Hi, \nIn our previously published notebook, We've added a new section for modeling the meta-information using TabNet, a novel high-performance and interpretable canonical deep tabular data learning architecture.
We've used an open-source TabNet implementation in PyTorch. To demonstrate the working pipeline of TabNet, just three features have been used (LB:~70). In addition, RandomizedSearchCV is also used to find the optimal hyper-param. And lastly, with the best param, Group-Stratify-KFold is applied for the model training.
Hope you like it. Cheers.
", "messages": [ { "id": "887526", "postDate": "06/15/2020 18:03:09", "content": "Hi, \nIn our previously published notebook, We've added a new section for modeling the meta-information using TabNet, a novel high-performance and interpretable canonical deep tabular data learning architecture.
We've used an open-source TabNet implementation in PyTorch. To demonstrate the working pipeline of TabNet, just three features have been used (LB:~70). In addition, RandomizedSearchCV is also used to find the optimal hyper-param. And lastly, with the best param, Group-Stratify-KFold is applied for the model training.
Hope you like it. Cheers.
", "rawMarkdown": "Hi, \nIn our previously published notebook, We've added a new section for modeling the **meta-information** using [`TabNet`](https://arxiv.org/pdf/1908.07442.pdf), a novel high-performance and interpretable canonical deep tabular data learning architecture. \n\n- [[TF.Keras] Melanoma Classification Starter, TabNet](https://www.kaggle.com/ipythonx/tf-keras-melanoma-classification-starter-tabnet)\n\nWe've used an open-source [`TabNet` implementation in **PyTorch**](https://github.com/dreamquark-ai/tabnet). To demonstrate the working pipeline of `TabNet`, just three features have been used (LB:~70). In addition, **RandomizedSearchCV** is also used to find the optimal hyper-param. And lastly, with the best param, **Group-Stratify-KFold** is applied for the model training.\n\n---\n\nHope you like it. Cheers.", "votes": null }, { "id": "887843", "postDate": "06/15/2020 23:26:27", "content": "Thanks for sharing. I have two questions, first, in your notebook, did you use Tensorflow-Keras for image data and PyTorch (TabNet) for tabular data? and second i see that you use a type of augmentation for image data,the size of the image dataset isn't enough? I'm still a novice so, i don't understimate your skills i'm just learning.
", "rawMarkdown": "Thanks for sharing. I have two questions, first, in your notebook, did you use Tensorflow-Keras for image data and PyTorch (TabNet) for tabular data? and second i see that you use a type of augmentation for image data,the size of the image dataset isn't enough? I'm still a novice so, i don't understimate your skills i'm just learning.", "votes": null }, { "id": "887992", "postDate": "06/16/2020 03:48:57", "content": "Yes, I've used tf.keras for image modeling and PyTorch for tabular. And please feel free to experiment with other data set. I've used only jpeg samples (no external data) for the demonstration purposes.
Thanks a lot for answer, good luck.
", "rawMarkdown": "Thanks a lot for answer, good luck.", "votes": null } ], "comments": [ { "id": 887843, "author_name": "hiramcho", "author_url": "", "post_date": "06/15/2020 23:26:27", "content": "Thanks for sharing. I have two questions, first, in your notebook, did you use Tensorflow-Keras for image data and PyTorch (TabNet) for tabular data? and second i see that you use a type of augmentation for image data,the size of the image dataset isn't enough? I'm still a novice so, i don't understimate your skills i'm just learning.
", "votes": null, "replies": [ { "id": 887992, "author_name": "ipythonx", "author_url": "", "post_date": "06/16/2020 03:48:57", "content": "Yes, I've used tf.keras for image modeling and PyTorch for tabular. And please feel free to experiment with other data set. I've used only jpeg samples (no external data) for the demonstration purposes.
Thanks a lot for answer, good luck.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "887526": "Hi, \nIn our previously published notebook, We've added a new section for modeling the **meta-information** using [`TabNet`](https://arxiv.org/pdf/1908.07442.pdf), a novel high-performance and interpretable canonical deep tabular data learning architecture. \n\n- [[TF.Keras] Melanoma Classification Starter, TabNet](https://www.kaggle.com/ipythonx/tf-keras-melanoma-classification-starter-tabnet)\n\nWe've used an open-source [`TabNet` implementation in **PyTorch**](https://github.com/dreamquark-ai/tabnet). To demonstrate the working pipeline of `TabNet`, just three features have been used (LB:~70). In addition, **RandomizedSearchCV** is also used to find the optimal hyper-param. And lastly, with the best param, **Group-Stratify-KFold** is applied for the model training.\n\n---\n\nHope you like it. Cheers.", "887843": "Thanks for sharing. I have two questions, first, in your notebook, did you use Tensorflow-Keras for image data and PyTorch (TabNet) for tabular data? and second i see that you use a type of augmentation for image data,the size of the image dataset isn't enough? I'm still a novice so, i don't understimate your skills i'm just learning.", "887992": "Yes, I've used `tf.keras` for image modeling and PyTorch for tabular. And please feel free to experiment with other data set. I've used only jpeg samples (no external data) for the demonstration purposes.", "888548": "Thanks a lot for answer, good luck." }, "source": "meta" }