{ "id": 359126, "title": "Use DNN models?", "url": "/competitions/tabular-playground-series-oct-2022/discussion/359126", "author_name": "", "post_date": "2022-10-10T20:27:00.139724900Z", "votes": 2, "comment_count": 5, "views": 0, "content": "

Anyone attempt use DNN models?

", "messages": [ { "id": "1981516", "postDate": "10/10/2022 20:27:00", "content": "

Anyone attempt use DNN models?

", "rawMarkdown": "Anyone attempt use DNN models?", "votes": null }, { "id": "1981729", "postDate": "10/11/2022 03:00:28", "content": "

@arasekaito You could give it a try. They're good but not the best for tabular data. Most people use XGBoost (SotA for tabular data), tree-based methods, and ensembles because they use less computational resources and perform better than DNN models.

", "rawMarkdown": "arasekaito You could give it a try. They're good but not the best for tabular data. Most people use XGBoost (SotA for tabular data), tree-based methods, and ensembles because they use less computational resources and perform better than DNN models.", "votes": null }, { "id": "1982154", "postDate": "10/11/2022 09:44:10", "content": "

@arasekaito yes i use neural network which is build with the Fastai V2 library. This library is based on pytorch and gives beginners a quite simple and easy access to use neural network for regression tasks. The main challenge for this competition is the sheer number of training data. These require some handling tricks.

\n

You can find my first attempt here: https://www.kaggle.com/code/casati8/kaggle-tps-2022-oct-fastai
\n@paddykb offers another notebook, which uses the older version v1 of Fastai.

\n

You can use these notebooks as a starting point for your own experiments.
\nRegards Casati

", "rawMarkdown": "arasekaito yes i use neural network which is build with the Fastai V2 library. This library is based on pytorch and gives beginners a quite simple and easy access to use neural network for regression tasks. The main challenge for this competition is the sheer number of training data. These require some handling tricks.\n\nYou can find my first attempt here: https://www.kaggle.com/code/casati8/kaggle-tps-2022-oct-fastai\n@paddykb offers another notebook, which uses the older version v1 of Fastai.\n\nYou can use these notebooks as a starting point for your own experiments.\nRegards Casati", "votes": null }, { "id": "1982182", "postDate": "10/11/2022 10:02:12", "content": "

@casati8 FastAI is a really good lib! Did you take the fastai course? (https://course.fast.ai)

", "rawMarkdown": "casati8 FastAI is a really good lib! Did you take the fastai course? (https://course.fast.ai)", "votes": null }, { "id": "1982234", "postDate": "10/11/2022 10:40:36", "content": "

Last year i have a lot of free time and read the book from Jeremy Howard 'Deep Learning for coders with Fastai & pytorch'. I refreshed my long-ago experience from my studies. The book is pedagogically really execlent. Later saw some videos Jeremy Howard posted on youtube. This year I'm participating in these competitions, for which I almost exclusively use the FastAi library. And because the spread of this library is not very high here, I post and publish some of my notebooks. If I have time and leisure.

", "rawMarkdown": "Last year i have a lot of free time and read the book from Jeremy Howard 'Deep Learning for coders with Fastai & pytorch'. I refreshed my long-ago experience from my studies. The book is pedagogically really execlent. Later saw some videos Jeremy Howard posted on youtube. This year I'm participating in these competitions, for which I almost exclusively use the FastAi library. And because the spread of this library is not very high here, I post and publish some of my notebooks. If I have time and leisure.", "votes": null }, { "id": "1982704", "postDate": "10/11/2022 15:07:53", "content": "

Thank you both for your replies. I will try it myself.

", "rawMarkdown": "Thank you both for your replies. I will try it myself.", "votes": null } ], "comments": [ { "id": 1981729, "author_name": "mrgabrielblins", "author_url": "", "post_date": "10/11/2022 03:00:28", "content": "

@arasekaito You could give it a try. They're good but not the best for tabular data. Most people use XGBoost (SotA for tabular data), tree-based methods, and ensembles because they use less computational resources and perform better than DNN models.

", "votes": null, "replies": [] }, { "id": 1982154, "author_name": "casati8", "author_url": "", "post_date": "10/11/2022 09:44:10", "content": "

@arasekaito yes i use neural network which is build with the Fastai V2 library. This library is based on pytorch and gives beginners a quite simple and easy access to use neural network for regression tasks. The main challenge for this competition is the sheer number of training data. These require some handling tricks.

\n

You can find my first attempt here: https://www.kaggle.com/code/casati8/kaggle-tps-2022-oct-fastai
\n@paddykb offers another notebook, which uses the older version v1 of Fastai.

\n

You can use these notebooks as a starting point for your own experiments.
\nRegards Casati

", "votes": null, "replies": [ { "id": 1982182, "author_name": "mrgabrielblins", "author_url": "", "post_date": "10/11/2022 10:02:12", "content": "

@casati8 FastAI is a really good lib! Did you take the fastai course? (https://course.fast.ai)

", "votes": null, "replies": [] }, { "id": 1982234, "author_name": "casati8", "author_url": "", "post_date": "10/11/2022 10:40:36", "content": "

Last year i have a lot of free time and read the book from Jeremy Howard 'Deep Learning for coders with Fastai & pytorch'. I refreshed my long-ago experience from my studies. The book is pedagogically really execlent. Later saw some videos Jeremy Howard posted on youtube. This year I'm participating in these competitions, for which I almost exclusively use the FastAi library. And because the spread of this library is not very high here, I post and publish some of my notebooks. If I have time and leisure.

", "votes": null, "replies": [] }, { "id": 1982704, "author_name": "arasekaito", "author_url": "", "post_date": "10/11/2022 15:07:53", "content": "

Thank you both for your replies. I will try it myself.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1981516": "Anyone attempt use DNN models?", "1981729": "arasekaito You could give it a try. They're good but not the best for tabular data. Most people use XGBoost (SotA for tabular data), tree-based methods, and ensembles because they use less computational resources and perform better than DNN models.", "1982154": "arasekaito yes i use neural network which is build with the Fastai V2 library. This library is based on pytorch and gives beginners a quite simple and easy access to use neural network for regression tasks. The main challenge for this competition is the sheer number of training data. These require some handling tricks.\n\nYou can find my first attempt here: https://www.kaggle.com/code/casati8/kaggle-tps-2022-oct-fastai\n@paddykb offers another notebook, which uses the older version v1 of Fastai.\n\nYou can use these notebooks as a starting point for your own experiments.\nRegards Casati", "1982182": "casati8 FastAI is a really good lib! Did you take the fastai course? (https://course.fast.ai)", "1982234": "Last year i have a lot of free time and read the book from Jeremy Howard 'Deep Learning for coders with Fastai & pytorch'. I refreshed my long-ago experience from my studies. The book is pedagogically really execlent. Later saw some videos Jeremy Howard posted on youtube. This year I'm participating in these competitions, for which I almost exclusively use the FastAi library. And because the spread of this library is not very high here, I post and publish some of my notebooks. If I have time and leisure.", "1982704": "Thank you both for your replies. I will try it myself." }, "source": "meta" }