I can see a lot of my fellow Kagglers have uploaded three submissions. As per my understanding, in round two we just need to run the rules we created during round one and find the outcome for the newly released test data set. However, even if we retrain the model using validation outcomes, we get only two models. Shall we not be allowed to submit only two predictions. How can we justify the third submission..?
", "messages": [ { "id": "110943", "postDate": "03/09/2016 20:42:14", "content": "I can see a lot of my fellow Kagglers have uploaded three submissions. As per my understanding, in round two we just need to run the rules we created during round one and find the outcome for the newly released test data set. However, even if we retrain the model using validation outcomes, we get only two models. Shall we not be allowed to submit only two predictions. How can we justify the third submission..?
", "rawMarkdown": "I can see a lot of my fellow Kagglers have uploaded three submissions. As per my understanding, in round two we just need to run the rules we created during round one and find the outcome for the newly released test data set. However, even if we retrain the model using validation outcomes, we get only two models. Shall we not be allowed to submit only two predictions. How can we justify the third submission..?", "votes": null }, { "id": "110945", "postDate": "03/09/2016 20:59:21", "content": "@Mohd, first, because of the way the scores were reset, everyone starts with 1 score of zero, so people with 3 scores, have only submitted 2 new scores. Second, there are a number of ways that one could innocently mess up and need to submit extra times. The simplest is just to submit the same model twice; I've certainly done that before, although not this time.
", "rawMarkdown": "Mohd, first, because of the way the scores were reset, everyone starts with 1 score of zero, so people with 3 scores, have only submitted 2 new scores. Second, there are a number of ways that one could innocently mess up and need to submit extra times. The simplest is just to submit the same model twice; I've certainly done that before, although not this time.", "votes": null }, { "id": "110991", "postDate": "03/10/2016 07:15:48", "content": "I made a third submission (indeed the first one - which is a very simple baseline model) which was a very basic sanity check that I did not do some trivial error. Incorrect handling of a new kind of error in the test data may also lead to more submissions - I'll have to do that I think.. (edit: fortunately not..)
\n\nAnother motivation for more submissions is curiosity how would they be scored on the full test set - even if one is not selecting them.
", "rawMarkdown": "I made a third submission (indeed the first one - which is a very simple baseline model) which was a very basic sanity check that I did not do some trivial error. Incorrect handling of a new kind of error in the test data may also lead to more submissions - I'll have to do that I think.. (edit: fortunately not..)\r\n\r\nAnother motivation for more submissions is curiosity how would they be scored on the full test set - even if one is not selecting them.", "votes": null } ], "comments": [ { "id": 110945, "author_name": "bitsofbits", "author_url": "", "post_date": "03/09/2016 20:59:21", "content": "@Mohd, first, because of the way the scores were reset, everyone starts with 1 score of zero, so people with 3 scores, have only submitted 2 new scores. Second, there are a number of ways that one could innocently mess up and need to submit extra times. The simplest is just to submit the same model twice; I've certainly done that before, although not this time.
", "votes": null, "replies": [] }, { "id": 110991, "author_name": "romualdj", "author_url": "", "post_date": "03/10/2016 07:15:48", "content": "I made a third submission (indeed the first one - which is a very simple baseline model) which was a very basic sanity check that I did not do some trivial error. Incorrect handling of a new kind of error in the test data may also lead to more submissions - I'll have to do that I think.. (edit: fortunately not..)
\n\nAnother motivation for more submissions is curiosity how would they be scored on the full test set - even if one is not selecting them.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "110943": "I can see a lot of my fellow Kagglers have uploaded three submissions. As per my understanding, in round two we just need to run the rules we created during round one and find the outcome for the newly released test data set. However, even if we retrain the model using validation outcomes, we get only two models. Shall we not be allowed to submit only two predictions. How can we justify the third submission..?", "110945": "Mohd, first, because of the way the scores were reset, everyone starts with 1 score of zero, so people with 3 scores, have only submitted 2 new scores. Second, there are a number of ways that one could innocently mess up and need to submit extra times. The simplest is just to submit the same model twice; I've certainly done that before, although not this time.", "110991": "I made a third submission (indeed the first one - which is a very simple baseline model) which was a very basic sanity check that I did not do some trivial error. Incorrect handling of a new kind of error in the test data may also lead to more submissions - I'll have to do that I think.. (edit: fortunately not..)\r\n\r\nAnother motivation for more submissions is curiosity how would they be scored on the full test set - even if one is not selecting them." }, "source": "meta" }