\n
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4921379%2F3855eeb1e01c79be6c98c0055fd9aaca%2Fd.PNG?generation=1707234454704674&alt=media\")
\n
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4921379%2F3665ca0f47da287b969cfc731cb6dbe5%2F111.PNG?generation=1707234477205115&alt=media\")
Most of the depth=0 numerical variable's distplot seems either left skewed or mostly single bar.
\n
\n
but last one tries to look like gaussian distribution.
\nAny reference or idea what these dist plots show to us?
\n-pics from my notebook-
Most of the depth=0 numerical variable's distplot seems either left skewed or mostly single bar.
\n
\n
but last one tries to look like gaussian distribution.
\nAny reference or idea what these dist plots show to us?
\n-pics from my notebook-
Hi @beckpro ! I think you have to deal with the outliers in this situation. First create boxplots and then use Isolation Forest.
", "rawMarkdown": "Hi @beckpro ! I think you have to deal with the outliers in this situation. First create boxplots and then use Isolation Forest.", "votes": null }, { "id": "2639121", "postDate": "02/06/2024 17:15:25", "content": "@beckpro, I remembered 2 other options that might help with these distributions. This is the Box-Cox method https://www.statology.org/box-cox-transformation-python/ and as an option you can also try binning, i.e. dividing the feature into several categories. For this you can use the optbinning library https://gnpalencia.org/optbinning/tutorials/tutorial_binary.html . I mainly use optbinning in my work when I build a model based on logistic regression.
", "rawMarkdown": "beckpro, I remembered 2 other options that might help with these distributions. This is the Box-Cox method https://www.statology.org/box-cox-transformation-python/ and as an option you can also try binning, i.e. dividing the feature into several categories. For this you can use the optbinning library https://gnpalencia.org/optbinning/tutorials/tutorial_binary.html . I mainly use optbinning in my work when I build a model based on logistic regression.", "votes": null } ], "comments": [ { "id": 2639056, "author_name": "bratkovskyevgeny", "author_url": "", "post_date": "02/06/2024 16:52:24", "content": "Hi @beckpro ! I think you have to deal with the outliers in this situation. First create boxplots and then use Isolation Forest.
", "votes": null, "replies": [] }, { "id": 2639121, "author_name": "bratkovskyevgeny", "author_url": "", "post_date": "02/06/2024 17:15:25", "content": "@beckpro, I remembered 2 other options that might help with these distributions. This is the Box-Cox method https://www.statology.org/box-cox-transformation-python/ and as an option you can also try binning, i.e. dividing the feature into several categories. For this you can use the optbinning library https://gnpalencia.org/optbinning/tutorials/tutorial_binary.html . I mainly use optbinning in my work when I build a model based on logistic regression.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2638983": "Most of the depth=0 numerical variable's distplot seems either left skewed or mostly single bar.\n\n\n\nbut last one tries to look like gaussian distribution.\nAny reference or idea what these dist plots show to us?\n-pics from my [notebook](https://www.kaggle.com/code/beckpro/depth-0-complete-eda-with-nice-descriptions)-", "2639056": "Hi @beckpro ! I think you have to deal with the outliers in this situation. First create boxplots and then use Isolation Forest.", "2639121": "beckpro, I remembered 2 other options that might help with these distributions. This is the Box-Cox method https://www.statology.org/box-cox-transformation-python/ and as an option you can also try binning, i.e. dividing the feature into several categories. For this you can use the optbinning library https://gnpalencia.org/optbinning/tutorials/tutorial_binary.html . I mainly use optbinning in my work when I build a model based on logistic regression." }, "source": "meta" }