![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2099265%2Fd7c302717ca90f526fd3017b4161bf0a%2FScreenshot%202024-02-10%20at%204.43.10%20AM.png?generation=1707569005681360&alt=media\")
For this competition we are provided with feature_definitions.csv file, that contains features and their descriptions. While doing some feature engineering, I found it useful to add two more columns to that table:
dtype, type to which a feature should be resolved; tables, a list of tables where a particular feature could be encountered.
dtype is pretty straightforward for P, A, M and D features, however, for T and L it should be deduced from the actual data. In addition, there is a feature called score_940 that has a description, but couldn't be found in the training dataset.
Here is a bar chart that summarizes dtypes analysis, the \"Missing\" dtype is the score_940 feature.
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2099265%2Ff57f8e53dd8685e4e5afc6971a0b4873%2FScreenshot%202024-02-10%20at%205.17.50%20AM.png?generation=1707571091623992&alt=media\")
I've uploaded the enhanced feature definitions as a Parquet file to kaggle datasets, feel free to use it. Hope you will find it helpful.
", "messages": [ { "id": 2645755, "postDate": "2024-02-10T12:51:09.393Z", "content": "For this competition we are provided with feature_definitions.csv file, that contains features and their descriptions. While doing some feature engineering, I found it useful to add two more columns to that table:
dtype, type to which a feature should be resolved; tables, a list of tables where a particular feature could be encountered.
dtype is pretty straightforward for P, A, M and D features, however, for T and L it should be deduced from the actual data. In addition, there is a feature called score_940 that has a description, but couldn't be found in the training dataset.
Here is a bar chart that summarizes dtypes analysis, the \"Missing\" dtype is the score_940 feature.
I've uploaded the enhanced feature definitions as a Parquet file to kaggle datasets, feel free to use it. Hope you will find it helpful.
", "rawMarkdown": "For this competition we are provided with `feature_definitions.csv` file, that contains features and their descriptions. While doing some feature engineering, I found it useful to add two more columns to that table:\n- `dtype`, type to which a feature should be resolved; \n- `tables`, a list of tables where a particular feature could be encountered.\n\n\n\n`dtype` is pretty straightforward for `P`, `A`, `M` and `D` features, however, for `T` and `L` it should be deduced from the actual data. In addition, there is a feature called `score_940` that has a description, but couldn't be found in the training dataset. \n\nHere is a bar chart that summarizes dtypes analysis, the \"Missing\" dtype is the `score_940` feature. \n\n\n\nI've uploaded the enhanced feature definitions as a [Parquet file](https://www.kaggle.com/datasets/kononenko/home-credit-enhanced-feature-definitions) to kaggle datasets, feel free to use it. Hope you will find it helpful. ", "votes": 42 }, { "id": 2690529, "postDate": "2024-03-10T16:12:26.157Z", "content": "Thanks, very helpful! Based on this answer https://money.stackexchange.com/a/142747, my guess regarding score_940 is that it is most likely a binary indicator showing if a person's credit rating is at least 940, so it makes sense it's not available.
Interesting, I guess you are right. The host actually said they decided not to include it at some point, so this feature is an artifact.
", "rawMarkdown": "Interesting, I guess you are right. The host actually [said](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/478491#2662440) they decided not to include it at some point, so this feature is an artifact." } ] }, { "id": 2725987, "postDate": "2024-03-31T21:36:53.073Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2648315, "postDate": "2024-02-12T08:02:04.617Z", "content": "Thank you very much it is helpfull
", "rawMarkdown": "Thank you very much it is helpfull", "votes": 2 } ], "comments": [ { "id": 2690529, "author_name": "Antonio Šajatović", "author_url": "", "post_date": "2024-03-10T16:12:26.157000", "content": "Thanks, very helpful! Based on this answer https://money.stackexchange.com/a/142747, my guess regarding score_940 is that it is most likely a binary indicator showing if a person's credit rating is at least 940, so it makes sense it's not available.
Interesting, I guess you are right. The host actually said they decided not to include it at some point, so this feature is an artifact.
", "votes": 0, "replies": [] } ] }, { "id": 2725987, "author_name": "", "author_url": "", "post_date": "2024-03-31T21:36:53.073000", "content": "", "votes": 0, "replies": [] }, { "id": 2648315, "author_name": "LucaMTB", "author_url": "", "post_date": "2024-02-12T08:02:04.617000", "content": "Thank you very much it is helpfull
", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "2645755": "For this competition we are provided with `feature_definitions.csv` file, that contains features and their descriptions. While doing some feature engineering, I found it useful to add two more columns to that table:\n- `dtype`, type to which a feature should be resolved; \n- `tables`, a list of tables where a particular feature could be encountered.\n\n\n\n`dtype` is pretty straightforward for `P`, `A`, `M` and `D` features, however, for `T` and `L` it should be deduced from the actual data. In addition, there is a feature called `score_940` that has a description, but couldn't be found in the training dataset. \n\nHere is a bar chart that summarizes dtypes analysis, the \"Missing\" dtype is the `score_940` feature. \n\n\n\nI've uploaded the enhanced feature definitions as a [Parquet file](https://www.kaggle.com/datasets/kononenko/home-credit-enhanced-feature-definitions) to kaggle datasets, feel free to use it. Hope you will find it helpful. ", "2690529": "Thanks, very helpful! Based on this answer https://money.stackexchange.com/a/142747, my guess regarding `score_940` is that it is most likely a binary indicator showing if a person's credit rating is at least 940, so it makes sense it's not available.", "2725987": "", "2648315": "Thank you very much it is helpfull" } }