How are y'all creating aggregate features from the merchants file? From my understanding train.csv contains a card_id which is the join key for the historical_transactions.csv file which also contains a card_id column. This is easy enough to aggregate, just historical_transactions.groupby([\"card_id\"])[\"purchase_amount\"].agg(np.mean) or whatever. However, the historical_transactions.csv file also contains a join key for the merchants.csv file with the merchant_id column in both files, which means it's not as easy to create aggregate features from the merchants.csv file. So my question is how are y'all doing a 3 table join (train <-> historical_transactions <-> merchants)? I think my primary issue is that I'm not familiar enough with pandas yet to know how to do this so any help would be greatly appreciated! Thanks!
How are y'all creating aggregate features from the merchants file? From my understanding train.csv contains a card_id which is the join key for the historical_transactions.csv file which also contains a card_id column. This is easy enough to aggregate, just historical_transactions.groupby([\"card_id\"])[\"purchase_amount\"].agg(np.mean) or whatever. However, the historical_transactions.csv file also contains a join key for the merchants.csv file with the merchant_id column in both files, which means it's not as easy to create aggregate features from the merchants.csv file. So my question is how are y'all doing a 3 table join (train <-> historical_transactions <-> merchants)? I think my primary issue is that I'm not familiar enough with pandas yet to know how to do this so any help would be greatly appreciated! Thanks!
Hello, I believe you can join historical transactions with merchants , using card_id, do your feature engineering at card_id level. Once done , you can join with train with card_id. Train csv contains one row per card_id, so that should not pose any problems. I did that way. Let me know if this works.
", "rawMarkdown": "Hello, I believe you can join historical transactions with merchants , using card_id, do your feature engineering at card_id level. Once done , you can join with train with card_id. Train csv contains one row per card_id, so that should not pose any problems. I did that way. Let me know if this works.", "votes": null }, { "id": "446287", "postDate": "12/27/2018 20:05:28", "content": "That did the trick. Thanks!
", "rawMarkdown": "That did the trick. Thanks!", "votes": null } ], "comments": [ { "id": 446176, "author_name": "anirbank", "author_url": "", "post_date": "12/27/2018 16:44:11", "content": "Hello, I believe you can join historical transactions with merchants , using card_id, do your feature engineering at card_id level. Once done , you can join with train with card_id. Train csv contains one row per card_id, so that should not pose any problems. I did that way. Let me know if this works.
", "votes": null, "replies": [ { "id": 446287, "author_name": "timothycwillard", "author_url": "", "post_date": "12/27/2018 20:05:28", "content": "That did the trick. Thanks!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "445845": "How are y'all creating aggregate features from the merchants file? From my understanding *train.csv* contains a `card_id` which is the join key for the *historical\\_transactions.csv* file which also contains a `card_id` column. This is easy enough to aggregate, just `historical_transactions.groupby([\"card_id\"])[\"purchase_amount\"].agg(np.mean)` or whatever. However, the *historical\\_transactions.csv* file also contains a join key for the *merchants.csv* file with the `merchant_id` column in both files, which means it's not as easy to create aggregate features from the *merchants.csv* file. So my question is how are y'all doing a 3 table join (*train* <-> *historical\\_transactions* <-> *merchants*)? I think my primary issue is that I'm not familiar enough with pandas yet to know how to do this so any help would be greatly appreciated! Thanks!", "446176": "Hello, I believe you can join historical transactions with merchants , using card_id, do your feature engineering at card_id level. Once done , you can join with train with card_id. Train csv contains one row per card_id, so that should not pose any problems. I did that way. Let me know if this works.", "446287": "That did the trick. Thanks!" }, "source": "meta" }