{ "id": 334391, "title": "Best Single NN Score", "url": "/competitions/amex-default-prediction/discussion/334391", "author_name": "", "post_date": "2022-07-01T11:00:37.474975Z", "votes": 25, "comment_count": 8, "views": 0, "content": "

What is the highest score you can achieve by using a single NN?

\n

Till now its for me the following NN:

\n

UPDATE

\n
Model: Self developed dense NN with 8 hidden layers \nCV: 0.785\n
\n

I have seen others achieve quite good scores with TabNet and some RNN models. I'm trying to develop my own model to get >0.79 but I was not able to do so yet.

", "messages": [ { "id": "1839365", "postDate": "07/01/2022 11:00:37", "content": "

What is the highest score you can achieve by using a single NN?

\n

Till now its for me the following NN:

\n

UPDATE

\n
Model: Self developed dense NN with 8 hidden layers \nCV: 0.785\n
\n

I have seen others achieve quite good scores with TabNet and some RNN models. I'm trying to develop my own model to get >0.79 but I was not able to do so yet.

", "rawMarkdown": "What is the highest score you can achieve by using a single NN?\n\nTill now its for me the following NN:\n\n**UPDATE**\n\n```\nModel: Self developed dense NN with 8 hidden layers \nCV: 0.785\n```\n\nI have seen others achieve quite good scores with TabNet and some RNN models. I'm trying to develop my own model to get >0.79 but I was not able to do so yet.", "votes": null }, { "id": "1839398", "postDate": "07/01/2022 11:24:10", "content": "

my current best NN is CV:0.793 / LB:0.794
\nSome people posted 0.795 on LB possible

", "rawMarkdown": "my current best NN is CV:0.793 / LB:0.794\nSome people posted 0.795 on LB possible", "votes": null }, { "id": "1839532", "postDate": "07/01/2022 13:35:59", "content": "

my current best NN is CV:0.791 / LB:0.791

", "rawMarkdown": "my current best NN is CV:0.791 / LB:0.791", "votes": null }, { "id": "1839687", "postDate": "07/01/2022 16:18:03", "content": "

My current best NN with 5 dense layers has cv and lb 0.790 and is published here. Maybe 11 layers is too much.

", "rawMarkdown": "My current best NN with 5 dense layers has cv and lb 0.790 and is [published here](https://www.kaggle.com/code/ambrosm/amex-keras-quickstart-1-training). Maybe 11 layers is too much.", "votes": null }, { "id": "1840677", "postDate": "07/02/2022 13:33:09", "content": "

LB 794
\nTransformers

", "rawMarkdown": "LB 794\nTransformers", "votes": null }, { "id": "1840735", "postDate": "07/02/2022 14:19:47", "content": "

That's big numbers! nice!

", "rawMarkdown": "That's big numbers! nice!", "votes": null }, { "id": "1841251", "postDate": "07/03/2022 00:01:25", "content": "

Transformer
\nCV : 0.792
\nLB : 0.793

", "rawMarkdown": "Transformer\nCV : 0.792\nLB : 0.793", "votes": null }, { "id": "1843047", "postDate": "07/04/2022 14:03:42", "content": "

the feature for the NN model is the same as LGB/XGB ?

", "rawMarkdown": "the feature for the NN model is the same as LGB/XGB ?", "votes": null }, { "id": "1882689", "postDate": "08/03/2022 11:31:30", "content": "

My first NN is 0.787

", "rawMarkdown": "My first NN is 0.787", "votes": null } ], "comments": [ { "id": 1839398, "author_name": "raddar", "author_url": "", "post_date": "07/01/2022 11:24:10", "content": "

my current best NN is CV:0.793 / LB:0.794
\nSome people posted 0.795 on LB possible

", "votes": null, "replies": [ { "id": 1843047, "author_name": "transwarp", "author_url": "", "post_date": "07/04/2022 14:03:42", "content": "

the feature for the NN model is the same as LGB/XGB ?

", "votes": null, "replies": [] } ] }, { "id": 1839532, "author_name": "sunyuri", "author_url": "", "post_date": "07/01/2022 13:35:59", "content": "

my current best NN is CV:0.791 / LB:0.791

", "votes": null, "replies": [] }, { "id": 1839687, "author_name": "ambrosm", "author_url": "", "post_date": "07/01/2022 16:18:03", "content": "

My current best NN with 5 dense layers has cv and lb 0.790 and is published here. Maybe 11 layers is too much.

", "votes": null, "replies": [] }, { "id": 1840677, "author_name": "xiaowangiiiii", "author_url": "", "post_date": "07/02/2022 13:33:09", "content": "

LB 794
\nTransformers

", "votes": null, "replies": [ { "id": 1840735, "author_name": "raddar", "author_url": "", "post_date": "07/02/2022 14:19:47", "content": "

That's big numbers! nice!

", "votes": null, "replies": [] } ] }, { "id": 1841251, "author_name": "zakopur0", "author_url": "", "post_date": "07/03/2022 00:01:25", "content": "

Transformer
\nCV : 0.792
\nLB : 0.793

", "votes": null, "replies": [] }, { "id": 1882689, "author_name": "bjjiang", "author_url": "", "post_date": "08/03/2022 11:31:30", "content": "

My first NN is 0.787

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1839365": "What is the highest score you can achieve by using a single NN?\n\nTill now its for me the following NN:\n\n**UPDATE**\n\n```\nModel: Self developed dense NN with 8 hidden layers \nCV: 0.785\n```\n\nI have seen others achieve quite good scores with TabNet and some RNN models. I'm trying to develop my own model to get >0.79 but I was not able to do so yet.", "1839398": "my current best NN is CV:0.793 / LB:0.794\nSome people posted 0.795 on LB possible", "1839532": "my current best NN is CV:0.791 / LB:0.791", "1839687": "My current best NN with 5 dense layers has cv and lb 0.790 and is [published here](https://www.kaggle.com/code/ambrosm/amex-keras-quickstart-1-training). Maybe 11 layers is too much.", "1840677": "LB 794\nTransformers", "1840735": "That's big numbers! nice!", "1841251": "Transformer\nCV : 0.792\nLB : 0.793", "1843047": "the feature for the NN model is the same as LGB/XGB ?", "1882689": "My first NN is 0.787" }, "source": "meta" }