![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1583716%2F93c8aeef41ac1205ac48acfe7c8dbef5%2FCaptura%20de%20pantalla%202024-05-21%20224049.png?generation=1716324083867018&alt=media\")
As shown in this notebook, there are features with extremely different distributions when comparing the train and test sets. For example, features state_q0001_[0-12] are completely different, and the same happens for features state_q0002_[0-29] and state_q0003_[0-15]. Additionally, some pbuf_ozone features have pretty different distributions.
\nThis is an example taken from the notebook, for features of the state_q0001 family:
\n
How are you managing these features in your models? Are you excluding them completely? Are the records with the missing values hidden in the original low-res dataset? Are these features scaled improperly in the test set, and is this why they are so different?
", "messages": [ { "id": 2828054, "postDate": "2024-05-21T20:41:42.657Z", "content": "As shown in this notebook, there are features with extremely different distributions when comparing the train and test sets. For example, features state_q0001_[0-12] are completely different, and the same happens for features state_q0002_[0-29] and state_q0003_[0-15]. Additionally, some pbuf_ozone features have pretty different distributions.
\nThis is an example taken from the notebook, for features of the state_q0001 family:
\n
How are you managing these features in your models? Are you excluding them completely? Are the records with the missing values hidden in the original low-res dataset? Are these features scaled improperly in the test set, and is this why they are so different?
", "rawMarkdown": "As shown in [this notebook](https://www.kaggle.com/code/docxian/leap-climsim-visual-eda#Targets-and-Features), there are features with extremely different distributions when comparing the train and test sets. For example, features state_q0001_[0-12] are completely different, and the same happens for features state_q0002_[0-29] and state_q0003_[0-15]. Additionally, some pbuf_ozone features have pretty different distributions.\n\nThis is an example taken from the notebook, for features of the state_q0001 family:\n\n\n\nHow are you managing these features in your models? Are you excluding them completely? Are the records with the missing values hidden in the original low-res dataset? Are these features scaled improperly in the test set, and is this why they are so different?", "votes": 11 }, { "id": 2837074, "postDate": "2024-05-26T09:02:05.720Z", "content": "How to solve this problem?
", "rawMarkdown": "How to solve this problem?", "votes": 2 }, { "id": 2839880, "postDate": "2024-05-27T18:42:52.767Z", "content": "Actually the tendency from 0-12 are not considered except temperature. I would not consider the state variables from the pressure levels 0-12 except temperature. I'll try this in the next days
", "rawMarkdown": "Actually the tendency from 0-12 are not considered except temperature. I would not consider the state variables from the pressure levels 0-12 except temperature. I'll try this in the next days " } ], "comments": [ { "id": 2837074, "author_name": "Zhuoqun Li", "author_url": "", "post_date": "2024-05-26T09:02:05.720000", "content": "How to solve this problem?
", "votes": 2, "replies": [] }, { "id": 2839880, "author_name": "HappyKiter", "author_url": "", "post_date": "2024-05-27T18:42:52.767000", "content": "Actually the tendency from 0-12 are not considered except temperature. I would not consider the state variables from the pressure levels 0-12 except temperature. I'll try this in the next days
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2828054": "As shown in [this notebook](https://www.kaggle.com/code/docxian/leap-climsim-visual-eda#Targets-and-Features), there are features with extremely different distributions when comparing the train and test sets. For example, features state_q0001_[0-12] are completely different, and the same happens for features state_q0002_[0-29] and state_q0003_[0-15]. Additionally, some pbuf_ozone features have pretty different distributions.\n\nThis is an example taken from the notebook, for features of the state_q0001 family:\n\n\n\nHow are you managing these features in your models? Are you excluding them completely? Are the records with the missing values hidden in the original low-res dataset? Are these features scaled improperly in the test set, and is this why they are so different?", "2837074": "How to solve this problem?", "2839880": "Actually the tendency from 0-12 are not considered except temperature. I would not consider the state variables from the pressure levels 0-12 except temperature. I'll try this in the next days " } }