{ "id": 102025, "title": "Overfiting or calculating score of compettion is reason ??", "url": "/competitions/champs-scalar-coupling/discussion/102025", "author_name": "", "post_date": "2019-07-30T12:54:33.745269500Z", "votes": 3, "comment_count": 7, "views": 0, "content": "

When i submit to competition, my score only be 0.045....Does that mean overfiting ??

\n\n

\"\"

", "messages": [ { "id": "588347", "postDate": "07/30/2019 12:54:33", "content": "

When i submit to competition, my score only be 0.045....Does that mean overfiting ??

\n\n

\"\"

", "rawMarkdown": "When i submit to competition, my score only be 0.045....Does that mean overfiting ?? \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2497784%2F1452c3ff673ac4b8c6522dbab53b709f%2F67396250_636346380184363_7287381920063684608_n.png?generation=1564491097631255&alt=media)", "votes": null }, { "id": "588385", "postDate": "07/30/2019 13:55:01", "content": "

I have screwed up in pipeline of preparing data set when combining groups with merge. When I screw up the test ID's or do a random shuffle to the test rows I get results like yours.

", "rawMarkdown": "I have screwed up in pipeline of preparing data set when combining groups with merge. When I screw up the test ID's or do a random shuffle to the test rows I get results like yours.", "votes": null }, { "id": "588749", "postDate": "07/31/2019 01:37:48", "content": "

How do you handle ?

", "rawMarkdown": "How do you handle ?", "votes": null }, { "id": "588815", "postDate": "07/31/2019 04:28:30", "content": "

Not well :) Lots of print statements to figure out were things are going wrong. I use JMP and Excel along with Python. Currently trying to get a merge working - if you have some merges start digging is all I can advise. Been digging for 4 hours so far on my merge issue.

", "rawMarkdown": "Not well :) Lots of print statements to figure out were things are going wrong. I use JMP and Excel along with Python. Currently trying to get a merge working - if you have some merges start digging is all I can advise. Been digging for 4 hours so far on my merge issue.", "votes": null }, { "id": "589428", "postDate": "07/31/2019 21:52:11", "content": "

Are you using predicting on any of the fc/sd/dso/etc? Or any oof?\nA common cause could be using the truth when training, then using predictions when predicting. The model might heavily rely on the fc being 100% accurate for example, but your test ones have a lot of error / are predictions.

", "rawMarkdown": "Are you using predicting on any of the fc/sd/dso/etc? Or any oof?\nA common cause could be using the truth when training, then using predictions when predicting. The model might heavily rely on the fc being 100% accurate for example, but your test ones have a lot of error / are predictions.", "votes": null }, { "id": "589509", "postDate": "08/01/2019 02:26:09", "content": "

Thank you !

", "rawMarkdown": "Thank you !", "votes": null }, { "id": "591532", "postDate": "08/03/2019 21:36:12", "content": "

@duykhanh99 did you found a reason for the low score? I have a very similar problem. High CV score and much lower PL.

", "rawMarkdown": "duykhanh99 did you found a reason for the low score? I have a very similar problem. High CV score and much lower PL.", "votes": null }, { "id": "591552", "postDate": "08/03/2019 22:49:03", "content": "

Fixed my issue. After making merge with Pandas the order of records slightly changed. So, when I made the final X_test it was only partially correctly ordered. Debugged by making 'id' column preserved and checked it was in a correct sequence (sum(df['id'].diff() < 0) == 0)

", "rawMarkdown": "Fixed my issue. After making merge with Pandas the order of records slightly changed. So, when I made the final X_test it was only partially correctly ordered. Debugged by making 'id' column preserved and checked it was in a correct sequence (sum(df['id'].diff() < 0) == 0)", "votes": null } ], "comments": [ { "id": 588385, "author_name": "pcjimmmy", "author_url": "", "post_date": "07/30/2019 13:55:01", "content": "

I have screwed up in pipeline of preparing data set when combining groups with merge. When I screw up the test ID's or do a random shuffle to the test rows I get results like yours.

", "votes": null, "replies": [ { "id": 588749, "author_name": "duykhanh99", "author_url": "", "post_date": "07/31/2019 01:37:48", "content": "

How do you handle ?

", "votes": null, "replies": [] }, { "id": 588815, "author_name": "pcjimmmy", "author_url": "", "post_date": "07/31/2019 04:28:30", "content": "

Not well :) Lots of print statements to figure out were things are going wrong. I use JMP and Excel along with Python. Currently trying to get a merge working - if you have some merges start digging is all I can advise. Been digging for 4 hours so far on my merge issue.

", "votes": null, "replies": [] } ] }, { "id": 589428, "author_name": "roushrsh", "author_url": "", "post_date": "07/31/2019 21:52:11", "content": "

Are you using predicting on any of the fc/sd/dso/etc? Or any oof?\nA common cause could be using the truth when training, then using predictions when predicting. The model might heavily rely on the fc being 100% accurate for example, but your test ones have a lot of error / are predictions.

", "votes": null, "replies": [ { "id": 589509, "author_name": "duykhanh99", "author_url": "", "post_date": "08/01/2019 02:26:09", "content": "

Thank you !

", "votes": null, "replies": [] } ] }, { "id": 591532, "author_name": "criskiev", "author_url": "", "post_date": "08/03/2019 21:36:12", "content": "

@duykhanh99 did you found a reason for the low score? I have a very similar problem. High CV score and much lower PL.

", "votes": null, "replies": [ { "id": 591552, "author_name": "criskiev", "author_url": "", "post_date": "08/03/2019 22:49:03", "content": "

Fixed my issue. After making merge with Pandas the order of records slightly changed. So, when I made the final X_test it was only partially correctly ordered. Debugged by making 'id' column preserved and checked it was in a correct sequence (sum(df['id'].diff() < 0) == 0)

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "588347": "When i submit to competition, my score only be 0.045....Does that mean overfiting ?? \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2497784%2F1452c3ff673ac4b8c6522dbab53b709f%2F67396250_636346380184363_7287381920063684608_n.png?generation=1564491097631255&alt=media)", "588385": "I have screwed up in pipeline of preparing data set when combining groups with merge. When I screw up the test ID's or do a random shuffle to the test rows I get results like yours.", "588749": "How do you handle ?", "588815": "Not well :) Lots of print statements to figure out were things are going wrong. I use JMP and Excel along with Python. Currently trying to get a merge working - if you have some merges start digging is all I can advise. Been digging for 4 hours so far on my merge issue.", "589428": "Are you using predicting on any of the fc/sd/dso/etc? Or any oof?\nA common cause could be using the truth when training, then using predictions when predicting. The model might heavily rely on the fc being 100% accurate for example, but your test ones have a lot of error / are predictions.", "589509": "Thank you !", "591532": "duykhanh99 did you found a reason for the low score? I have a very similar problem. High CV score and much lower PL.", "591552": "Fixed my issue. After making merge with Pandas the order of records slightly changed. So, when I made the final X_test it was only partially correctly ordered. Debugged by making 'id' column preserved and checked it was in a correct sequence (sum(df['id'].diff() < 0) == 0)" }, "source": "meta" }