I'm a bit confused because I'm trying to combine the predictions of several models but when submitting I get the error \"submission csv not found\". However, each of the models I use had no problem when I submitted them individually and my notebook has no bugs. I really don't know how to solve this problem
", "messages": [ { "id": "1127390", "postDate": "12/26/2020 13:57:33", "content": "I'm a bit confused because I'm trying to combine the predictions of several models but when submitting I get the error \"submission csv not found\". However, each of the models I use had no problem when I submitted them individually and my notebook has no bugs. I really don't know how to solve this problem
", "rawMarkdown": "I'm a bit confused because I'm trying to combine the predictions of several models but when submitting I get the error \"submission csv not found\". However, each of the models I use had no problem when I submitted them individually and my notebook has no bugs. I really don't know how to solve this problem", "votes": null }, { "id": "1127400", "postDate": "12/26/2020 14:08:35", "content": "well combining predictions of multiple models shouldn't actually be an issue until your notebook at the end creates one submission.csv following the pattern as mentioned. Kindly check that whether your notebook is creating one with the defined format or not.
", "rawMarkdown": "well combining predictions of multiple models shouldn't actually be an issue until your notebook at the end creates one submission.csv following the pattern as mentioned. Kindly check that whether your notebook is creating one with the defined format or not.", "votes": null }, { "id": "1127905", "postDate": "12/27/2020 02:32:01", "content": "effectively my notebook creates a submission.csv file at the end. That is why I really do not understand why when submitting I got this error.
", "rawMarkdown": "effectively my notebook creates a submission.csv file at the end. That is why I really do not understand why when submitting I got this error.", "votes": null }, { "id": "1127933", "postDate": "12/27/2020 03:54:57", "content": "Root causing \"submission not found\" can be a real pain. Sadly most of the competitions in last year have used this form of prevention of cheating and test data set probing. For folks who don't spend all their spare hours on Kaggle getting this error worked out can be a real frustration.
\nIMO the hosts and Kaggle did a huge dis-service for us by only including one image in the public test. That single image is really not sufficient to insure that you notebook has \"no bugs\".
\nMy advice - run your notebook inference using the train data set - this will let you see if your prediction code can handle multiple images. Best to do this with the \"run all\" and than step through first. If it looks good, than see if it survives a submit/save. If it's still ok - than change the code back to looking at the test folder. Hopefully by this point you found the source of your error.
\nOtherwise:
\nYou will need to read lots of discussion posts in this and some past competitions to gather a list of possible root causes. If your not worried about your kernel being a super winner, than making it public and coming back to discussion with a link might get a few helpful folks to look at your code. The root cause of submission not found is too wide a list for anyone to generate the right fix without seeing the code.
I finally understood what was going wrong with my notebook. For my first attempt of assembling, I simply select the majority vote item between all my models. Now as there was 5 models and 5 possible classes, it was possible that the majority item was not unique which raises an exception when you use the \"mode\" method from \"statistics\" package. In my notebook after committing, I got the submission.csv file because all the 5 models predict class 4 and therefore there was not conflict using the \"mode\" function. So I suppose that in the real test datasets, it is possible that after my 5 models give their predictions there are more than one mode which raises an error. So, to address this problem there are two solutions. First I can build a custom function to select the majority item (what I did and It worked), or build a 6th model (what I've just done).
", "rawMarkdown": "I finally understood what was going wrong with my notebook. For my first attempt of assembling, I simply select the majority vote item between all my models. Now as there was 5 models and 5 possible classes, it was possible that the majority item was not unique which raises an exception when you use the \"mode\" method from \"statistics\" package. In my notebook after committing, I got the submission.csv file because all the 5 models predict class 4 and therefore there was not conflict using the \"mode\" function. So I suppose that in the real test datasets, it is possible that after my 5 models give their predictions there are more than one mode which raises an error. So, to address this problem there are two solutions. First I can build a custom function to select the majority item (what I did and It worked), or build a 6th model (what I've just done).", "votes": null }, { "id": "1132829", "postDate": "12/30/2020 17:54:27", "content": "Glad you figured something out - I recently added vote to my main kernels but only running them on local PC - you might have saved me some grief by letting us know your root cause.
\nGood luck!
", "rawMarkdown": "Glad you figured something out - I recently added vote to my main kernels but only running them on local PC - you might have saved me some grief by letting us know your root cause.\n\nGood luck!", "votes": null } ], "comments": [ { "id": 1127400, "author_name": "haashaatif", "author_url": "", "post_date": "12/26/2020 14:08:35", "content": "well combining predictions of multiple models shouldn't actually be an issue until your notebook at the end creates one submission.csv following the pattern as mentioned. Kindly check that whether your notebook is creating one with the defined format or not.
", "votes": null, "replies": [ { "id": 1127905, "author_name": "danielarmel", "author_url": "", "post_date": "12/27/2020 02:32:01", "content": "effectively my notebook creates a submission.csv file at the end. That is why I really do not understand why when submitting I got this error.
", "votes": null, "replies": [] } ] }, { "id": 1127933, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/27/2020 03:54:57", "content": "Root causing \"submission not found\" can be a real pain. Sadly most of the competitions in last year have used this form of prevention of cheating and test data set probing. For folks who don't spend all their spare hours on Kaggle getting this error worked out can be a real frustration.
\nIMO the hosts and Kaggle did a huge dis-service for us by only including one image in the public test. That single image is really not sufficient to insure that you notebook has \"no bugs\".
\nMy advice - run your notebook inference using the train data set - this will let you see if your prediction code can handle multiple images. Best to do this with the \"run all\" and than step through first. If it looks good, than see if it survives a submit/save. If it's still ok - than change the code back to looking at the test folder. Hopefully by this point you found the source of your error.
\nOtherwise:
\nYou will need to read lots of discussion posts in this and some past competitions to gather a list of possible root causes. If your not worried about your kernel being a super winner, than making it public and coming back to discussion with a link might get a few helpful folks to look at your code. The root cause of submission not found is too wide a list for anyone to generate the right fix without seeing the code.
I finally understood what was going wrong with my notebook. For my first attempt of assembling, I simply select the majority vote item between all my models. Now as there was 5 models and 5 possible classes, it was possible that the majority item was not unique which raises an exception when you use the \"mode\" method from \"statistics\" package. In my notebook after committing, I got the submission.csv file because all the 5 models predict class 4 and therefore there was not conflict using the \"mode\" function. So I suppose that in the real test datasets, it is possible that after my 5 models give their predictions there are more than one mode which raises an error. So, to address this problem there are two solutions. First I can build a custom function to select the majority item (what I did and It worked), or build a 6th model (what I've just done).
", "votes": null, "replies": [] }, { "id": 1132829, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/30/2020 17:54:27", "content": "Glad you figured something out - I recently added vote to my main kernels but only running them on local PC - you might have saved me some grief by letting us know your root cause.
\nGood luck!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1127390": "I'm a bit confused because I'm trying to combine the predictions of several models but when submitting I get the error \"submission csv not found\". However, each of the models I use had no problem when I submitted them individually and my notebook has no bugs. I really don't know how to solve this problem", "1127400": "well combining predictions of multiple models shouldn't actually be an issue until your notebook at the end creates one submission.csv following the pattern as mentioned. Kindly check that whether your notebook is creating one with the defined format or not.", "1127905": "effectively my notebook creates a submission.csv file at the end. That is why I really do not understand why when submitting I got this error.", "1127933": "Root causing \"submission not found\" can be a real pain. Sadly most of the competitions in last year have used this form of prevention of cheating and test data set probing. For folks who don't spend all their spare hours on Kaggle getting this error worked out can be a real frustration. \n\nIMO the hosts and Kaggle did a huge dis-service for us by only including one image in the public test. That single image is really not sufficient to insure that you notebook has \"no bugs\".\n\nMy advice - run your notebook inference using the train data set - this will let you see if your prediction code can handle multiple images. Best to do this with the \"run all\" and than step through first. If it looks good, than see if it survives a submit/save. If it's still ok - than change the code back to looking at the test folder. Hopefully by this point you found the source of your error.\n\nOtherwise:\nYou will need to read lots of discussion posts in this and some past competitions to gather a list of possible root causes. If your not worried about your kernel being a super winner, than making it public and coming back to discussion with a link might get a few helpful folks to look at your code. The root cause of submission not found is too wide a list for anyone to generate the right fix without seeing the code.", "1132340": "I finally understood what was going wrong with my notebook. For my first attempt of assembling, I simply select the majority vote item between all my models. Now as there was 5 models and 5 possible classes, it was possible that the majority item was not unique which raises an exception when you use the \"mode\" method from \"statistics\" package. In my notebook after committing, I got the submission.csv file because all the 5 models predict class 4 and therefore there was not conflict using the \"mode\" function. So I suppose that in the real test datasets, it is possible that after my 5 models give their predictions there are more than one mode which raises an error. So, to address this problem there are two solutions. First I can build a custom function to select the majority item (what I did and It worked), or build a 6th model (what I've just done).", "1132829": "Glad you figured something out - I recently added vote to my main kernels but only running them on local PC - you might have saved me some grief by letting us know your root cause.\n\nGood luck!" }, "source": "meta" }