content": "I've read the other posts on resolving submission scoring error and nothing seems to be working. I've made my code to be flexible and work with any amount of files in the directory, I'm checking for frames when I open videos with opencv, I'm ensuring MTCNN is returning faces(and returning .5 confidence otherwise) and everything works correctly on the public validation(400 video) set. I'm confused as to what else could cause the run to fail. I'll share my code if necessary. Does anyone have any ideas?
\n\nThanks
", "rawMarkdown": "I've read the other posts on resolving submission scoring error and nothing seems to be working. I've made my code to be flexible and work with any amount of files in the directory, I'm checking for frames when I open videos with opencv, I'm ensuring MTCNN is returning faces(and returning .5 confidence otherwise) and everything works correctly on the public validation(400 video) set. I'm confused as to what else could cause the run to fail. I'll share my code if necessary. Does anyone have any ideas?\n\nThanks", "votes": null }, { "id": "728825", "postDate": "01/25/2020 10:43:48", "content": "I've just shared my inference kernel. Maybe you can pick up some tips from there?
", "rawMarkdown": "I've just shared [my inference kernel](https://www.kaggle.com/humananalog/inference-demo). Maybe you can pick up some tips from there?", "votes": null }, { "id": "728984", "postDate": "01/25/2020 14:38:05", "content": "I too have just started to get this issue. Kinda annoying when you wait 9 hours just to get a flaw that is illogical.
\n\nI originally had int as my output, but I thought this was not what was requested.
\n\nI may go back just to see if I get any score at all.
", "rawMarkdown": "I too have just started to get this issue. Kinda annoying when you wait 9 hours just to get a flaw that is illogical.\n\nI originally had int as my output, but I thought this was not what was requested. \n\nI may go back just to see if I get any score at all.", "votes": null }, { "id": "729317", "postDate": "01/26/2020 02:27:03", "content": "Solved by putting a bunch of try blocks around each part that could fail. Now I've got a massive difference between CV and LB score to figure out
", "rawMarkdown": "Solved by putting a bunch of try blocks around each part that could fail. Now I've got a massive difference between CV and LB score to figure out", "votes": null }, { "id": "730919", "postDate": "01/28/2020 05:52:33", "content": "Be sure that in the kernel during the public validation that you have at least one video file that is corrupt. I made a bad video by renaming a small csv file as an .mp4. Include this bad file as part of your data set or turn one of the existing files.
\n\nThan add that file to the list of 400 - run the script within the kernel. When the bad file is reached you will likely get an error message that you have not trapped and/or the count of values in your submission will not be the 400 or so that you expect and/or the value for the bad file will be outside the 0 to 1 range.
\n\nThis was how I finally tracked down my errors (had two that were very stupid logic errors in my traps) and after a month got a successful submission on my own code.
", "rawMarkdown": "Be sure that in the kernel during the public validation that you have at least one video file that is corrupt. I made a bad video by renaming a small csv file as an .mp4. Include this bad file as part of your data set or turn one of the existing files.\n\nThan add that file to the list of 400 - run the script within the kernel. When the bad file is reached you will likely get an error message that you have not trapped and/or the count of values in your submission will not be the 400 or so that you expect and/or the value for the bad file will be outside the 0 to 1 range.\n\nThis was how I finally tracked down my errors (had two that were very stupid logic errors in my traps) and after a month got a successful submission on my own code.", "votes": null }, { "id": "732046", "postDate": "01/29/2020 12:20:37", "content": "I was struggling with this Submission Scoring Error for a long time - finally the problem was that I didn't include all the file names in the submission. Apparently I filtered some files (probably corrupted or without faces?) while generating the torch DataSet.
", "rawMarkdown": "I was struggling with this Submission Scoring Error for a long time - finally the problem was that I didn't include all the file names in the submission. Apparently I filtered some files (probably corrupted or without faces?) while generating the torch DataSet.", "votes": null } ], "comments": [ { "id": 728825, "author_name": "humananalog", "author_url": "", "post_date": "01/25/2020 10:43:48", "content": "I've just shared my inference kernel. Maybe you can pick up some tips from there?
", "votes": null, "replies": [] }, { "id": 728984, "author_name": "quiteconfused", "author_url": "", "post_date": "01/25/2020 14:38:05", "content": "I too have just started to get this issue. Kinda annoying when you wait 9 hours just to get a flaw that is illogical.
\n\nI originally had int as my output, but I thought this was not what was requested.
\n\nI may go back just to see if I get any score at all.
", "votes": null, "replies": [] }, { "id": 729317, "author_name": "sahiljain5125", "author_url": "", "post_date": "01/26/2020 02:27:03", "content": "Solved by putting a bunch of try blocks around each part that could fail. Now I've got a massive difference between CV and LB score to figure out
", "votes": null, "replies": [] }, { "id": 730919, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/28/2020 05:52:33", "content": "Be sure that in the kernel during the public validation that you have at least one video file that is corrupt. I made a bad video by renaming a small csv file as an .mp4. Include this bad file as part of your data set or turn one of the existing files.
\n\nThan add that file to the list of 400 - run the script within the kernel. When the bad file is reached you will likely get an error message that you have not trapped and/or the count of values in your submission will not be the 400 or so that you expect and/or the value for the bad file will be outside the 0 to 1 range.
\n\nThis was how I finally tracked down my errors (had two that were very stupid logic errors in my traps) and after a month got a successful submission on my own code.
", "votes": null, "replies": [] }, { "id": 732046, "author_name": "odedgolden", "author_url": "", "post_date": "01/29/2020 12:20:37", "content": "I was struggling with this Submission Scoring Error for a long time - finally the problem was that I didn't include all the file names in the submission. Apparently I filtered some files (probably corrupted or without faces?) while generating the torch DataSet.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "728764": "I've read the other posts on resolving submission scoring error and nothing seems to be working. I've made my code to be flexible and work with any amount of files in the directory, I'm checking for frames when I open videos with opencv, I'm ensuring MTCNN is returning faces(and returning .5 confidence otherwise) and everything works correctly on the public validation(400 video) set. I'm confused as to what else could cause the run to fail. I'll share my code if necessary. Does anyone have any ideas?\n\nThanks", "728825": "I've just shared [my inference kernel](https://www.kaggle.com/humananalog/inference-demo). Maybe you can pick up some tips from there?", "728984": "I too have just started to get this issue. Kinda annoying when you wait 9 hours just to get a flaw that is illogical.\n\nI originally had int as my output, but I thought this was not what was requested. \n\nI may go back just to see if I get any score at all.", "729317": "Solved by putting a bunch of try blocks around each part that could fail. Now I've got a massive difference between CV and LB score to figure out", "730919": "Be sure that in the kernel during the public validation that you have at least one video file that is corrupt. I made a bad video by renaming a small csv file as an .mp4. Include this bad file as part of your data set or turn one of the existing files.\n\nThan add that file to the list of 400 - run the script within the kernel. When the bad file is reached you will likely get an error message that you have not trapped and/or the count of values in your submission will not be the 400 or so that you expect and/or the value for the bad file will be outside the 0 to 1 range.\n\nThis was how I finally tracked down my errors (had two that were very stupid logic errors in my traps) and after a month got a successful submission on my own code.", "732046": "I was struggling with this Submission Scoring Error for a long time - finally the problem was that I didn't include all the file names in the submission. Apparently I filtered some files (probably corrupted or without faces?) while generating the torch DataSet." }, "source": "meta" }