{ "id": 491974, "title": "how to add categarical values when we have duplicate case id's (ie: applprev data set of credit previous data)", "url": "/competitions/home-credit-credit-risk-model-stability/discussion/491974", "author_name": "", "post_date": "2024-04-08T05:07:15.063791200Z", "votes": null, "comment_count": 2, "views": 0, "content": "

applprev data set of credit previous data has multiple case id's of previous credit tansational data which are categarical like 1. cacccardblochreas_147M (Card blocking reason) , 2. conts_type_509L (Person contact type in previous application), 3.credacc_cards_status_52L(Card status of the previous credit account).

\n

Data understanding : https://www.kaggle.com/code/kunduruanil/credit-risk-data-understanding

\n

Feature Extration : https://www.kaggle.com/code/kunduruanil/create-training-dataset

", "messages": [ { "id": "2741039", "postDate": "04/08/2024 05:07:15", "content": "

applprev data set of credit previous data has multiple case id's of previous credit tansational data which are categarical like 1. cacccardblochreas_147M (Card blocking reason) , 2. conts_type_509L (Person contact type in previous application), 3.credacc_cards_status_52L(Card status of the previous credit account).

\n

Data understanding : https://www.kaggle.com/code/kunduruanil/credit-risk-data-understanding

\n

Feature Extration : https://www.kaggle.com/code/kunduruanil/create-training-dataset

", "rawMarkdown": "applprev data set of credit previous data has multiple case id's of previous credit tansational data which are categarical like 1. cacccardblochreas_147M (Card blocking reason) , 2. conts_type_509L (Person contact type in previous application), 3.credacc_cards_status_52L(Card status of the previous credit account).\n\nData understanding : [https://www.kaggle.com/code/kunduruanil/credit-risk-data-understanding](url)\n\nFeature Extration : [https://www.kaggle.com/code/kunduruanil/create-training-dataset](url)", "votes": null }, { "id": "2741999", "postDate": "04/08/2024 17:31:16", "content": "

general procedure is to get a feature file from applprev data such that every row if one case_id than to use join to add these features to the base file.
\nnow to get the feature file we need to aggregate using groupby over case_id. in there u can use various conditions like take the top most value or return True if certain value is present in the list, or count the occurrence of a specific value in the list, or u can use some other logic/condition of ur choice to aggregate the list to one value.
\nU can perform this multiple times to get multiple features from a single feature
\nmore u extract more information u get but also more noise. so u need get balance between which feature u are extracting and how many.

", "rawMarkdown": "general procedure is to get a feature file from applprev data such that every row if one case_id than to use join to add these features to the base file.\nnow to get the feature file we need to aggregate using groupby over case_id. in there u can use various conditions like take the top most value or return True if certain value is present in the list, or count the occurrence of a specific value in the list, or u can use some other logic/condition of ur choice to aggregate the list to one value.\nU can perform this multiple times to get multiple features from a single feature \nmore u extract more information u get but also more noise. so u need get balance between which feature u are extracting and how many.", "votes": null }, { "id": "2742190", "postDate": "04/08/2024 19:32:36", "content": "

You can calculate mode, count number of unique values, or perform one-hot encoding and then sum up values.

", "rawMarkdown": "You can calculate mode, count number of unique values, or perform one-hot encoding and then sum up values.", "votes": null } ], "comments": [ { "id": 2741999, "author_name": "shreyas9181", "author_url": "", "post_date": "04/08/2024 17:31:16", "content": "

general procedure is to get a feature file from applprev data such that every row if one case_id than to use join to add these features to the base file.
\nnow to get the feature file we need to aggregate using groupby over case_id. in there u can use various conditions like take the top most value or return True if certain value is present in the list, or count the occurrence of a specific value in the list, or u can use some other logic/condition of ur choice to aggregate the list to one value.
\nU can perform this multiple times to get multiple features from a single feature
\nmore u extract more information u get but also more noise. so u need get balance between which feature u are extracting and how many.

", "votes": null, "replies": [] }, { "id": 2742190, "author_name": "eivolkova", "author_url": "", "post_date": "04/08/2024 19:32:36", "content": "

You can calculate mode, count number of unique values, or perform one-hot encoding and then sum up values.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2741039": "applprev data set of credit previous data has multiple case id's of previous credit tansational data which are categarical like 1. cacccardblochreas_147M (Card blocking reason) , 2. conts_type_509L (Person contact type in previous application), 3.credacc_cards_status_52L(Card status of the previous credit account).\n\nData understanding : [https://www.kaggle.com/code/kunduruanil/credit-risk-data-understanding](url)\n\nFeature Extration : [https://www.kaggle.com/code/kunduruanil/create-training-dataset](url)", "2741999": "general procedure is to get a feature file from applprev data such that every row if one case_id than to use join to add these features to the base file.\nnow to get the feature file we need to aggregate using groupby over case_id. in there u can use various conditions like take the top most value or return True if certain value is present in the list, or count the occurrence of a specific value in the list, or u can use some other logic/condition of ur choice to aggregate the list to one value.\nU can perform this multiple times to get multiple features from a single feature \nmore u extract more information u get but also more noise. so u need get balance between which feature u are extracting and how many.", "2742190": "You can calculate mode, count number of unique values, or perform one-hot encoding and then sum up values." }, "source": "meta" }