Since some of the categorical variables have such high carnality, I would like to try target encoding them. But, I would like to generate them out of sample like discussed here https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260. Does anybody know of a package in R that does the target encoding out of sample? I tried making my own function, but did not have much success.
\n\nThanks in advance!
", "messages": [ { "id": "398859", "postDate": "10/04/2018 17:36:25", "content": "Since some of the categorical variables have such high carnality, I would like to try target encoding them. But, I would like to generate them out of sample like discussed here https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260. Does anybody know of a package in R that does the target encoding out of sample? I tried making my own function, but did not have much success.
\n\nThanks in advance!
", "rawMarkdown": "Since some of the categorical variables have such high carnality, I would like to try target encoding them. But, I would like to generate them out of sample like discussed here [https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260][1]. Does anybody know of a package in R that does the target encoding out of sample? I tried making my own function, but did not have much success.\n\nThanks in advance!\n\n\n [1]: https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260", "votes": null }, { "id": "398980", "postDate": "10/04/2018 22:40:01", "content": "I don't know if there is a package to do this in R. However, you can use reticulate and import some python function to apply target encoding .
", "rawMarkdown": "I don't know if there is a package to do this in R. However, you can use reticulate and import some python function to apply target encoding .", "votes": null }, { "id": "399006", "postDate": "10/05/2018 01:33:55", "content": "data.table solution:
\n\ntmp <- train[,mean(totals.transactionRevenue), by=CATEGORICAL_VAR]\ntrain <- merge(train, tmp, by=CATEGORICAL_VAR, all.x=TRUE)\ntest <- merge(test, tmp, by=CATEGORICAL_VAR, all.x=TRUE)\nThanks for the advice!
", "rawMarkdown": "Thanks for the advice!", "votes": null }, { "id": "399575", "postDate": "10/06/2018 05:55:05", "content": "Thank you! I'd like to implement the target encoding out of sample. Like a K-Fold sort of thing. I'll see if I can take this code and modify it.
", "rawMarkdown": "Thank you! I'd like to implement the target encoding out of sample. Like a K-Fold sort of thing. I'll see if I can take this code and modify it.", "votes": null }, { "id": "399688", "postDate": "10/06/2018 13:35:22", "content": "Maybe this can be helpful: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-munging/target-encoding.html
", "rawMarkdown": "Maybe this can be helpful: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-munging/target-encoding.html", "votes": null }, { "id": "399719", "postDate": "10/06/2018 15:17:51", "content": "There was a public kernel by @raddar with target encoding with KFold regularisation:
\n\nhttps://www.kaggle.com/raddar/raddar-extratrees
\n\nThis was one of top results in the Mersedes Benz comp. I actually do not know R, but I was following the duscussion, which led to this kernel
", "rawMarkdown": "There was a public kernel by @raddar with target encoding **with KFold regularisation**:\n\nhttps://www.kaggle.com/raddar/raddar-extratrees\n\n This was one of top results in the Mersedes Benz comp. I actually do not know R, but I was following the duscussion, which led to this kernel", "votes": null } ], "comments": [ { "id": 398980, "author_name": "paulofelipe", "author_url": "", "post_date": "10/04/2018 22:40:01", "content": "I don't know if there is a package to do this in R. However, you can use reticulate and import some python function to apply target encoding .
", "votes": null, "replies": [ { "id": 399574, "author_name": "obrienmitch94", "author_url": "", "post_date": "10/06/2018 05:53:37", "content": "Thanks for the advice!
", "votes": null, "replies": [] } ] }, { "id": 399006, "author_name": "returnofsputnik", "author_url": "", "post_date": "10/05/2018 01:33:55", "content": "data.table solution:
\n\ntmp <- train[,mean(totals.transactionRevenue), by=CATEGORICAL_VAR]\ntrain <- merge(train, tmp, by=CATEGORICAL_VAR, all.x=TRUE)\ntest <- merge(test, tmp, by=CATEGORICAL_VAR, all.x=TRUE)\nThank you! I'd like to implement the target encoding out of sample. Like a K-Fold sort of thing. I'll see if I can take this code and modify it.
", "votes": null, "replies": [] } ] }, { "id": 399688, "author_name": "paulofelipe", "author_url": "", "post_date": "10/06/2018 13:35:22", "content": "Maybe this can be helpful: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-munging/target-encoding.html
", "votes": null, "replies": [] }, { "id": 399719, "author_name": "mlisovyi", "author_url": "", "post_date": "10/06/2018 15:17:51", "content": "There was a public kernel by @raddar with target encoding with KFold regularisation:
\n\nhttps://www.kaggle.com/raddar/raddar-extratrees
\n\nThis was one of top results in the Mersedes Benz comp. I actually do not know R, but I was following the duscussion, which led to this kernel
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "398859": "Since some of the categorical variables have such high carnality, I would like to try target encoding them. But, I would like to generate them out of sample like discussed here [https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260][1]. Does anybody know of a package in R that does the target encoding out of sample? I tried making my own function, but did not have much success.\n\nThanks in advance!\n\n\n [1]: https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/66260", "398980": "I don't know if there is a package to do this in R. However, you can use reticulate and import some python function to apply target encoding .", "399006": "data.table solution:\n\n tmp <- train[,mean(totals.transactionRevenue), by=CATEGORICAL_VAR]\n train <- merge(train, tmp, by=CATEGORICAL_VAR, all.x=TRUE)\n test <- merge(test, tmp, by=CATEGORICAL_VAR, all.x=TRUE)", "399574": "Thanks for the advice!", "399575": "Thank you! I'd like to implement the target encoding out of sample. Like a K-Fold sort of thing. I'll see if I can take this code and modify it.", "399688": "Maybe this can be helpful: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-munging/target-encoding.html", "399719": "There was a public kernel by @raddar with target encoding **with KFold regularisation**:\n\nhttps://www.kaggle.com/raddar/raddar-extratrees\n\n This was one of top results in the Mersedes Benz comp. I actually do not know R, but I was following the duscussion, which led to this kernel" }, "source": "meta" }