When you are submitting your submission dataframe (submission.csv), leave the NaN's as is! Do not fill them with an empty string (\"\")! I made that mistake and wasted 5-6 submissions.
When you are submitting your submission dataframe (submission.csv), leave the NaN's as is! Do not fill them with an empty string (\"\")! I made that mistake and wasted 5-6 submissions.
Strange but empty string worked fine for me before. However, while submitting the latest version of my kernel (https://www.kaggle.com/opanichev/tf2-0-qa-binary-classification-baseline?scriptVersionId=23035897) I've started to get \"Submission Scoring Error\" with no further explanation. Maybe organizers have changed a scoring script. Submission file looks same as before except several predictions now have new values. Together with a double code run (commit and submit) Kaggle's kernel competition again makes me totally frustrated.
", "rawMarkdown": "Strange but empty string worked fine for me before. However, while submitting the latest version of my kernel (https://www.kaggle.com/opanichev/tf2-0-qa-binary-classification-baseline?scriptVersionId=23035897) I've started to get \"Submission Scoring Error\" with no further explanation. Maybe organizers have changed a scoring script. Submission file looks same as before except several predictions now have new values. Together with a double code run (commit and submit) Kaggle's kernel competition again makes me totally frustrated.", "votes": null }, { "id": "666042", "postDate": "11/05/2019 16:57:09", "content": "I've tried to resubmit my results with NaNs instead of empty strings but got the same scoring error. I think empty string should work fine.
", "rawMarkdown": "I've tried to resubmit my results with NaNs instead of empty strings but got the same scoring error. I think empty string should work fine.", "votes": null }, { "id": "666164", "postDate": "11/05/2019 20:08:26", "content": "That's really weird. I tried submitting the default file (all NaN's) and it gave a result (0.00). But when I submit something with \"\"s (and actual ranges as well), I get submission error. I'd like to get to the bottom of this problem, but I already used up all my submission for today, so I can only find out tomorrow.
", "rawMarkdown": "That's really weird. I tried submitting the default file (all NaN's) and it gave a result (0.00). But when I submit something with \"\"s (and actual ranges as well), I get submission error. I'd like to get to the bottom of this problem, but I already used up all my submission for today, so I can only find out tomorrow.", "votes": null }, { "id": "666713", "postDate": "11/06/2019 12:07:04", "content": "It seems I've figured out what my problem was related to. I have prepared a dataset with extracted features for train and test sets and used it to train the model. When I was committing the code everything was fine, but on the submission stage I've got a scoring error. Now I moved the code for feature extraction for test part back to the code and it work again. It seems that during submission we have a new larger json file with test data and your code should handle this.
", "rawMarkdown": "It seems I've figured out what my problem was related to. I have prepared a dataset with extracted features for train and test sets and used it to train the model. When I was committing the code everything was fine, but on the submission stage I've got a scoring error. Now I moved the code for feature extraction for test part back to the code and it work again. It seems that during submission we have a new larger json file with test data and your code should handle this.", "votes": null } ], "comments": [ { "id": 665985, "author_name": "opanichev", "author_url": "", "post_date": "11/05/2019 15:57:58", "content": "Strange but empty string worked fine for me before. However, while submitting the latest version of my kernel (https://www.kaggle.com/opanichev/tf2-0-qa-binary-classification-baseline?scriptVersionId=23035897) I've started to get \"Submission Scoring Error\" with no further explanation. Maybe organizers have changed a scoring script. Submission file looks same as before except several predictions now have new values. Together with a double code run (commit and submit) Kaggle's kernel competition again makes me totally frustrated.
", "votes": null, "replies": [ { "id": 666164, "author_name": "xhlulu", "author_url": "", "post_date": "11/05/2019 20:08:26", "content": "That's really weird. I tried submitting the default file (all NaN's) and it gave a result (0.00). But when I submit something with \"\"s (and actual ranges as well), I get submission error. I'd like to get to the bottom of this problem, but I already used up all my submission for today, so I can only find out tomorrow.
", "votes": null, "replies": [] } ] }, { "id": 666042, "author_name": "opanichev", "author_url": "", "post_date": "11/05/2019 16:57:09", "content": "I've tried to resubmit my results with NaNs instead of empty strings but got the same scoring error. I think empty string should work fine.
", "votes": null, "replies": [] }, { "id": 666713, "author_name": "opanichev", "author_url": "", "post_date": "11/06/2019 12:07:04", "content": "It seems I've figured out what my problem was related to. I have prepared a dataset with extracted features for train and test sets and used it to train the model. When I was committing the code everything was fine, but on the submission stage I've got a scoring error. Now I moved the code for feature extraction for test part back to the code and it work again. It seems that during submission we have a new larger json file with test data and your code should handle this.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "665960": "When you are submitting your submission dataframe (`submission.csv`), leave the NaN's as is! Do not fill them with an empty string (\"\")! I made that mistake and wasted 5-6 submissions.", "665985": "Strange but empty string worked fine for me before. However, while submitting the latest version of my kernel (https://www.kaggle.com/opanichev/tf2-0-qa-binary-classification-baseline?scriptVersionId=23035897) I've started to get \"Submission Scoring Error\" with no further explanation. Maybe organizers have changed a scoring script. Submission file looks same as before except several predictions now have new values. Together with a double code run (commit and submit) Kaggle's kernel competition again makes me totally frustrated.", "666042": "I've tried to resubmit my results with NaNs instead of empty strings but got the same scoring error. I think empty string should work fine.", "666164": "That's really weird. I tried submitting the default file (all NaN's) and it gave a result (0.00). But when I submit something with \"\"s (and actual ranges as well), I get submission error. I'd like to get to the bottom of this problem, but I already used up all my submission for today, so I can only find out tomorrow.", "666713": "It seems I've figured out what my problem was related to. I have prepared a dataset with extracted features for train and test sets and used it to train the model. When I was committing the code everything was fine, but on the submission stage I've got a scoring error. Now I moved the code for feature extraction for test part back to the code and it work again. It seems that during submission we have a new larger json file with test data and your code should handle this." }, "source": "meta" }