{ "id": 246759, "title": "Here comes the 0.01 gap! One digit to change them all.", "url": "/competitions/seti-breakthrough-listen/discussion/246759", "author_name": "Riccardo", "post_date": "2021-06-16T20:14:23.506000", "votes": 2, "comment_count": 3, "views": 0, "content": "

So far classical CNN approaches have led us to 0.98+ in PB. Are you ready to find the breakthrough for the 0.999+?
\nThis shows how a simple digit can whisper a lot! And the time invested in EDA should be greater than the time required by the models to training (at least some times 😂😂).

", "messages": [ { "id": 1353066, "postDate": "2021-06-16T20:14:23.507Z", "content": "

So far classical CNN approaches have led us to 0.98+ in PB. Are you ready to find the breakthrough for the 0.999+?
\nThis shows how a simple digit can whisper a lot! And the time invested in EDA should be greater than the time required by the models to training (at least some times 😂😂).

", "rawMarkdown": "So far classical CNN approaches have led us to 0.98+ in PB. Are you ready to find the breakthrough for the 0.999+? \nThis shows how a simple digit can whisper a lot! And the time invested in EDA should be greater than the time required by the models to training (at least some times 😂😂).", "votes": 2 }, { "id": 1353087, "postDate": "2021-06-16T21:17:45.777Z", "content": "

that fight for 1 digit public lb increase will lead to a lot of kaggler to overfit the public lb so badly and that too for 20% dataset(in csv submission competition like this large ensembe lead to public lb overfitting so fast) and they will remain very happy to see them up in the lb until private lb is revealed.
\nlet us trust our CV

", "rawMarkdown": "that fight for 1 digit public lb increase will lead to a lot of kaggler to overfit the public lb so badly and that too for 20% dataset(in csv submission competition like this large ensembe lead to public lb overfitting so fast) and they will remain very happy to see them up in the lb until private lb is revealed.\n**let us trust our CV**", "votes": 1, "replies": [ { "id": 1353096, "postDate": "2021-06-16T21:37:24.677Z", "content": "

That's right in a lot of comps! Trusting your CV is important.
\nBut in this comp I've seen 0.99+ with very low number of subs. This does not look pretty lb overfit to me.
\nI think these results were obtained through careful analysis of the data, and thus why this comp shoud be a Kernel comp as mentioned in this discussion suggested by @serigne and many others.

", "rawMarkdown": "That's right in a lot of comps! Trusting your CV is important.\nBut in this comp I've seen 0.99+ with very low number of subs. This does not look pretty lb overfit to me.\nI think these results were obtained through careful analysis of the data, and thus why this comp shoud be a Kernel comp as mentioned [in this discussion](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/238758) suggested by @serigne and many others.", "votes": 3 }, { "id": 1353130, "postDate": "2021-06-16T22:00:23.127Z", "content": "

If you can achieve 0.999 on public LB you will be also able to achieve that on private LB.

", "rawMarkdown": "If you can achieve 0.999 on public LB you will be also able to achieve that on private LB.", "votes": 6 } ] } ], "comments": [ { "id": 1353087, "author_name": "Mobassir", "author_url": "", "post_date": "2021-06-16T21:17:45.777000", "content": "

that fight for 1 digit public lb increase will lead to a lot of kaggler to overfit the public lb so badly and that too for 20% dataset(in csv submission competition like this large ensembe lead to public lb overfitting so fast) and they will remain very happy to see them up in the lb until private lb is revealed.
\nlet us trust our CV

", "votes": 1, "replies": [ { "id": 1353096, "author_name": "Riccardo", "author_url": "", "post_date": "2021-06-16T21:37:24.677000", "content": "

That's right in a lot of comps! Trusting your CV is important.
\nBut in this comp I've seen 0.99+ with very low number of subs. This does not look pretty lb overfit to me.
\nI think these results were obtained through careful analysis of the data, and thus why this comp shoud be a Kernel comp as mentioned in this discussion suggested by @serigne and many others.

", "votes": 3, "replies": [] }, { "id": 1353130, "author_name": "Psi", "author_url": "", "post_date": "2021-06-16T22:00:23.127000", "content": "

If you can achieve 0.999 on public LB you will be also able to achieve that on private LB.

", "votes": 6, "replies": [] } ] } ], "raw_markdown_by_id": { "1353066": "So far classical CNN approaches have led us to 0.98+ in PB. Are you ready to find the breakthrough for the 0.999+? \nThis shows how a simple digit can whisper a lot! And the time invested in EDA should be greater than the time required by the models to training (at least some times 😂😂).", "1353087": "that fight for 1 digit public lb increase will lead to a lot of kaggler to overfit the public lb so badly and that too for 20% dataset(in csv submission competition like this large ensembe lead to public lb overfitting so fast) and they will remain very happy to see them up in the lb until private lb is revealed.\n**let us trust our CV**" } }