Spent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.
\nIf you're an NN noob like me, KerasTuner is a good way to experiment with a bunch of different architectures and see how they perform on your dataset. You can use these findings to have a backbone for your NN. And then add in the finer details like the right learning schedule, skip connections etc. to improve your model further.
\nI have explored this in the notebook below
\nhttps://www.kaggle.com/code/illidan7/kerastuner-find-the-mlp-for-you/notebook
\nI'm still working on getting a good MLP model too. Happy to learn and share! I look forward to any feedback!
\nCredits:
\nSide note:
\nI tried to put together a test inference section as well. But for some reason, I keep getting a memory error when I try to read in the test file before I can split it into chunks. I have commented out that part of the code
Spent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.
\nIf you're an NN noob like me, KerasTuner is a good way to experiment with a bunch of different architectures and see how they perform on your dataset. You can use these findings to have a backbone for your NN. And then add in the finer details like the right learning schedule, skip connections etc. to improve your model further.
\nI have explored this in the notebook below
\nhttps://www.kaggle.com/code/illidan7/kerastuner-find-the-mlp-for-you/notebook
\nI'm still working on getting a good MLP model too. Happy to learn and share! I look forward to any feedback!
\nCredits:
\nSide note:
\nI tried to put together a test inference section as well. But for some reason, I keep getting a memory error when I try to read in the test file before I can split it into chunks. I have commented out that part of the code
\n\nSpent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.
\n
This hits wayyy too close to home
", "rawMarkdown": "> Spent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.\n\nThis hits wayyy too close to home", "votes": null }, { "id": "1892104", "postDate": "08/09/2022 21:03:09", "content": "Haha I feel you bro
\nIt's been a fun competition, learned a lot 😃 Just gotta keep moving forward. No ragrets
", "rawMarkdown": "Haha I feel you bro\n\nIt's been a fun competition, learned a lot 😃 Just gotta keep moving forward. No ragrets", "votes": null }, { "id": "1893254", "postDate": "08/10/2022 16:52:12", "content": "Thanks need to check this .. is this kind of a Neural arch search ?
", "rawMarkdown": "Thanks need to check this .. is this kind of a Neural arch search ?", "votes": null }, { "id": "1894467", "postDate": "08/11/2022 14:13:20", "content": "Yeah it is. On the notebook, it tries architectures up to 3 layers deep, with up to 1024 neurons in each layer. You can expand the scope of the search by changing the parameters inside the make_MLP function.
\nHope it is useful!
", "rawMarkdown": "Yeah it is. On the notebook, it tries architectures up to 3 layers deep, with up to 1024 neurons in each layer. You can expand the scope of the search by changing the parameters inside the make_MLP function. \n\nHope it is useful!", "votes": null }, { "id": "1896794", "postDate": "08/13/2022 06:47:48", "content": "thanks for youe topic
", "rawMarkdown": "thanks for youe topic", "votes": null }, { "id": "1903465", "postDate": "08/17/2022 13:02:06", "content": "Thanks for your sharing
", "rawMarkdown": "Thanks for your sharing", "votes": null } ], "comments": [ { "id": 1892092, "author_name": "mrandri19", "author_url": "", "post_date": "08/09/2022 20:46:17", "content": "\n\nSpent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.
\n
This hits wayyy too close to home
", "votes": null, "replies": [ { "id": 1892104, "author_name": "illidan7", "author_url": "", "post_date": "08/09/2022 21:03:09", "content": "Haha I feel you bro
\nIt's been a fun competition, learned a lot 😃 Just gotta keep moving forward. No ragrets
", "votes": null, "replies": [] } ] }, { "id": 1893254, "author_name": "gauravbrills", "author_url": "", "post_date": "08/10/2022 16:52:12", "content": "Thanks need to check this .. is this kind of a Neural arch search ?
", "votes": null, "replies": [ { "id": 1894467, "author_name": "illidan7", "author_url": "", "post_date": "08/11/2022 14:13:20", "content": "Yeah it is. On the notebook, it tries architectures up to 3 layers deep, with up to 1024 neurons in each layer. You can expand the scope of the search by changing the parameters inside the make_MLP function.
\nHope it is useful!
", "votes": null, "replies": [] } ] }, { "id": 1896794, "author_name": "", "author_url": "", "post_date": "08/13/2022 06:47:48", "content": "thanks for youe topic
", "votes": null, "replies": [] }, { "id": 1903465, "author_name": "kristinryuvve", "author_url": "", "post_date": "08/17/2022 13:02:06", "content": "Thanks for your sharing
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1891714": "Spent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble. \n\nIf you're an NN noob like me, KerasTuner is a good way to experiment with a bunch of different architectures and see how they perform on your dataset. You can use these findings to have a backbone for your NN. And then add in the finer details like the right learning schedule, skip connections etc. to improve your model further.\n\nI have explored this in the notebook below\n\nhttps://www.kaggle.com/code/illidan7/kerastuner-find-the-mlp-for-you/notebook\n\nI'm still working on getting a good MLP model too. Happy to learn and share! I look forward to any feedback!\n\n\n____________________\nCredits:\n- https://www.kaggle.com/code/fchollet/moa-keras-kerastuner-best-practices\n- https://www.tensorflow.org/tutorials/keras/keras_tuner\n- https://www.kaggle.com/datasets/raddar/amex-data-integer-dtypes-parquet-format\n\n__________________\n\n\n**Side note**: \nI tried to put together a test inference section as well. But for some reason, I keep getting a memory error when I try to read in the test file before I can split it into chunks. I have commented out that part of the code", "1892092": "> Spent all your time on feature engineering and tree models? You are not alone. Just about 2 weeks left and it's time to take a shot at building that perfect NN to round out your ensemble.\n\nThis hits wayyy too close to home", "1892104": "Haha I feel you bro\n\nIt's been a fun competition, learned a lot 😃 Just gotta keep moving forward. No ragrets", "1893254": "Thanks need to check this .. is this kind of a Neural arch search ?", "1894467": "Yeah it is. On the notebook, it tries architectures up to 3 layers deep, with up to 1024 neurons in each layer. You can expand the scope of the search by changing the parameters inside the make_MLP function. \n\nHope it is useful!", "1896794": "thanks for youe topic", "1903465": "Thanks for your sharing" }, "source": "meta" }