Has anyone successfully used the data from merchants.csv and noticed significant difference in performance with and without that data ? I personally found very little improvement in performance after using merchants.csv. This data can offer a significant number of newly engineered categorical features, can anyone list down some of the important ones ?
\n\nAlso, some of the merchant_ids appear multiple times in the data, with different combinations of group_ids and other categorical variables , what would be the best way to handle such repetitions in order to properly append them to train and test data ?
", "messages": [ { "id": "463040", "postDate": "01/29/2019 09:48:25", "content": "Has anyone successfully used the data from merchants.csv and noticed significant difference in performance with and without that data ? I personally found very little improvement in performance after using merchants.csv. This data can offer a significant number of newly engineered categorical features, can anyone list down some of the important ones ?
\n\nAlso, some of the merchant_ids appear multiple times in the data, with different combinations of group_ids and other categorical variables , what would be the best way to handle such repetitions in order to properly append them to train and test data ?
", "rawMarkdown": "Has anyone successfully used the data from merchants.csv and noticed significant difference in performance with and without that data ? I personally found very little improvement in performance after using merchants.csv. This data can offer a significant number of newly engineered categorical features, can anyone list down some of the important ones ? \n\nAlso, some of the merchant_ids appear multiple times in the data, with different combinations of group_ids and other categorical variables , what would be the best way to handle such repetitions in order to properly append them to train and test data ?", "votes": null }, { "id": "464054", "postDate": "01/31/2019 05:59:09", "content": "Hi Rajaswa. Yes i tried using those features as well. It requires a lot of data cleaning. i think around 41 ids have multiple rows. You can use your best judgement to collate them! Try i just tried mean of numerical_1,2 .. no improvement
", "rawMarkdown": "Hi Rajaswa. Yes i tried using those features as well. It requires a lot of data cleaning. i think around 41 ids have multiple rows. You can use your best judgement to collate them! Try i just tried mean of numerical_1,2 .. no improvement", "votes": null }, { "id": "464857", "postDate": "02/01/2019 16:48:23", "content": "For the duplicated ids, I just removed one of them and left the other. I then added it to the transaction data frame by merging on merchant id. Then, I aggregated it like the other features of transaction data frame. Since then, I have removed all of the features and it made by model more accurate and less complex. There may be some important features that you can extract, but the cost is greater than the benefit.
", "rawMarkdown": "For the duplicated ids, I just removed one of them and left the other. I then added it to the transaction data frame by merging on merchant id. Then, I aggregated it like the other features of transaction data frame. Since then, I have removed all of the features and it made by model more accurate and less complex. There may be some important features that you can extract, but the cost is greater than the benefit.", "votes": null }, { "id": "465079", "postDate": "02/02/2019 09:34:04", "content": "very little help, no good way to use.
", "rawMarkdown": "very little help, no good way to use.", "votes": null }, { "id": "465100", "postDate": "02/02/2019 10:15:25", "content": "A quick question, does min-max scaling/standardizing features help ?
", "rawMarkdown": "A quick question, does min-max scaling/standardizing features help ?", "votes": null }, { "id": "465193", "postDate": "02/02/2019 15:29:56", "content": "i just using sum /mean/nunique. i can't add more aggregate features, they are easy to be noise to my model
", "rawMarkdown": "i just using sum /mean/nunique. i can't add more aggregate features, they are easy to be noise to my model", "votes": null }, { "id": "478818", "postDate": "02/26/2019 17:09:18", "content": "https://www.kaggle.com/econdata/elo-17
", "rawMarkdown": "https://www.kaggle.com/econdata/elo-17", "votes": null } ], "comments": [ { "id": 464054, "author_name": "prashanththangavel", "author_url": "", "post_date": "01/31/2019 05:59:09", "content": "Hi Rajaswa. Yes i tried using those features as well. It requires a lot of data cleaning. i think around 41 ids have multiple rows. You can use your best judgement to collate them! Try i just tried mean of numerical_1,2 .. no improvement
", "votes": null, "replies": [] }, { "id": 464857, "author_name": "petersk20", "author_url": "", "post_date": "02/01/2019 16:48:23", "content": "For the duplicated ids, I just removed one of them and left the other. I then added it to the transaction data frame by merging on merchant id. Then, I aggregated it like the other features of transaction data frame. Since then, I have removed all of the features and it made by model more accurate and less complex. There may be some important features that you can extract, but the cost is greater than the benefit.
", "votes": null, "replies": [] }, { "id": 465079, "author_name": "codlife", "author_url": "", "post_date": "02/02/2019 09:34:04", "content": "very little help, no good way to use.
", "votes": null, "replies": [ { "id": 465100, "author_name": "prashanththangavel", "author_url": "", "post_date": "02/02/2019 10:15:25", "content": "A quick question, does min-max scaling/standardizing features help ?
", "votes": null, "replies": [] }, { "id": 465193, "author_name": "codlife", "author_url": "", "post_date": "02/02/2019 15:29:56", "content": "i just using sum /mean/nunique. i can't add more aggregate features, they are easy to be noise to my model
", "votes": null, "replies": [] } ] }, { "id": 478818, "author_name": "econdata", "author_url": "", "post_date": "02/26/2019 17:09:18", "content": "", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "463040": "Has anyone successfully used the data from merchants.csv and noticed significant difference in performance with and without that data ? I personally found very little improvement in performance after using merchants.csv. This data can offer a significant number of newly engineered categorical features, can anyone list down some of the important ones ? \n\nAlso, some of the merchant_ids appear multiple times in the data, with different combinations of group_ids and other categorical variables , what would be the best way to handle such repetitions in order to properly append them to train and test data ?", "464054": "Hi Rajaswa. Yes i tried using those features as well. It requires a lot of data cleaning. i think around 41 ids have multiple rows. You can use your best judgement to collate them! Try i just tried mean of numerical_1,2 .. no improvement", "464857": "For the duplicated ids, I just removed one of them and left the other. I then added it to the transaction data frame by merging on merchant id. Then, I aggregated it like the other features of transaction data frame. Since then, I have removed all of the features and it made by model more accurate and less complex. There may be some important features that you can extract, but the cost is greater than the benefit.", "465079": "very little help, no good way to use.", "465100": "A quick question, does min-max scaling/standardizing features help ?", "465193": "i just using sum /mean/nunique. i can't add more aggregate features, they are easy to be noise to my model", "478818": "https://www.kaggle.com/econdata/elo-17" }, "source": "meta" }